Financial resource recommendation method and device for small and medium-sized enterprises, and storage medium

Abstract

The application discloses a small and medium-sized enterprise-oriented financial resource recommendation method and device and a storage medium, comprising: distributing enterprise characteristic data in parallel to multiple heterogeneous special agents in a multi-agent cooperation architecture for dynamic cooperation processing to obtain a confidence score, a dynamic risk assessment report and a financial health score; adopting the confidence score to deploy model parameters of a confidence weighted fusion model to adjust enterprise dimension contribution weights of the dynamic risk assessment report and the financial report; generating an enterprise comprehensive portrait through the adjustment result; and vectorizing the enterprise comprehensive portrait to obtain an enterprise characteristic vector and mapping the enterprise characteristic vector to a high-dimensional vector space of a financial product knowledge graph for multiple rounds of progressive matching to generate a personalized financial product recommendation scheme. The application constructs a comprehensive enterprise portrait through multi-agent cooperation, realizes accurate product matching by combining a knowledge graph to provide product recommendation, effectively improves the accuracy and personalization of financing recommendation, and significantly improves the financing success rate of small and medium-sized enterprises.

Description

Technical Field

[0001] This application relates to the field of financial data processing, and in particular to a method, device and storage medium for recommending financial resources to small and medium-sized enterprises. Background Technology

[0002] Currently, financial resource recommendation services for SMEs face numerous technological bottlenecks, hindering the accuracy and efficiency of financing matching. The fundamental technical challenge lies in the failure to construct an organic system capable of deeply integrating multi-source information, conducting intelligent collaborative analysis, and executing dynamic optimization matching, rather than simply displaying information or filtering by rules.

[0003] Specifically, enterprise data comes from diverse sources and has a heterogeneous structure, encompassing multimodal information such as financial statements, operational documents, and industry data. Traditional methods lack effective technical means to uniformly clean, standardize, and semantically integrate this data, resulting in single-dimensional enterprise feature vectors with limited representational capabilities, failing to comprehensively depict the true state of enterprises. At the data analysis architecture level, financial, risk, and demand analysis modules often operate independently, lacking cross-module cross-validation and collaborative reasoning mechanisms. In the core matching stage, only hard screening of product access conditions can be achieved, failing to quantitatively calculate and comprehensively weigh multiple dimensions such as risk tolerance, financing costs, term structure, and flexibility of terms. Its matching logic lacks an intelligent decision-making process capable of semantic similarity calculation and dynamic optimization in a unified vector space, resulting in crude and impractical recommendations that cannot provide enterprises with accurate financial resource recommendation services.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device and storage medium for recommending financial resources for small and medium-sized enterprises (SMEs), aiming to solve the technical problems of information asymmetry and low accuracy of matching financial products in existing SME financing.

[0006] To achieve the above objectives, this application proposes a method for recommending financial resources to small and medium-sized enterprises (SMEs), the method comprising:

[0007] Obtain standardized enterprise characteristic data;

[0008] The enterprise feature data is distributed in parallel to multiple heterogeneous dedicated agents in the multi-agent collaborative architecture for dynamic collaborative processing. Based on the processing results, a confidence score, a dynamic risk assessment report, and a financial health score are obtained.

[0009] The model parameters of the confidence-weighted fusion model are deployed using the confidence score, and the enterprise dimension contribution weights of the dynamic risk assessment report and the financial health score are adjusted based on the confidence-weighted fusion model.

[0010] A comprehensive enterprise profile is generated based on the adjustment results of the enterprise dimension contribution weights.

[0011] The comprehensive profile of the enterprise is vectorized to obtain the enterprise feature vector, and the enterprise feature vector is mapped to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching.

[0012] A personalized financial product recommendation scheme is generated based on the matching results of the multi-round progressive matching.

[0013] Output the personalized financial product recommendation scheme.

[0014] In one embodiment, the heterogeneous dedicated intelligent agents include a basic information intelligent agent, a financial analysis intelligent agent, an operational status intelligent agent, a risk assessment intelligent agent, and a demand analysis intelligent agent. The step of distributing the enterprise characteristic data in parallel to multiple heterogeneous dedicated intelligent agents in a multi-agent collaborative architecture for dynamic collaborative processing includes:

[0015] The hierarchical industry classification model and industry knowledge graph pre-set by the basic information intelligent agent are used to perform semantic matching on the enterprise feature data, generate enterprise sub-industry tags, and share the enterprise sub-industry tags with the financial analysis intelligent agent and the business status intelligent agent in real time.

[0016] The financial analysis agent and the operational status agent simultaneously receive the enterprise's subdivided industry tags, process the enterprise's subdivided industry tags to obtain a financial health score and a life cycle stage identifier, and share the financial health score and operational stage identifier with the demand analysis agent and the risk assessment agent.

[0017] The risk assessment agent selects the corresponding risk assessment model based on the received life cycle stage identifier, scans the financial indicators processed by the financial analysis agent through the risk assessment model, and generates a dynamic risk assessment report from the scanning results.

[0018] The demand analysis agent comprehensively processes the enterprise's subdivided industry tags, life cycle stage identifiers, and financial health scores, and predicts the scale and preferences of the enterprise's potential financing needs through multiple regression analysis.

[0019] In one embodiment, the steps of vectorizing the comprehensive enterprise profile to obtain enterprise feature vectors, and mapping the enterprise feature vectors to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching include:

[0020] The corresponding matching dimension parameters are set for each matching stage of the multi-round progressive matching by a pre-configured progressive matching strategy;

[0021] Based on the matching dimension parameters, a multi-round progressive matching is performed in the high-dimensional vector space.

[0022] In one embodiment, the step of performing multi-round progressive matching based on the matching dimension parameter in the high-dimensional vector space includes:

[0023] The first round of matching is performed based on hard constraints, and Boolean filtering is conducted through the set basic admission conditions. A preliminary set of products is generated based on the filtering results.

[0024] The second round of matching is performed based on the quantitative scoring dimension. The weighted cosine similarity between the enterprise feature vector and the feature vector of each financial product in the initial screening product set on the core dimension is calculated. The target product sequence is generated based on the weighted cosine similarity.

[0025] The final matching round is performed based on the strategy optimization dimension. Real-time supply data from the financial market and policy guidance factors are introduced as dynamic adjustment weights to comprehensively reorder the target product series. The final matching result is obtained by calculating the comprehensive reordering result through a comprehensive risk-return model.

[0026] In one embodiment, the step of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching includes:

[0027] Key decision data is extracted during the multi-round progressive matching process, and the key decision data is mapped and filled with a preset natural language template.

[0028] Based on the enterprise size and industry attributes in the comprehensive enterprise profile, the terminology granularity and level of detail of the natural language template are adaptively adjusted to generate explanatory text for the personalized financial product recommendation scheme.

[0029] In one embodiment, the step of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching includes:

[0030] The enterprise feature vector and the product feature vector in the matching results are compared dimensionally, and the difference between the enterprise feature vector and the product feature vector is determined based on the comparison results.

[0031] Based on the aforementioned gap, the strategy rule base is invoked to generate financial optimization suggestions;

[0032] Strategic recommendations for the personalized financial product recommendation scheme are generated by using real-time financial market data and the aforementioned financial optimization suggestions.

[0033] In one embodiment, before the step of vectorizing the comprehensive enterprise profile to obtain an enterprise feature vector and mapping the enterprise feature vector to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching, the method further includes:

[0034] Information on multiple financial products is obtained through a pre-defined API interface, and the information on multiple financial products is parsed and standardized to generate financial product feature information.

[0035] Based on a predefined ontology model, the feature information of the financial products is mapped to nodes and attributes in a knowledge graph to form the financial product knowledge graph.

[0036] In one embodiment, the step of mapping the financial product feature information to nodes and attributes in a knowledge graph based on a predefined ontology model to form the financial product knowledge graph includes:

[0037] Calculate the multidimensional feature vectors between each product node in the knowledge graph;

[0038] Based on the dimensions of the multidimensional feature vector, corresponding product association relationships are created, and the product association network of the knowledge graph is constructed through the product association relationships.

[0039] Furthermore, to achieve the above objectives, this application also proposes a financial resource recommendation device for small and medium-sized enterprises (SMEs), the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the financial resource recommendation method for SMEs as described above.

[0040] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the financial resource recommendation method for small and medium-sized enterprises as described above.

[0041] One or more technical solutions proposed in this application have at least the following technical effects:

[0042] The technical solution of this application involves: acquiring standardized enterprise characteristic data; distributing the enterprise characteristic data in parallel to multiple heterogeneous dedicated intelligent agents in a multi-agent collaborative architecture for dynamic collaborative processing; obtaining confidence scores, dynamic risk assessment reports, and financial health scores based on the processing results; deploying model parameters of a confidence-weighted fusion model using the confidence scores; adjusting the enterprise dimension contribution weights of the dynamic risk assessment report and the financial health score based on the confidence-weighted fusion model; generating a comprehensive enterprise profile based on the adjusted enterprise dimension contribution weights; vectorizing the comprehensive enterprise profile to obtain enterprise feature vectors; mapping the enterprise feature vectors to a high-dimensional vector space of a financial product knowledge graph for multi-round progressive matching; generating personalized financial product recommendation schemes based on the matching results of the multi-round progressive matching; and outputting the personalized financial product recommendation schemes. This application constructs a comprehensive enterprise profile through multi-agent collaboration and combines it with a knowledge graph to achieve accurate product matching and provide product recommendations, effectively improving the accuracy and personalization of financing recommendations and significantly increasing the success rate of financing for SMEs. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an embodiment of the financial resource recommendation method for SMEs in this application.

[0046] Figure 2 This is a detailed step diagram of step S20 above;

[0047] Figure 3 This is a detailed step diagram of step S50 above;

[0048] Figure 4 This is a schematic diagram of the hardware operating environment involved in the financial resource recommendation method for SMEs in this application embodiment.

[0049] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0051] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0052] The main solution of this application embodiment is as follows: Obtain standardized enterprise feature data; distribute the enterprise feature data in parallel to multiple heterogeneous dedicated intelligent agents in a multi-agent collaborative architecture for dynamic collaborative processing, and obtain confidence scores, dynamic risk assessment reports, and financial health scores based on the processing results; deploy the model parameters of a confidence-weighted fusion model using the confidence scores, and adjust the enterprise dimension contribution weights of the dynamic risk assessment report and the financial health score based on the confidence-weighted fusion model; generate a comprehensive enterprise profile based on the adjustment results of the enterprise dimension contribution weights; vectorize the comprehensive enterprise profile to obtain enterprise feature vectors, and map the enterprise feature vectors to the high-dimensional vector space of a financial product knowledge graph for multi-round progressive matching; generate a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching; and output the personalized financial product recommendation scheme.

[0053] The current financial resource recommendation service for SMEs faces challenges such as difficulty in integrating multi-source heterogeneous data, insufficient collaboration among analysis modules, and matching mechanisms that rely solely on rule-based screening rather than intelligent optimization. These issues result in limited representation of enterprise characteristics and insufficient comprehensive evaluation, making it difficult to achieve accurate product recommendations.

[0054] This application provides a solution that constructs a comprehensive enterprise profile through multi-agent collaboration and combines it with a knowledge graph to achieve precise product matching and provide product recommendations. This effectively improves the accuracy and personalization of financing recommendations and significantly increases the success rate of financing for SMEs.

[0055] Based on this, the embodiments of this application provide a method for recommending financial resources for small and medium-sized enterprises, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the financial resource recommendation method for SMEs according to this application. In this embodiment, the financial resource recommendation method for SMEs includes steps S10 to S70:

[0056] Step S10: Obtain standardized enterprise characteristic data;

[0057] In this embodiment, standardized enterprise characteristic data is acquired. This acquisition process is achieved by collecting raw enterprise data from multiple sources. Specifically, these multiple sources include official authoritative channels, professional data provided by commercial databases and third-party credit reporting platforms, financial reports and operating data voluntarily submitted by enterprises, shared data from partners, and other publicly available market information. Due to the diverse data sources, the acquired enterprise data exhibits significant differences in format, encoding rules, update frequency, and units of measurement, displaying typical heterogeneous characteristics. Therefore, it must undergo rigorous standardization processing before it can be used for subsequent analysis. Thus, the acquired enterprise data is defined as raw multimodal data.

[0058] The raw multimodal data undergoes standardization processing through data cleaning. This standardization process includes cleaning, standardization, and normalization. The cleaning process identifies and addresses noise, errors, missing values, and inconsistencies in the raw multimodal data, thereby cleaning the data. The specific standardization process includes checking the logical rationality of the data through validity verification, identifying and removing outliers that significantly deviate from the normal range using rule engines or statistical methods, repairing missing fields using interpolation, imputation, or model-based prediction methods, and eliminating duplicate records through entity resolution technology to ensure the uniqueness of enterprise entities.

[0059] After data cleaning, the cleaned raw multimodal data undergoes a standardization process. This is because the raw multimodal data comes from different database systems, file formats, or API interfaces, each with its own structure and semantic definition. Therefore, the standardization process parses and maps the raw multimodal data to a unified mediating model or data schema, thereby standardizing the expression of fields such as date, amount, and percentage. Furthermore, unstructured or semi-structured data is parsed and extracted, transforming it into structured fields.

[0060] Next, the original multimodal data after format conversion underwent index normalization, scaling the values ​​to a uniform scale through mathematical transformations. Normalization ensures that the data indicators of the original multimodal data are all within a comparable numerical range, eliminating the influence of dimensions and laying a reliable data foundation for subsequent multi-dimensional collaborative analysis, agent processing, and precise matching. This standardization process improves data quality and fundamentally solves the problems of "information silos" and low matching efficiency caused by data heterogeneity, ensuring the comparability and computability of enterprise characteristic data.

[0061] Based on the original multimodal data after indicator normalization, extract the data fields with enterprise characteristics as the enterprise characteristic data, or use the original multimodal data after indicator normalization as the enterprise characteristic data.

[0062] Step S20: Distribute the enterprise feature data in parallel to multiple heterogeneous dedicated agents in the multi-agent collaborative architecture for dynamic collaborative processing, and obtain confidence scores, dynamic risk assessment reports and financial health scores based on the processing results.

[0063] The standardized enterprise feature data described in step S10 above is input into a multi-agent collaborative architecture for in-depth analysis and processing. The core of this architecture lies in the process of decoupling and fusing the enterprise feature data through multiple agent modules with different algorithms in a parallel and collaborative manner. The processing flow of this architecture begins with a "parallel distribution" mechanism.

[0064] Specifically, the enterprise standard data is not sent sequentially to each heterogeneous dedicated intelligent agent. Instead, through a preset scheduling mechanism, a complete copy of the enterprise standard data, or a subset of relevant data based on the enterprise standard data obtained through routing, is simultaneously distributed to all relevant heterogeneous dedicated intelligent agents. This parallel distribution operation relies on an underlying distributed message middleware or data bus, ensuring that data arrives at the processing nodes of each heterogeneous dedicated intelligent agent with extremely low latency and near-synchronous delivery. This fundamentally eliminates the data synchronization delay and blocking problems inherent in traditional serial processing, laying a solid foundation for the system's high throughput and real-time response.

[0065] After the enterprise feature data is distributed in parallel, each of the heterogeneous specialized intelligent agents, based on its professional domain knowledge base and built-in algorithm model, performs parallel processing on the received enterprise feature data. Specifically, the basic information analysis intelligent agent focuses on analyzing the enterprise's business registration information, equity structure, intellectual property holdings, etc., and uses natural language processing and classification models to generate accurate enterprise industry classification labels and business scope descriptions; the financial evaluation intelligent agent deeply mines the enterprise's balance sheet, income statement, cash flow statement and other financial records, and calculates a comprehensive financial health score through a series of financial ratio analyses, trend prediction models and health assessment algorithms to quantify the enterprise's solvency, profitability and operational efficiency; the risk prediction intelligent agent integrates the enterprise's internal financial data and external macroeconomic environment information, uses statistical learning and machine learning models to dynamically assess the credit risk, market risk and operational risk that the enterprise may face now and in the future, and generates a structured dynamic risk assessment report.

[0066] It is important to clarify that the dynamic collaborative processing of the enterprise characteristic data by the multiple heterogeneous dedicated intelligent agents essentially involves real-time, bidirectional information exchange and collaborative computation between these agents as needed during the processing. Communication is achieved through predefined real-time data exchange protocols and collaboration rules. Specifically, if the risk prediction agent detects an anomaly in a financial indicator during the assessment process, it can request a more detailed historical calculation process or relevant supporting data for that indicator from the financial assessment agent in real time. Conversely, when calculating the score, the financial assessment agent may also instantly obtain the latest industry classification labels generated by the basic information agent to ensure the accuracy of the scoring model parameters. In essence, after processing the enterprise characteristic data, the multiple heterogeneous dedicated intelligent agents generate a dynamic risk assessment report and a financial health score based on the enterprise.

[0067] Based on the processing of the enterprise feature data by the heterogeneous specialized intelligent agent, the system outputs analysis results and provides a confidence score calculated based on the determinism of its internal model or data quality assessment. This confidence score indicates the reliability of the intelligent agent's current output.

[0068] Step S30: Deploy the model parameters of the confidence-weighted fusion model using the confidence score, and adjust the enterprise dimension contribution weights of the dynamic risk assessment report and the financial health score based on the confidence-weighted fusion model;

[0069] Based on the output results of heterogeneous specialized agents in the multi-agent collaborative architecture, the model parameters are deployed and dynamic weights are adjusted. Specifically, the confidence scores generated by each heterogeneous specialized agent are used to configure the configuration parameters of the confidence-weighted fusion model to achieve an intelligent fusion mechanism. Furthermore, the contribution weights of the dynamic risk assessment report and financial report in constructing a comprehensive enterprise profile are dynamically adjusted based on the confidence-weighted fusion model.

[0070] Specifically, in the process of deploying the model parameters of the confidence-weighted fusion model using confidence scores, the confidence scores output by the heterogeneous dedicated agents are collected, and these confidence scores are initialized as key guiding parameters for the weight learning module. An initial importance bias is pre-assigned to the data dimensions output by the corresponding heterogeneous dedicated agents based on these confidence scores. For example, if the confidence score output by the financial assessment agent is higher than that of other agents, the initial value of the parameter corresponding to the financial health score dimension is increased when initializing the weight learning module.

[0071] Based on the confidence-weighted fusion model with pre-initialized parameters, the contribution weight of the enterprise dimension is dynamically adjusted. Specifically, the confidence-weighted fusion model continuously receives real-time data streams and confidence scores from each agent. In actual operation, through the differentiable learning mechanism within the confidence-weighted fusion model, the contribution weight of the data dimension corresponding to the confidence score is appropriately increased. Furthermore, if the confidence of a certain dimension, such as a sub-indicator in a dynamic risk assessment report, decreases due to data quality fluctuations or increased model uncertainty, the confidence-weighted fusion model smoothly reduces the contribution weight of this dimension parameter in the final decision, and seeks supplementary information based on other high-confidence dimensions for cross-validation and compensation to avoid errors.

[0072] Step S40: Generate a comprehensive enterprise profile based on the adjustment results of the enterprise dimension contribution weights;

[0073] The final weight allocation scheme output by the confidence-weighted fusion model is analyzed, and the multi-dimensional evaluation results are calculated by fusing the weight allocation scheme. Specifically, the original evaluation results produced by each agent are aligned and standardized, and the resulting dynamic weights are used as fusion coefficients to generate a comprehensive enterprise profile. Specifically, if the confidence level of the financial data in the current period is extremely high, the weight is increased, and the financial health score of the comprehensive enterprise profile is strengthened based on the key indicators of the financial health score.

[0074] Deep integration of information structure is performed based on the weighted calculations represented by the weighting allocation scheme. Specifically, high-weighted dimensions serve as the backbone of the comprehensive enterprise profile, while lower-weighted dimensions act as important supplements and verifications. Furthermore, through this deep fusion of dynamic weights, a unified comprehensive enterprise profile is constructed. This comprehensive enterprise profile employs a hybrid representation approach. Specifically, the main body of the comprehensive enterprise profile is a highly structured data framework, listing core conclusions and quantitative scores for dimensions such as enterprise operating status, financial health, and risk level. Simultaneously, processed unstructured features can be embedded into the comprehensive enterprise profile. These feature vectors retain the subtle differences and potential patterns of the original multimodal data, providing an information foundation for subsequent deep matching and analysis.

[0075] Step S50: The comprehensive profile of the enterprise is vectorized to obtain the enterprise feature vector, and the enterprise feature vector is mapped to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching.

[0076] The generated comprehensive enterprise profile is vectorized, transforming it into a high-dimensional feature vector. This high-dimensional feature vector is then matched against a pre-configured financial product knowledge graph. A graph neural network maps the high-dimensional feature vector to a high-dimensional vector space for matching.

[0077] In the matching process of the high-dimensional vector space, the high-dimensional feature vector converted from the comprehensive profile of the enterprise is matched with the financial product knowledge graph in multiple rounds of progressive matching. This progressive matching can be carried out in stages and at multiple levels to obtain the final matching result.

[0078] Specifically, based on the most stable and fundamental attributes in the comprehensive enterprise profile, such as the industry category, years of establishment, registered capital, and required financing amount and term, the products in the financial product knowledge graph are initially screened. This initial screening quickly eliminates obviously inconsistent products, narrows down the candidate product set, and improves the efficiency of subsequent matching.

[0079] The candidate product pool after the initial screening enters the second round of matching, which is a matching stage based on semantic similarity calculation. In this matching stage, the distance or similarity between the vectorized enterprise feature vector and the vector of each candidate financial product in high-dimensional space is calculated in depth, so as to quantify the degree of fit between enterprises and products from a semantic level, and the candidate products are ranked according to the similarity score.

[0080] The third round of matching involves dynamic optimization and re-ranking. In this matching phase, real-time changing external market factors and business rules are introduced as adjustment variables to weight and recalibrate the similarity ranking results generated in the second round. Through a multi-round, progressively refined matching strategy, an optimal list of financial product recommendations can be output that meets the company's hard requirements, possesses strong semantic relevance, and sensitively reflects dynamic market changes, while maintaining efficiency.

[0081] Based on the above matching operations, and built upon a pre-created financial product knowledge graph, prior to the steps of vectorizing the comprehensive enterprise profile to obtain enterprise feature vectors and mapping these enterprise feature vectors to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching, the following steps are also included:

[0082] Information on multiple financial products is obtained through a pre-defined API interface, and the information on multiple financial products is parsed and standardized to generate financial product feature information.

[0083] Based on a predefined ontology model, the feature information of the financial products is mapped to nodes and attributes in a knowledge graph to form the financial product knowledge graph.

[0084] In this embodiment, before vectorizing the comprehensive enterprise profile and mapping it to the financial product knowledge graph for matching, the construction and updating of the financial product knowledge graph are performed in advance. Through a pre-configured set of application programming interfaces (APIs), raw financial product information, including product terms, interest rate descriptions, and applicable conditions, is periodically obtained from multiple data platforms. The obtained raw financial product information is then standardized.

[0085] Specifically, the standardization process includes data parsing, cleaning, and transformation. The original financial product information is parsed to extract core feature fields such as product name, type, interest rate, term, amount, guarantee method, and applicable industry. A rule-based and algorithm-based cleaning module identifies and filters invalid records, redundant information, and outliers in the original financial product information, and fills in or marks missing necessary fields. Then, a format conversion module transforms the diverse and inconsistent original financial product information into a predefined standardized format, for example, converting all interest rate representations into a unified annualized interest rate standard, ensuring the comparability of all product information.

[0086] The standardized financial product feature information is input into a knowledge graph construction engine based on a predefined ontology model. The predefined ontology model predefines the core concepts, attributes, and hierarchical relationships of the financial product domain. Based on this predefined ontology model, the knowledge graph construction engine maps each standardized original financial product information to an entity node in the knowledge graph. The attributes of these entity nodes correspond to the dimensional definitions of the ontology model. Simultaneously, by calculating the semantic relevance of the product features of the original financial product information in a high-dimensional vector space, various types of associations, such as "similar products" and "substitute products," are established between different nodes, thereby forming a semantically rich and structured financial product knowledge graph. Specifically, this process is defined as the steps of mapping the financial product feature information to nodes and attributes in the knowledge graph based on the predefined ontology model to form the financial product knowledge graph, including:

[0087] Calculate the multidimensional feature vectors between each product node in the knowledge graph;

[0088] Based on the dimensions of the multidimensional feature vector, corresponding product association relationships are created, and the product association network of the knowledge graph is constructed through the product association relationships.

[0089] In constructing a financial product knowledge graph, product relationships are established by calculating multi-dimensional feature vectors between product nodes in the knowledge graph, thereby constructing the product relationship network of the financial product knowledge graph. Specifically, based on standardized financial product feature information, a series of core dimensional features are extracted, including interest rate characteristics, risk level, applicable enterprise size, financing term, guarantee method, and repayment flexibility. A specific vector generation algorithm is used to calculate the corresponding multi-dimensional feature vector for each product node. These multi-dimensional feature vectors are used to comprehensively represent the product's distribution position in the multi-dimensional attribute space.

[0090] Based on this, associations are automatically created by calculating the similarity of multi-dimensional feature vectors between product nodes. The association creation process adopts a dynamic threshold control mechanism, which adaptively adjusts the judgment criteria for associations according to different types of association dimensions. When it is determined that the vector similarity between two product nodes in a specific dimension exceeds their corresponding threshold, the system automatically creates an association edge with type and strength attributes in the knowledge graph.

[0091] By systematically establishing all eligible product relationships, a product relationship network is constructed for the financial product knowledge graph.

[0092] Step S60: Generate a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching.

[0093] Based on the ordered candidate product list generated through multiple rounds of progressive matching, the process of generating personalized financial product recommendation solutions is initiated. Specifically, the matching results obtained through quantitative calculations are transformed into personalized recommendation reports that are structurally complete, well-reasoned, easy to apply in business, and compliant with regulatory requirements, thereby integrating decision-making logic with interpretability.

[0094] This generation process comprehensively considers the matching score ranking in the ordered candidate product list, the complementarity between products, the risk diversification principles of financial institutions, and the potential diverse needs of enterprises to select the most representative recommended combination. The recommended combination includes core recommended products and several alternative solutions, which can meet the financing needs of enterprises at different levels or as alternatives.

[0095] By employing natural language generation technology, abstract matching data and logical relationships are transformed into easily understandable written recommendation reasons, enhancing the interpretability and operability of the recommendation scheme. During this generation process, detailed product matching analysis clearly elucidates the specific points of convergence between each recommended product and the characteristics of the enterprise. For example, it explains that "this innovative loan product is recommended primarily based on the enterprise's high-tech industry label, asset-light operating model, and the continuous growth trend of current R&D investment. These characteristics are highly consistent with the product's design goal of supporting technology-based enterprises to expand innovation," providing detailed and specific representations. It also provides specific applicability explanations, explaining from the perspective of actual enterprise application how the product's key terms match the enterprise's current cash flow situation and financing plans, and may offer usage suggestions. Furthermore, the personalized financial product recommendation scheme also includes necessary risk warnings, pointing out the potential restrictions, market risks, and potential considerations related to the enterprise's own financial situation when applying for the product, ensuring the prudence and compliance of the recommendation content.

[0096] Furthermore, this generation process is constantly validated and guided by actual business rules and constraints. Through a built-in series of risk control rules and compliance clauses, it ensures that the generated personalized financial product recommendation solutions not only improve data-level matching accuracy but also possess business feasibility. The final personalized financial product recommendation solution is a comprehensive decision support document that deeply integrates quantitative matching results, a semantic interpretation framework, and real-world business logic, providing financial institution account managers and corporate decision-makers with a final output that combines accuracy, insight, and operability.

[0097] Specifically, this generation process can be defined as the steps of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching, including:

[0098] Key decision data is extracted during the multi-round progressive matching process, and the key decision data is mapped and filled with a preset natural language template.

[0099] Based on the enterprise size and industry attributes in the comprehensive enterprise profile, the terminology granularity and level of detail of the natural language template are adaptively adjusted to generate explanatory text for the personalized financial product recommendation scheme.

[0100] In the process of generating personalized financial product recommendations based on multi-round progressive matching results, corresponding explanatory text is generated simultaneously to enhance the transparency and comprehensibility of the recommendations. During this generation process, key decision data is extracted from the multi-round progressive matching, including the core conditions used in each round of selection, the similarity index derived from quantitative matching, and the market or policy parameters used in the dynamic adjustment phase. This key decision data is then intelligently mapped and filled with pre-set natural language templates.

[0101] The natural language template adopts a modular design with built-in variable slots and conditional judgment logic, allowing it to adaptively select appropriate expression structures based on the data type of the input key decision data. Furthermore, based on the enterprise size and industry attributes included in the comprehensive enterprise profile, the terminology granularity and level of detail of the natural language template are adaptively adjusted. For example, for large enterprises or those in specialized fields such as finance and technology, more professional and refined terminology is adopted to form explanatory texts and provide in-depth analysis; while for SMEs or traditional industry enterprises, more concise, straightforward, and easy-to-understand suggestions are output.

[0102] This generation process can also be defined as the steps of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching, including:

[0103] The enterprise feature vector and the product feature vector in the matching results are compared dimensionally, and the difference between the enterprise feature vector and the product feature vector is determined based on the comparison results.

[0104] Based on the aforementioned gap, the strategy rule base is invoked to generate financial optimization suggestions;

[0105] Strategic recommendations for the personalized financial product recommendation scheme are generated by using real-time financial market data and the aforementioned financial optimization suggestions.

[0106] In the process of generating personalized financial product recommendation schemes based on multi-round progressive matching results, further guiding strategic suggestions are generated. During the generation of these strategies, the enterprise feature vectors from the matching results are placed in the same vector space for detailed dimensional comparison. By calculating the differences in key dimensions, the specific gap between the enterprise's current state and the ideal requirements of the product is quantitatively identified.

[0107] Based on the identified specific gaps, a pre-defined strategy rule base is invoked. This rule base contains strategies and knowledge for various typical financial optimization scenarios, such as cash flow management, capital structure optimization, and financing cost control. Furthermore, based on the specific dimensions and magnitude of the gaps, a corresponding optimization suggestion generation mechanism is automatically matched and triggered.

[0108] Simultaneously, the system acquires real-time financial market data, including market interest rates, regulatory policies, and industry prosperity indices, through an interface. It then generates final strategic recommendations by integrating the aforementioned quantitative gap analysis, financial optimization suggestions, and real-time market information. These recommendations include a list of recommended products and further provide enterprises with targeted guidance on how to improve their financial situation to better suit financial products and how to seize the optimal application opportunity based on the current market environment. This significantly enhances the practicality, timeliness, and strategic value of the recommended solutions.

[0109] Step S70: Output the personalized financial product recommendation scheme.

[0110] The generated personalized financial product recommendations are then output. This output process is also a delivery process, ensuring interpretability and traceability. Specifically, the output process includes: application programming interface (API) transmission for machine system integration, seamlessly connecting the recommendations to the financial institution's subsequent business process systems using standardized data formats; and a graphical interface display for business personnel or enterprise users, transforming professional content into more intuitive and easy-to-understand charts, text, and hierarchical menus through visual dashboards or interactive reporting interfaces.

[0111] The specific output scheme clearly lists the recommended product names and their key attributes through a financial product recommendation list. The output also includes detailed decision-making support information, clarifying the internal logic of the product recommendations. This output includes explanations of core indicators used to quantify the degree of matching, the weighting and rationale assigned to each enterprise dimension during model fusion, and an overall confidence assessment of the reliability of the recommendation conclusions. Through this output, users not only understand the recommendation results themselves but also gain a deeper understanding of the reasons behind them, effectively enhancing the transparency and credibility of the recommendation system. This clearly structured and comprehensive output mechanism ensures that personalized recommendation schemes can be successfully translated from theoretical calculations into practically guiding business decision-making bases.

[0112] In this embodiment, a comprehensive enterprise profile is constructed through multi-agent collaboration, and product recommendations are provided by combining knowledge graphs to achieve accurate product matching. This effectively improves the accuracy and personalization of financing recommendations and significantly increases the success rate of financing for SMEs.

[0113] Further details can be found here. Figure 2 , Figure 2 This is a detailed step diagram of step S20 above, as shown. Figure 2Including steps S21-S24, the heterogeneous dedicated intelligent agents include a basic information intelligent agent, a financial analysis intelligent agent, an operational status intelligent agent, a risk assessment intelligent agent, and a demand analysis intelligent agent. The step of distributing the enterprise characteristic data in parallel to multiple heterogeneous dedicated intelligent agents in a multi-agent collaborative architecture for dynamic collaborative processing includes:

[0114] Step S21: The hierarchical industry classification model and industry knowledge graph pre-set by the basic information intelligent agent are used to perform semantic matching on the enterprise feature data, generate enterprise sub-industry labels, and share the enterprise sub-industry labels with the financial analysis intelligent agent and the business status intelligent agent in real time.

[0115] Step S22: The financial analysis agent and the operational status agent simultaneously receive the enterprise's subdivided industry tags, process the enterprise's subdivided industry tags to obtain a financial health score and a life cycle stage identifier, and share the financial health score and operational stage identifier with the demand analysis agent and the risk assessment agent.

[0116] Step S23: The risk assessment agent selects the corresponding risk assessment model according to the received life cycle stage identifier, scans the financial indicators processed by the financial analysis agent through the risk assessment model, and generates a dynamic risk assessment report from the scanning results.

[0117] Step S24: The demand analysis agent comprehensively processes the enterprise's subdivided industry tags, life cycle stage identifiers, and financial health scores, and predicts the scale and preferences of the enterprise's potential financing needs through multiple regression analysis.

[0118] In this embodiment, standardized enterprise feature data is distributed in parallel to multiple heterogeneous dedicated intelligent agents with different functions within a multi-agent collaborative architecture. These heterogeneous dedicated intelligent agents include a basic information agent, a financial analysis agent, an operational status agent, a risk assessment agent, and a demand analysis agent. Each agent dynamically and collaboratively analyzes the enterprise data based on its respective specialized processing algorithm. The standardized enterprise feature dataset is synchronously transmitted to all heterogeneous dedicated intelligent agents through an underlying distributed communication mechanism, ensuring that each agent processes data simultaneously. This avoids the data blocking and latency issues inherent in traditional serial processing, supporting the system's high efficiency and real-time response. Furthermore, the corresponding heterogeneous dedicated intelligent agent can be selected based on the specific data type of the enterprise feature data. That is, the heterogeneous dedicated intelligent agent is used to process the corresponding type of enterprise feature data; therefore, the corresponding heterogeneous dedicated intelligent agent can be selected based on the specific data type of the enterprise feature data.

[0119] Based on the dynamic collaborative processing of multiple heterogeneous specialized intelligent agents, the enterprise feature data can be processed sequentially, specifically:

[0120] The basic information intelligent agent, based on a pre-configured hierarchical industry classification model and industry knowledge graph, performs deep semantic analysis and matching on enterprise feature data such as business scope and registration information to accurately generate enterprise-specific industry tags. Based on these generated enterprise-specific industry tags, they are shared with the financial analysis intelligent agent and the operational status intelligent agent via a real-time data exchange protocol, providing crucial industry context for subsequent analysis.

[0121] After simultaneously receiving the enterprise's subdivided industry tags, the financial analysis agent and the operational status agent initiate specialized analysis tasks. The financial analysis agent, based on the industry benchmarks and characteristics corresponding to the enterprise's subdivided industry tags, invokes a financial indicator calculation engine to perform in-depth calculations and trend analysis on multiple financial data points of the enterprise, ultimately outputting a dynamic financial health score. Simultaneously, the operational status agent, referencing the development patterns of similar industry enterprises, determines the enterprise's current development stage based on methods such as time series analysis and generates a corresponding lifecycle stage identifier. Based on the generated financial health score and lifecycle stage identifier, these are immediately shared with the demand analysis agent and the risk assessment agent.

[0122] Upon receiving the lifecycle stage identifier, the risk assessment agent adaptively selects a suitable risk assessment model from its model library based on the indicated enterprise development stage. Using the selected risk assessment model, the agent performs multi-scenario scanning and stress testing on various financial indicators processed by the financial analysis agent, generating a comprehensive and dynamic risk assessment report tailored to the characteristics of the enterprise's stage.

[0123] Meanwhile, the demand analysis intelligent agent comprehensively gathers and processes intermediate results from multiple sources, including enterprise-specific industry tags, lifecycle stage indicators, and financial health scores. Applying predictive algorithms such as multiple regression analysis, and integrating common industry patterns with enterprise-specific characteristics, it quantitatively predicts the scale and preferences of enterprises' potential financing needs.

[0124] In this embodiment, through the aforementioned parallel processing and dynamic collaboration mechanism of real-time interaction, each heterogeneous specialized intelligent agent fully leverages its in-depth analytical capabilities in its professional field. Through continuous sharing and feedback of intermediate data, an organic analytical whole is formed, providing multi-dimensional, in-depth, and highly collaborative evaluation results for the construction of a comprehensive enterprise profile.

[0125] Further, see Figure 3 , Figure 3 This is a detailed step diagram of step S50 above, as shown. Figure 3The process includes steps S51-S52, which involves vectorizing the comprehensive enterprise profile to obtain enterprise feature vectors, and mapping these enterprise feature vectors to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching.

[0126] Step S51: Set corresponding matching dimension parameters for each round of the multi-round progressive matching stage using a pre-configured progressive matching strategy;

[0127] Step S52: Perform multi-round progressive matching in the high-dimensional vector space based on the matching dimension parameters.

[0128] In this embodiment, after vectorizing the comprehensive enterprise profile and mapping it to a high-dimensional vector space, multiple rounds of progressive matching are performed to accurately screen financial products. In this multi-round progressive matching, a pre-configured progressive matching strategy sets specific matching dimension parameters for each of the sequential matching stages, thereby improving the efficiency and accuracy of matching through a layered and progressively refined screening logic.

[0129] The progressive matching strategy specifically involves configuring matching dimension parameters for the core screening dimensions in the first, second, and final matching rounds. Specifically, the first round of matching is based on hard constraint screening, with matching dimension parameters set as basic entry conditions, such as the minimum registered capital requirement for financial products, years of continuous operation, and specific industry access lists. This quickly eliminates candidate products that do not meet the basic conditions, significantly narrowing the candidate pool. The second round of matching enters the quantitative scoring screening stage, where matching dimension parameters are set as core quantitative indicators that best measure the fit between the company and the product, such as the company's debt-to-equity ratio, cash flow fluctuations, and revenue growth rate. In the second round of matching, products that pass the initial screening are finely ranked through weighted similarity calculations. The final round of matching is a dynamic strategy adjustment, where matching dimension parameters are used to integrate real-time changes in external market factors, such as the latest fluctuations in industry market interest rates and temporarily introduced interest subsidy policy coefficients, to ensure that the recommendation results sensitively reflect market dynamics.

[0130] Subsequently, based on the matching dimension parameters set in the above-mentioned hierarchical configuration, multiple rounds of progressive matching are actually performed in the high-dimensional vector space. The entire matching process strictly follows a progressive strategy: the first round of matching uses mechanisms such as Boolean operations to quickly filter all products based on hard constraint parameters, generating a preliminary subset of products that meet the conditions. The second round of matching is based on the results of the first round, deeply calculating the weighted similarity between the enterprise feature vector and each product vector in the product subset on a specific quantitative dimension, and sorting them according to the scores to generate a priority product list. The final round of matching introduces dynamic market parameters as adjustment factors to finally calibrate and re-sort the priority product list generated in the second round, such as improving the ranking of products that are more advantageous in the current market environment.

[0131] By employing a progressive matching mechanism, the matching process is ensured to consider both basic admission criteria and deep integration of advanced quantitative features and real-time market changes. This not only guarantees processing efficiency but also significantly improves the accuracy, timeliness, and practicality of the final recommendation results.

[0132] Furthermore, the implementation of the aforementioned multi-round progressive matching, namely the step of performing multi-round progressive matching based on the matching dimension parameter in the high-dimensional vector space, includes:

[0133] The first round of matching is performed based on hard constraints, and Boolean filtering is conducted through the set basic admission conditions. A preliminary set of products is generated based on the filtering results.

[0134] The second round of matching is performed based on the quantitative scoring dimension. The weighted cosine similarity between the enterprise feature vector and the feature vector of each financial product in the initial screening product set on the core dimension is calculated. The target product sequence is generated based on the weighted cosine similarity.

[0135] The final matching round is performed based on the strategy optimization dimension. Real-time supply data from the financial market and policy guidance factors are introduced as dynamic adjustment weights to comprehensively reorder the target product series. The final matching result is obtained by calculating the comprehensive reordering result through a comprehensive risk-return model.

[0136] In this embodiment, during the process of performing multiple rounds of progressive matching based on the matching dimension parameters, a phased and hierarchical fine-tuning strategy is adopted to gradually refine the matching results in the high-dimensional vector space composed of enterprise feature vectors and financial product vectors.

[0137] The entire matching process begins with the first round of matching, which is based on hard constraints. A strict Boolean logic screening is performed based on a set of preset basic eligibility criteria, such as core hard indicators like registered capital, years of continuous operation, specific industry affiliation, and minimum credit rating. The goal of this stage is to quickly filter out all financial products that do not meet the basic eligibility requirements, thereby significantly narrowing the candidate pool and generating a preliminary set of products that meet the initial threshold requirements, laying the foundation for subsequent in-depth matching.

[0138] After obtaining the initial product set, the second round of matching begins. This second round is based on a quantitative scoring dimension, calculating the weighted cosine similarity between the vectorized enterprise feature vector and the feature vector of each financial product in the initial set across multiple core dimensions. Furthermore, corresponding weights of the cosine similarity function are assigned to different dimensions, ultimately calculating a comprehensive matching score for each product. All products are then ranked using these comprehensive matching scores to generate a target product sequence arranged by matching priority.

[0139] The final round of matching is initiated. This final round of matching is based on strategy optimization, and the matching results are calibrated by introducing external dynamic factors. Real-time dynamic supply data from the financial market and the latest policy guidance factors are acquired and quantified into dynamically adjusted weights. These dynamically adjusted weights are then used to reorder the target product sequence generated in the second round. This reordering process is implemented using a comprehensive risk-return model. This model calculates a fully optimized final matching result list by weighing the potential returns, risk levels, and alignment with the current market environment and policy guidance of each product, thereby ensuring that the recommended results are accurate, timely, and policy-compliant.

[0140] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the financial resource recommendation method for small and medium-sized enterprises in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0141] This application provides a financial resource recommendation device for small and medium-sized enterprises (SMEs). The device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the financial resource recommendation method for SMEs described in Embodiment 1 above.

[0142] The following is for reference. Figure 4The diagram illustrates a structural schematic of a financial resource recommendation device for SMEs suitable for implementing embodiments of this application. The financial resource recommendation device for SMEs in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The financial resource recommendation device for SMEs shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0143] like Figure 4 As shown, the financial resource recommendation device for SMEs may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the financial resource recommendation device for SMEs. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the financial resource recommendation device for SMEs to exchange data with other devices wirelessly or via wired communication. Although the figure shows a financial resource recommendation device for SMEs with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0144] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0145] The financial resource recommendation device for SMEs provided in this application adopts the financial resource recommendation method for SMEs in the above embodiments, which can solve the technical problems of information asymmetry and low accuracy of financial product matching in existing SME financing. Compared with the prior art, the beneficial effects of the financial resource recommendation device for SMEs provided in this application are the same as those of the financial resource recommendation method for SMEs provided in the above embodiments, and other technical features of the financial resource recommendation device for SMEs are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0146] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0147] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0148] This application provides a storage medium, which is a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the financial resource recommendation method for small and medium-sized enterprises in the above embodiments.

[0149] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0150] The aforementioned computer-readable storage medium may be included in a financial resource recommendation device for SMEs; or it may exist independently and not be assembled into a financial resource recommendation device for SMEs.

[0151] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the financial resource recommendation device for SMEs, the financial resource recommendation device for SMEs implements the technical content of the above-described embodiment of the financial resource recommendation method for SMEs.

[0152] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0154] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0155] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for recommending financial resources for SMEs. This solves the technical problems of information asymmetry and low accuracy in matching financial products in existing SME financing systems. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the financial resource recommendation method for SMEs provided in the above embodiments, and will not be elaborated upon here.

Claims

1. A method for recommending financial resources to small and medium-sized enterprises, characterized in that, The method for recommending financial resources to SMEs includes the following steps: Obtain standardized enterprise characteristic data; The enterprise feature data is distributed in parallel to multiple heterogeneous dedicated agents in the multi-agent collaborative architecture for dynamic collaborative processing. Based on the processing results, a confidence score, a dynamic risk assessment report, and a financial health score are obtained. The model parameters of the confidence-weighted fusion model are deployed using the confidence score, and the enterprise dimension contribution weights of the dynamic risk assessment report and the financial health score are adjusted based on the confidence-weighted fusion model. A comprehensive enterprise profile is generated based on the adjustment results of the enterprise dimension contribution weights. The comprehensive enterprise profile is vectorized to obtain enterprise feature vectors, and these enterprise feature vectors are mapped to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching. A pre-configured progressive matching strategy is used to set corresponding matching dimension parameters for each round of the multi-round progressive matching. Multi-round progressive matching is then performed in the high-dimensional vector space based on these matching dimension parameters. A personalized financial product recommendation scheme is generated based on the matching results of the multi-round progressive matching. Output the personalized financial product recommendation scheme; The step of performing multi-round progressive matching based on the matching dimension parameter in the high-dimensional vector space includes: The first round of matching is performed based on hard constraints, and Boolean filtering is conducted through the set basic admission conditions. A preliminary set of products is generated based on the filtering results. The second round of matching is performed based on the quantitative scoring dimension. The weighted cosine similarity between the enterprise feature vector and the feature vector of each financial product in the initial screening product set on the core dimension is calculated. The target product sequence is generated based on the weighted cosine similarity. The final matching round is performed based on the strategy optimization dimension. Real-time supply data from the financial market and policy guidance factors are introduced as dynamic adjustment weights to comprehensively reorder the target product series. The final matching result is obtained by calculating the comprehensive reordering result through a comprehensive risk-return model.

2. The financial resource recommendation method for SMEs as described in claim 1, characterized in that, The heterogeneous dedicated intelligent agents include basic information intelligent agents, financial analysis intelligent agents, operational status intelligent agents, risk assessment intelligent agents, and demand analysis intelligent agents. The steps of distributing the enterprise characteristic data in parallel to multiple heterogeneous dedicated intelligent agents in a multi-agent collaborative architecture for dynamic collaborative processing include: The hierarchical industry classification model and industry knowledge graph pre-set by the basic information intelligent agent are used to perform semantic matching on the enterprise feature data, generate enterprise sub-industry tags, and share the enterprise sub-industry tags with the financial analysis intelligent agent and the business status intelligent agent in real time. The financial analysis agent and the operational status agent simultaneously receive the enterprise's subdivided industry tags, process the enterprise's subdivided industry tags to obtain a financial health score and a life cycle stage identifier, and share the financial health score and operational stage identifier with the demand analysis agent and the risk assessment agent. The risk assessment agent selects the corresponding risk assessment model based on the received life cycle stage identifier, scans the financial indicators processed by the financial analysis agent through the risk assessment model, and generates a dynamic risk assessment report from the scanning results. The demand analysis agent comprehensively processes the enterprise's subdivided industry tags, life cycle stage identifiers, and financial health scores, and predicts the scale and preferences of the enterprise's potential financing needs through multiple regression analysis.

3. The financial resource recommendation method for SMEs as described in claim 1, characterized in that, The step of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching includes: Key decision data is extracted during the multi-round progressive matching process, and the key decision data is mapped and filled with a preset natural language template. Based on the enterprise size and industry attributes in the comprehensive enterprise profile, the terminology granularity and level of detail of the natural language template are adaptively adjusted to generate explanatory text for the personalized financial product recommendation scheme.

4. The financial resource recommendation method for SMEs as described in claim 1, characterized in that, The step of generating a personalized financial product recommendation scheme based on the matching results of the multi-round progressive matching includes: The enterprise feature vector and the product feature vector in the matching results are compared dimensionally, and the difference between the enterprise feature vector and the product feature vector is determined based on the comparison results. Based on the aforementioned gap, the strategy rule base is invoked to generate financial optimization suggestions; Strategic recommendations for the personalized financial product recommendation scheme are generated by using real-time financial market data and the aforementioned financial optimization suggestions.

5. The financial resource recommendation method for SMEs as described in claim 1, characterized in that, Before the step of vectorizing the comprehensive enterprise profile to obtain an enterprise feature vector, and mapping the enterprise feature vector to the high-dimensional vector space of the financial product knowledge graph for multi-round progressive matching, the method further includes: Information on multiple financial products is obtained through a pre-defined API interface, and the information on multiple financial products is parsed and standardized to generate financial product feature information. Based on a predefined ontology model, the feature information of the financial products is mapped to nodes and attributes in a knowledge graph to form the financial product knowledge graph.

6. The financial resource recommendation method for SMEs as described in claim 5, characterized in that, The step of mapping the financial product feature information to nodes and attributes in a knowledge graph based on a predefined ontology model to form the financial product knowledge graph includes: Calculate the multidimensional feature vectors between each product node in the knowledge graph; Based on the dimensions of the multidimensional feature vector, corresponding product association relationships are created, and the product association network of the knowledge graph is constructed through the product association relationships.

7. A financial resource recommendation device for small and medium-sized enterprises, characterized in that, The financial resource recommendation device for SMEs stores a computer program, which, when executed by a processor, implements the financial resource recommendation method for SMEs as described in any one of claims 1-6.

8. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the financial resource recommendation method for small and medium-sized enterprises as described in any one of claims 1-6.

Images