Method and system for generating financial information graph of user through financing platform

Abstract

The invention discloses a method and system for generating a financial information graph of a user through a financing platform, and relates to the technical field of financial information. The method is used for solving the problems of weak user operation data association and inaccurate risk assessment in a traditional financing platform. Firstly, multi-dimensional analysis based on transaction semantic recognition is performed on a user bank flow, transaction scene features are extracted, and a supply chain core enterprise fund flow path is recognized; then, a cash flow-supply chain cooperation efficiency coefficient is calculated in combination with order account period information, and enterprise fund turnover efficiency is quantified; then, constructing a dynamic weighted supply chain network topology based on the coefficient, and identifying key fragile nodes through a graph neural network; and finally, fusing cooperation efficiency and risk characteristics, predicting enterprise performance capability through a dynamic attention mechanism, generating an interactive dynamic information graph of a fund flow path, a supply chain state and a risk result, and realizing intelligent analysis and visual presentation of a financing platform on an enterprise operation state.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of financial information technology, in particular to a method and system for generating a user's financial information graph through a financing platform. BACKGROUND

[0002] With the popularization of financing platforms and digital credit services, the financing activities of enterprises and individuals show the characteristics of high frequency, fragmentation and diversification. The matching process between the supply and demand of funds gradually shifts from offline manual review to online intelligent analysis based on data driving. Financing platforms integrate users' financing records, repayment information and credit performance to form rich data assets, providing a new data basis for financial risk management and credit assessment. At the same time, users' financing behavior is no longer a single and isolated event, but a complex dynamic process containing time series characteristics, credit behavior changes and market environment influences. How to identify behavior patterns from financing data through intelligent algorithms and reveal the correlation between financing cycle, interest rate fluctuations and credit characteristics in a visual form has become an important research direction in the field of financial technology.

[0003] For example, the Chinese invention patent with publication number CN120045883B, specifically a method and system for generating a user's financial information graph through a financing platform, includes the following steps: based on user behavior data, extracting the financing amount, financing time, financing interest rate and repayment period in the user's financing record, and calling the timestamp to sort the financing behavior record in chronological order. In the present application, by dynamically extracting the amount, time, interest rate and repayment period in the user's financing data and integrating time series analysis, continuous tracking and deep mining of financing behavior are realized, the fluctuation rules of financing amount and interest rate are effectively identified, and by combining user repayment behavior and credit characteristics, a feature coupling matrix and a time weight distribution model are generated, which significantly improves the accuracy and comprehensiveness of user behavior prediction. By combining dynamic parameters, time weights and credit characteristics, an intuitive financial information graph can be generated, providing clearer data support and visual expression for financing decisions.

[0004] However, this type of method still has substantial deficiencies in technical implementation: its analysis range mainly focuses on the financing behavior records of a single user, ignoring the structural association between user's fund flow and their supply chain or transaction network, making it difficult to reveal the cross-subject fund operation logic; secondly, the feature coupling matrix generated by this method mainly reflects the statistical relationship between financing parameters, without introducing deep feature expression based on semantic understanding or graph structure learning, resulting in limited model explanation of complex financing behavior; in addition, the generated information graph is mostly in static display form, lacking multi-dimensional interaction and time evolution analysis capability, making it difficult to reflect the dynamic change rules of financing behavior in continuous cycles, thus limiting the application effect in risk identification and performance prediction scenarios. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a method and system for generating a user's financial information graph through a financing platform, which solves the problems of the above background art.

[0006] To achieve the above object, the present application is implemented by the following technical scheme: a method for generating a user's financial information graph through a financing platform, comprising the following steps: S1. performing multi-dimensional analysis on user bank flow data based on transaction semantic recognition, extracting transaction scenario features in the flow summary through natural language processing technology, identifying the fund flow path related to the core enterprise of the supply chain, and extracting the periodic fluctuation characteristics of operating cash flow based on time series analysis; S2. based on the fund flow path and the periodic fluctuation characteristics, combining the order account period information in the supply chain transaction data, analyzing the matching relationship between the fund payment time and the order delivery cycle, constructing a cash flow-supply chain coordination efficiency coefficient, and quantifying the fund turnover efficiency of enterprise operation cycle; S3. taking the cash flow-supply chain coordination efficiency coefficient as the weight of each enterprise entity node in the supply chain network, constructing a dynamically weighted supply chain network topology, learning the embedded representation of the network node through a graph neural network, and identifying the key vulnerable nodes in the supply chain network by analyzing the topological structure change of the weighted network in the continuous time window; S4. cross-dimension feature fusion of the key vulnerable nodes and the coordination efficiency coefficient, through a dynamic attention mechanism to evaluate the contribution weight of different risk factors, predict the comprehensive performance ability of the user enterprise, and generate an interactive dynamic information graph containing fund flow path, supply chain network state and risk prediction results.

[0007] Further, the multi-dimensional analysis of the user bank flow data based on transaction semantic recognition, the extraction of the transaction scenario features in the flow summary through natural language processing technology, and the identification of the fund flow path related to the core enterprise of the supply chain are as follows: according to named entity recognition, the transaction counterparty name, commodity service type and transaction purpose key elements are extracted from the flow summary; the semantic similarity between the transaction summary and the supply chain typical scene is calculated through a pre-trained domain word vector model, and the transaction scenarios including purchase payment, sales return and logistics settlement are identified; the fund flow direction law between transaction counterparties is analyzed through association rule, a fund flow network atlas with the core enterprise as the hub is constructed, and the key path with direct or indirect fund flow with the core enterprise is identified according to the atlas connectivity analysis.

[0008] Further, the specific process of extracting the periodic fluctuation characteristics of operating cash flow based on time series analysis is as follows: multi-scale decomposition is performed on the operating cash flow sequence to obtain periodic terms representing different time scales, and autocorrelation function analysis is used to identify significant periodic lengths corresponding to business cycles in the cash flow sequence; based on the periodic terms and the identified periodic lengths, the amplitude characteristics of cash flow fluctuations are calculated, including the range and standard deviation of fluctuations within a period; based on the time series distribution of the periodic terms, phase characteristics are extracted, including the starting time point and peak position of the periodic fluctuations; based on the stability analysis of consecutive periodic terms, a periodic consistency index is calculated to represent the sustainability characteristics of the enterprise operating cycle.

[0009] Further, step S2 includes the following steps: based on the fund flow path and the periodic fluctuation characteristics, the order account period information in the supply chain transaction data is combined to establish the mapping relationship between the fund collection and payment time and the order delivery cycle; the dynamic matching degree of the fund turnover cycle and the business operation cycle is calculated to quantify the coordination efficiency in the time dimension, and the stability characteristics of the coordination efficiency are evaluated through multi-time scale sliding window analysis; the periodic consistency index and the time series matching degree characteristics are fused to construct a multi-dimensional cash flow-supply chain coordination efficiency coefficient.

[0010] Further, the cash flow-supply chain coordination efficiency coefficient is used as the weight of each enterprise entity node in the supply chain network to construct a dynamically weighted supply chain network topology as follows: based on the upstream and downstream transaction relationships in the supply chain, an initial network structure is constructed, where nodes represent enterprise entities and edges represent transaction relationships, and the cash flow-supply chain coordination efficiency coefficient of each enterprise entity is used as the node weight to reflect the fund turnover efficiency of the enterprise in the supply chain; the edge weights are dynamically adjusted according to the transaction amount and the transaction frequency, and high-frequency large-amount transactions correspond to higher edge weights; a network adjacency matrix is constructed containing node weights and edge weights, where the diagonal elements store the node weights and the non-diagonal elements store the edge weights, and by introducing the time dimension, a dynamic network topology sequence is constructed under a continuous time window to capture the evolution process of the network structure.

[0011] Further, the specific process of learning the embedding representation of network nodes through graph neural network and identifying the key vulnerable nodes in the supply chain network by analyzing the topological structure changes of the weighted network within a continuous time window is as follows: based on the dynamically weighted supply chain network topology, the network structure characteristics of each time slice are extracted, the degree centrality change trend of the nodes within the continuous time window is calculated, and the nodes with unstable connection relationships are identified; the propagation influence of the nodes is analyzed, the importance of the nodes is evaluated by simulating the influence of node failure on network connectivity, and the nodes with continuously declining coordination efficiency coefficients and key network positions are identified by combining the time series changes of node weights; by integrating the topological structure changes and the evolution of node weights, the nodes with high network importance and low coordination efficiency are determined as the key vulnerable nodes.

[0012] Further, the key vulnerable node recognition is combined with the coordination efficiency coefficient across the dimensional features, and the contribution weight of different risk factors is evaluated through a dynamic attention mechanism to predict the specific process of the comprehensive performance ability of the user enterprise as follows: a key vulnerable node feature vector is constructed, including the centrality index of the node in the network, the duration of vulnerability, and the impact range feature. The cash flow-supply chain coordination efficiency coefficient is combined with the key vulnerable node features in multiple dimensions to form a comprehensive feature matrix. The importance weight of different dimensions in the feature matrix is calculated through a multi-head attention mechanism, wherein the query vector is dynamically adjusted based on the historical performance. The deep interaction between the key vulnerable node features and the coordination efficiency coefficient is learned through a feature cross layer to generate a cross feature representation. The comprehensive feature matrix is weighted and fused based on the attention weight to highlight the feature dimensions with high contribution to the performance prediction. The weighted feature representation is input into a fully connected neural network, and the Softmax function is used to output the probability distribution of the comprehensive performance ability of the user enterprise in the future specific period.

[0013] Further, the specific process of generating an interactive dynamic information map containing the fund flow path, supply chain network state and risk prediction results is as follows: a hierarchical visualization architecture is constructed, including three main views of the basic network layer, the risk identification layer and the prediction display layer. In the basic network layer, the supply chain network topology is displayed through a force-directed layout algorithm, the size of the node represents the enterprise scale, the color depth of the node represents the coordination efficiency coefficient, and the edge thickness represents the transaction intensity. In the risk identification layer, the identified key vulnerable nodes are marked with a highlighted frame and a pulse animation, and their vulnerability score and impact range are displayed. In the prediction display layer, the comprehensive performance ability score is displayed through a ring-shaped dashboard, the historical score change is displayed with a trend curve, and the risk level is distinguished by color coding. A multi-layer linkage interaction mechanism is designed, and when the user clicks on any enterprise node, all fund flow paths, associated trading parties and risk transmission paths are highlighted simultaneously.

[0014] A system for generating user financial information graphs through a financing platform includes the following modules: a data parsing and feature extraction module, used to perform multi-dimensional analysis of user bank statement data based on transaction semantic recognition, extract transaction scenario features from statement summaries using natural language processing technology, identify the capital flow path related to core enterprises in the supply chain, and extract the periodic fluctuation characteristics of operating cash flow based on time series analysis; and a collaborative efficiency analysis module, used to analyze the matching relationship between fund collection and payment time and order delivery cycle based on capital flow path and periodic fluctuation characteristics, combined with order payment period information in supply chain transaction data, construct a cash flow-supply chain collaborative efficiency coefficient, and quantify the cash cycle of enterprise operations. The system includes several modules: a supply chain network analysis module, a risk prediction and visualization module, and a risk prediction and visualization module. The supply chain network analysis module uses the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network to construct a dynamically weighted supply chain network topology. It learns the embedded representation of network nodes through a graph neural network and identifies key vulnerable nodes in the supply chain network by analyzing the changes in the topology of the weighted network within a continuous time window. The risk prediction and visualization module integrates the identification of key vulnerable nodes with the collaboration efficiency coefficient across dimensions. It evaluates the contribution weight of different risk factors through a dynamic attention mechanism, predicts the comprehensive performance capability of user enterprises, and generates an interactive dynamic information graph that simultaneously includes the cash flow path, supply chain network status, and risk prediction results.

[0015] The present invention has the following beneficial effects:

[0016] (1) A method for generating a user's financial information graph through a financing platform. By performing multi-dimensional analysis of user bank transaction data based on transaction semantic recognition, combined with natural language processing technology and time series analysis, this method achieves semantic understanding and cyclical feature extraction of enterprise operating cash flow. This method can identify real transaction scenarios from transaction summaries, construct cash flow paths associated with core enterprises in the supply chain, and reflect the dynamic stability of enterprise cash flow through cyclical fluctuation analysis. This invention introduces transaction semantics and business context into feature expression, significantly improving the accuracy of identifying enterprise operating behavior and the interpretability of data. Furthermore, by constructing a cash flow-supply chain collaborative efficiency coefficient and establishing a mapping relationship between fund receipt and payment time and order delivery cycle, it achieves integrated analysis of enterprise capital turnover and business execution rhythm, effectively revealing the temporal coordination characteristics of enterprise operating cycles and providing a quantitative basis for capital efficiency assessment.

[0017] (2) A system for generating a financial information graph of users through a financing platform. This invention uses the cash flow-supply chain collaboration efficiency coefficient as node weights to construct a dynamically weighted supply chain network topology. It also learns the embedding representations of enterprise nodes at different time periods through a graph neural network, achieving adaptive capture of changes in the supply chain network structure. This design can identify key vulnerable nodes in situations of declining capital turnover efficiency or unstable transaction relationships, thereby revealing potential risk propagation channels in the supply chain. By fusing the characteristics of key vulnerable nodes with the collaboration efficiency coefficient across dimensions, and employing a dynamic attention mechanism to evaluate the impact weights of different risk factors, the accuracy and interpretability of predicting the comprehensive performance capabilities of enterprises are improved. The resulting interactive dynamic financial information graph presents the capital flow path, supply chain network status, and risk prediction results in a unified manner, making the risk structure visible and its evolution process traceable, providing real-time decision support and risk insight capabilities for the financing platform.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for generating a user's financial information graph through a financing platform according to the present invention.

[0020] Figure 2 This is a system flowchart of the present invention for generating a user's financial information graph through a financing platform. Detailed Implementation

[0021] This application embodiment solves the problems of single dimension of financing data analysis, weak capital correlation between upstream and downstream of the supply chain, and lagging risk identification in the prior art by providing a method and system for generating financial information graphs of users through a financing platform.

[0022] The overall concept of the solution in this application embodiment is as follows:

[0023] First, by performing transaction semantic recognition and time series analysis on users' bank transaction data, the system extracts the capital flow paths related to core enterprises in the supply chain and identifies the cyclical fluctuations in operating cash flow, thereby achieving a multi-dimensional portrayal of the company's actual business activities. Second, by combining supply chain transaction data, a cash flow-supply chain synergy efficiency coefficient is constructed to quantify the degree of matching between capital turnover efficiency and order execution rhythm. Furthermore, this synergy efficiency coefficient is introduced into a dynamic supply chain network, and graph neural networks are used to analyze the temporal correlation changes between network nodes to identify potential vulnerable nodes. Finally, through multi-dimensional feature fusion and a dynamic attention mechanism, the importance of risk factors is assessed, generating an interactive financial information graph integrating capital flow paths, supply chain network status, and risk prediction results. This enables the financing platform to dynamically and visually assess the user company's ability to fulfill its obligations.

[0024] Please see Figure 1 This invention provides a technical solution: a method for generating a user's financial information graph through a financing platform, comprising the following steps: S1. Performing multi-dimensional analysis on user bank statement data based on transaction semantic recognition, extracting transaction scenario features from the statement summary using natural language processing technology, identifying the capital flow path related to the core enterprise in the supply chain, and extracting the periodic fluctuation characteristics of operating cash flow based on time series analysis; S2. Based on the capital flow path and periodic fluctuation characteristics, combined with order payment period information in the supply chain transaction data, analyzing the matching relationship between the time of capital receipt and payment and the order delivery cycle, constructing a cash flow-supply chain collaborative efficiency coefficient, and quantifying the enterprise... S3. Using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network, a dynamically weighted supply chain network topology is constructed. The embedded representation of network nodes is learned through graph neural networks. By analyzing the changes in the topology of the weighted network within a continuous time window, key vulnerable nodes in the supply chain network are identified. S4. The identification of key vulnerable nodes and the collaboration efficiency coefficient are fused across dimensions. The contribution weight of different risk factors is evaluated through a dynamic attention mechanism to predict the comprehensive performance capability of user enterprises and generate an interactive dynamic information graph that simultaneously includes the cash flow path, supply chain network status, and risk prediction results.

[0025] In this implementation plan, S1: Transform the original bank transaction records (timestamps, amounts, summary text, etc.) into structured events and fund paths with business semantics, providing semantic input for subsequent time series and network analysis. First, word segmentation and named entity recognition are used to extract counterparties, goods / services, and transaction purposes; then, pre-trained domain word vectors or semantic matching models are used to calculate the similarity between the summary and typical scenarios for labeling; finally, a transaction graph is constructed with the transaction entity as nodes and the fund direction as edges, and the edge weights are determined using frequency / amount rules or weighted formulas to extract the direct / indirect paths to the core enterprise from the graph connectivity. Named entity recognition (NER): Identifies named entities (such as company names, product names) from text; word vectors: map words or phrases to vectors to measure semantic similarity; edge weights: numerical values ​​representing the transaction intensity in the graph (which can be obtained by combining frequency and amount). S2: Quantify the degree of time matching between "fund receipt and payment time" and "order delivery / payment period" to generate comparable single-enterprise time-business coupling indicators (collaboration efficiency coefficient). This coefficient measures whether cash flow is synchronized with business rhythm. Input / Output: Inputs are the cash path and cash flow cycle characteristics of S1, and order payment period information; outputs are the time series of the collaborative efficiency coefficient for each enterprise under the sliding time window. First, align the cash flow series and order events with a unified time granularity (resampling or time window matching); attempt to map each payment to the corresponding order (based on the counterparty, amount range, and time window); calculate matching degree indicators (e.g., mapping success rate, payment-delivery time difference distribution statistics); synthesize the matching degree, cycle consistency, and volatility into a collaborative efficiency coefficient according to predetermined weights or a fusion model. The weights or fusion strategy are determined through historical sample optimization. Time resampling: unifies data from different time granularities to a common granularity; matching degree: a statistic measuring the degree of correspondence between payment events and order events; sliding window: an observation window that moves over time to capture short-term and long-term characteristics. S3: Inject the individual collaborative efficiency into the supply chain graph to obtain a weighted network that evolves over time and identify key nodes that are both "structurally important" and "financially vulnerable". Input / Output: Inputs include collaboration efficiency coefficients, supply chain transaction relationships, and transaction amounts / frequency; outputs include a dynamic adjacency matrix sequence, node embedding representations, and a list of critical vulnerable nodes and their vulnerability scores. Implementation Highlights: A time-slice graph is constructed using collaboration efficiency as node weights and amount / frequency as edge weights; a graph neural network (or time-series graph model) is used to learn the low-dimensional embeddings of each node to encode topology and attributes; and the systemic impact is assessed by combining centrality changes, the decline trajectory of collaboration efficiency, and failure propagation simulations to identify critical vulnerable nodes. Graph Neural Network (GNN): A model that aggregates and learns information on a graph structure, outputting node embeddings (i.e., low-dimensional vectors); Vulnerable Nodes: Nodes that perform poorly on several indicators (network importance and financial health) and whose failure amplifies systemic risk; Propagation Simulation: Simulating node failures on the graph to assess the cascading effects.S4: Integrates network layer and time layer features, calculates the contribution of each risk factor, outputs interpretable performance capability predictions, and displays the funding path, network status, and risk predictions in an interactive infographic. Input / Output: Inputs include key vulnerable node features, collaborative efficiency sequences, node embeddings, and historical performance characteristics of the enterprise; outputs include performance capability probability / risk scores and metadata (node / edge states, attention weights, etc.) used to drive visualization. Implementation Highlights: After constructing a comprehensive feature matrix, a multi-head / dynamic attention mechanism is used to calculate the weights of each dimension (the query can be dynamically adjusted based on historical performance), followed by feature cross-referencing and a fully connected prediction head to output the probability distribution; simultaneously, the attention weights and prediction results are mapped to a three-layer visualization view (basic network layer, risk identification layer, prediction layer), and interactive time sliders and node clicks are provided to replay network evolution and view factor contributions. Attention Mechanism: A module that assigns importance weights to different input dimensions; Multi-head Attention: Parallel learning of multiple interaction modes to enhance expressive power; Interactive Infographic: A clickable, time-replayable visualization interface displaying interpretive indicators. This method couples "semantic parsing of flow text" with "cash flow time characteristics" and "supply chain network structure" at the time-network level, and achieves interpretable performance prediction and interactive visualization through graph learning and attention mechanisms.

[0026] Specifically, the process of performing multi-dimensional analysis of user bank transaction data based on transaction semantic recognition, and extracting transaction scenario features from transaction summaries using natural language processing technology to identify the fund flow paths related to core enterprises in the supply chain is as follows: Key elements such as counterparty names, product / service types, and transaction purposes are extracted from the transaction summaries based on named entity recognition; the semantic similarity between the transaction summaries and typical supply chain scenarios is calculated using a pre-trained domain word vector model to identify transaction scenarios including procurement payments, sales receipts, and logistics settlements; the fund flow patterns between counterparties are analyzed using association rules to construct a fund flow network graph with the core enterprise as the hub; and key paths with direct or indirect fund transactions with the core enterprise are identified based on graph connectivity analysis.

[0027] In this implementation plan, to improve the accuracy of transaction semantic recognition, the text samples from bank transaction records that have undergone word segmentation and part-of-speech tagging are first input into a named entity recognition model to extract a set of semantic entities containing the transaction subject, transaction purpose, and the name of the goods or services. Subsequently, a pre-trained domain word vector model is used to calculate the semantic vector for each transaction record, and its similarity is compared with the labeled supply chain scenario vectors. Semantic similarity can be calculated using the cosine similarity function, expressed as: Among them, R p,q Transaction summary vector V p With typical scenario vector V qsemantic similarity; V p : A vector representation of the transaction summary, obtained through a word vector model; V q : Represents a semantic vector for typical supply chain scenarios (such as procurement payments and sales revenue collection); ||V p ||,||V q ||: These represent the magnitudes of the two vectors. When the semantic similarity R... p,q When the value exceeds a set threshold τ1, the transaction summary is considered a match for scenario q. The threshold τ1 is determined by training semantic classification on labeled historical samples and using the similarity threshold corresponding to the maximum F1 score on the validation set as the final value. After scenario recognition is completed, a transaction pair node set N = {n1, n2, ..., n} is constructed based on the counterparty names. K}, and with the fund transfer relationship as the directed edge E={(n i ,n j Construct a transaction graph. The weight w of each edge... i,j The formula for representing the combined intensity of the frequency and amount of capital flows is as follows: Among them, w i,j Node n i With n j The boundary weight between; T i,j Average transaction time interval; A i,j : Cumulative transaction amount; A max : Maximum cumulative transaction amount in the sample, used for normalization; α: Balance parameter, used to weigh the impact of transaction frequency and amount, with a value range of [0,1]. The parameter α is determined by using cross-validation to achieve the optimal ratio of network connectivity to edge weight variance. After construction, connectivity analysis is used to calculate the connection from each node to the core enterprise node n. c The set of reachable paths, with path weights selected based on the principle of least resistance, ensures that the capital flow path reflects the true capital transmission relationship between enterprises.

[0028] Specifically, the process of extracting the cyclical fluctuation characteristics of operating cash flow based on time series analysis is as follows: The operating cash flow series is decomposed into multiple scales to obtain periodic terms representing different time scales; significant period lengths corresponding to the business cycle are identified in the cash flow series through autocorrelation function analysis; based on the periodic terms and the identified period lengths, the amplitude characteristics of cash flow fluctuations are calculated, including the range and standard deviation of fluctuations within the cycle; based on the temporal distribution of the periodic terms, phase characteristics are extracted, including the starting point and peak position of the periodic fluctuations; based on the stability analysis of continuous periodic terms, a period consistency index is calculated to characterize the sustainability of the enterprise's operating cycle.

[0029] In this implementation plan, the user's operating cash flow time series is denoted as... Where Ct Let C represent the operating cash flow at time t. A multi-scale decomposition algorithm is used to analyze C. t Decomposition yields several periodic components {S} m Each component represents a periodic term at a different time scale. Multi-scale decomposition can be achieved using Empirical Mode Decomposition (EMD) or wavelet packet decomposition. Subsequently, the autocorrelation value r(L) under different lag lengths L is calculated using the autocorrelation function, as shown in the formula: Where, r(L): autocorrelation coefficient with lag length L; C t Operating cash flow at time t; The mean of the cash flow series; T: total number of time points. When r(L) is within a certain lag length L * When a significant peak occurs, the determination period length is L. * Based on the identified period length L * Calculate the fluctuation amplitude characteristic Δ for each period. m , defined as: Δ m =max(S) m )-min(S m ); and periodic standard deviation: Where, Δ m : The range of the m-th periodic term; σ m : The standard deviation of the m-th periodic term; This periodic term has a period length of L. * The average value within the range. Phase characteristics are used to describe the timing of the occurrence of periodic wave peaks. The phase offset we is defined. m for: Among them, we m This represents the phase position of the cyclical fluctuation, corresponding to the time offset of the cash flow peak. Finally, to measure the stability of the cycle, the cycle consistency index η is calculated. m It is defined as the average of the phase and amplitude similarity within adjacent periods: Where, η m : Periodic consistency index; we m,k : Phase of the kth period; Δ m,k : The fluctuation amplitude of the k-th period; K: The total number of periods. When η mWhen the value approaches 1, it indicates that the cyclical fluctuations are stable and highly regular. To determine the judgment threshold τ2, the inflection point method is used to determine the boundary value that causes significant clustering of the cyclical consistency distribution by analyzing the cash flow time series of multiple sample companies. In this implementation scheme, the cycle length, amplitude characteristics, phase characteristics, and cycle consistency index of operating cash flow are obtained through the above process, providing time-series input characteristics for subsequent supply chain collaborative analysis. Compared with the analysis method that only uses static financial indicators, this scheme can reveal the inherent regularity of cash flow in the time dimension and the correspondence between it and the company's operating rhythm.

[0030] Specifically, step S2 includes the following steps: based on the flow path and cyclical fluctuation characteristics of funds, and combined with the order payment period information in the supply chain transaction data, establish a mapping relationship between the time of fund collection and payment and the order delivery cycle; calculate the dynamic matching degree between the fund turnover cycle and the business operation cycle, quantify the collaborative efficiency in the time dimension, and evaluate the stability characteristics of collaborative efficiency through multi-time scale sliding window analysis; and integrate the cycle consistency index and time series matching degree characteristics to construct a multi-dimensional cash flow-supply chain collaborative efficiency coefficient.

[0031] In this implementation scheme, firstly, each transaction event in the fund transfer path obtained in step S1 is matched with the order records in the supply chain transaction data based on the counterparty identifier and transaction time, forming a correspondence between fund receipt and payment events and order delivery events. Under this correspondence, let the time series of fund receipt and payment events be {t}. a1 ,t a2 ,…,t am The corresponding order delivery event time series is {t}. b1 ,t b2 ,…,t bm Then, the average time difference and variance of the two time series can be calculated to represent the consistency between the collection and payment time and the delivery cycle. To quantitatively describe the dynamic matching degree between the capital turnover cycle and the business operation cycle, this embodiment defines a matching degree function: Among them, M r Time-matching metrics are used to quantify the synchronicity between cash flow and business flow; t au : The timestamp of the uth fund transfer event; t bu : The order delivery timestamp corresponding to the uth payment / receipt event; Δ l The maximum time span of the sample time window, used to normalize the scale of different periods. This indicator shows that the closer the fund payment and receipt activities are to the delivery cycle, the better M becomes. r A value approaching 1 indicates a high degree of coordination between fund utilization and business operations. Furthermore, to analyze the trend of coordinated changes across different time scales, the matching degree index sequence M... r (t) Multi-timescale sliding window analysis is employed. The sliding window length L...s Below, calculate the mean matching degree within the window. With standard deviation σ s The collaborative stability index is defined as: Among them, S q Collaborative stability index; Mean matching degree within the sliding window; σ s Standard deviation of the matching degree fluctuation. This indicator measures the stability of the matching between capital and business cycles at different time scales; the smaller the fluctuation, the more stable the company's operating rhythm. Finally, the cycle consistency index P... e (Operating cash flow cycle stability indicator extracted from S1) and time matching degree M r Cooperative stability S q By integrating these elements, a comprehensive cash flow-supply chain synergy efficiency coefficient E can be constructed. f The formula is: E f =α1M r +α2S q +α3P e Among them, E f The cash flow-supply chain synergy efficiency coefficient reflects the overall synergy level between funds and business operations. α1, α2, and α3 are weighting coefficients used to balance the importance of different feature dimensions. The weighting coefficients are determined by performing grid search optimization based on the synergy efficiency prediction accuracy of historical transaction samples, with maximizing prediction performance as the objective function to determine the optimal parameter combination. Through the above calculation process, a multi-dimensional characterization of the time-based synergy between enterprise fund receipts and payments and supply chain business activities is achieved, providing a quantitative foundation for the subsequent construction of a dynamic supply chain network.

[0032] Specifically, the process of constructing a dynamically weighted supply chain network topology by using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network is as follows: An initial network structure is constructed based on upstream and downstream transaction relationships in the supply chain, where nodes represent enterprise entities and edges represent transaction relationships. The cash flow-supply chain collaboration efficiency coefficient of each enterprise entity is used as the node weight, reflecting the enterprise's capital turnover efficiency in the supply chain. The edge weights are dynamically adjusted according to the transaction amount and frequency, with higher edge weights corresponding to high-frequency, large-value transactions. A network adjacency matrix containing node and edge weights is constructed, where diagonal elements store node weights and off-diagonal elements store edge weights. By introducing a time dimension, a dynamic network topology sequence under continuous time windows is constructed to capture the evolution of the network structure.

[0033] In this implementation plan, a dynamically weighted supply chain network topology is constructed by using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network. The specific process is as follows: First, an initial supply chain network is constructed based on upstream and downstream transaction records between enterprises. Nodes in this network represent enterprise entities, and edges represent pairs of enterprises with transaction relationships. For each node v... x Its node weight w x Cash flow - supply chain collaboration efficiency coefficient E obtained from step S2 f (x) is assigned a value to characterize the company's capital turnover efficiency and business coordination ability. In determining the edge weights, let company v... x With v y If there are multiple transaction events, then define the edge weight: r xy =λ1·freq xy +λ2·val xy ; where r xy Enterprise v x With v y Edge weights between; freq xy : The frequency of transactions between the two companies within the statistical time window; val xy λ1 and λ2 are the average transaction amounts between the two companies, used to balance the importance of transaction frequency and transaction amount, determined through a minimum error search based on historical samples. Subsequently, a weighted adjacency matrix containing node and edge weights is constructed: A = [a...]. xy ] n×n , Where A: the weighted adjacency matrix of the supply chain network; a xy : The weighted value of a node or edge in the network; n: The total number of enterprise nodes in the network. To capture the dynamic evolution of the network over time, a time dimension t is introduced, within a continuous time window [t1, t2, ..., t]. T Generate a dynamic adjacency matrix sequence within [the context]. Each A (t) Representing the supply chain structure within a specific time period, continuous sequences are used to reflect the evolution of capital flow and transaction structure over time, providing dynamic input features for subsequent identification of vulnerable nodes.

[0034] Specifically, the process of learning the embedded representation of network nodes through graph neural networks and identifying key vulnerable nodes in the supply chain network by analyzing the topological changes of the weighted network within continuous time windows is as follows: Based on the dynamically weighted supply chain network topology, network structure features are extracted for each time slice, the degree centrality change trend of nodes within continuous time windows is calculated, and nodes with unstable connections are identified; the propagation influence of nodes is analyzed, and their importance is assessed by simulating the impact of node failure on network connectivity; combined with the time series changes of node weights, nodes with continuously decreasing collaborative efficiency coefficients and critical network positions are identified; and by comprehensively considering topological changes and node weight evolution, nodes with high network importance and low collaborative efficiency are identified as key vulnerable nodes.

[0035] In this implementation scheme, firstly, the dynamically weighted network topology is input into the graph neural network (GNN) model to learn the structural embedding representation of each node in the time dimension. Let the node features at time t be... Its update formula is: in, node v x The eigenvector at time t; node v x The set of neighbors; Neighbor node v y For node v x The propagation weight, based on the edge weight r xy Calculation of the rate of change within the time window; W h W s : Convolution weight matrix; σ(·): Non-linear activation function. Propagation weights Determined adaptively in the following manner: in, Boundary weights at time t; The efficiency difference between nodes; δ: attenuation coefficient, used to adjust the impact of efficiency difference on propagation intensity, determined through gradient search on the validation set. After obtaining the temporal embeddings of each node, the trend of node centrality change within a continuous time window is analyzed. The node degree centrality change rate is defined as: Among them, C x :node v x The rate of change of centrality is used to measure the stability of connectivity relationships; The weighted degree of a node at time t. Simultaneously, node importance is evaluated through simulated node failure experiments (i.e., removing a node and its connecting edges from the network and calculating the change in the remaining network connectivity). The impact of node failure is defined as follows: Among them, L x :node v x Impact degree; L0: average shortest path length of the original network; Lx Remove node v x The average shortest path length of the network is then determined. A node is identified as a critical vulnerable node when it simultaneously meets the criteria of "significantly decreasing collaborative efficiency coefficient," "high centrality change rate," and "significant failure impact." This identification result reflects high-risk entities that are both centrally located in the supply chain and exhibit anomalies in cash flow coordination, providing a basis for subsequent risk prediction and visualization.

[0036] Specifically, the process of cross-dimensional feature fusion of key vulnerable node identification and collaborative efficiency coefficient, and evaluation of the contribution weights of different risk factors through a dynamic attention mechanism to predict the comprehensive performance capability of user enterprises is as follows: A key vulnerable node feature vector is constructed, including the node's centrality index in the network, vulnerability duration, and impact range characteristics. The cash flow-supply chain collaborative efficiency coefficient is combined with the key vulnerable node features in multiple dimensions to form a comprehensive feature matrix. The importance weights of different dimensions in the feature matrix are calculated through a multi-head attention mechanism, where the query vector is dynamically adjusted based on historical performance. A feature cross-layer is used to learn the deep interaction relationship between key vulnerable node features and collaborative efficiency coefficient, generating a cross-feature representation. The comprehensive feature matrix is ​​weighted and fused based on attention weights to highlight feature dimensions that contribute highly to the prediction of performance capability. The weighted feature representation is input into a fully connected neural network, and the Softmax function outputs the probability distribution of the user enterprise's comprehensive performance capability in a specific future period.

[0037] In this implementation scheme, the following features are encoded as numerical vectors and then concatenated into a matrix: node centrality vector, vulnerability duration vector, impact range index vector, collaborative efficiency time series summary (e.g., statistics from recent windows), and supplementary historical performance indicators, mainly including standardization, missing value handling, and alignment. Attention calculation: The query vector is defined as... The key matrix is The value matrix is The single-head attention weight matrix is ​​then expressed as: in, Attention weight row vector; Θ: Query vector dynamically generated based on historical performance; Ke: Key matrix obtained by key mapping from feature matrix G; N: Value matrix obtained by value mapping from feature matrix G (used for subsequent weighted fusion); d θ : Query vector dimension, used to scale attention scores for stable training. Parameter explanation (semicolon-separated): Θ; represents the current model focus, output from a small feedforward network via the historical performance error sequence; K; represents the key representation of each feature dimension, obtained from G through a linear mapping; N; represents the value representation of each feature dimension, obtained from G through a linear mapping; d θThe dimension of the query vector is used for numerical stabilization. For multi-head attention, there are H parallel heads. Each head calculates its attention according to the above formula, and the outputs are concatenated. Then, a linear mapping is performed to obtain the final attention fusion representation. The fusion representation is denoted as Z and is calculated as follows: in, The attention weight of the h-th head; N h The value matrix corresponding to the h-th head; W o The linear projection matrix output by the head. Parameter explanation: The attention weight of the h-th head; N h The value matrix corresponding to the h-th head; W o This is used to map the multi-head stitching result back to the target dimension's projection matrix. Feature crossing and higher-order interactions: To learn the higher-order interactions between collaborative efficiency and vulnerable node features, a cross-feature matrix C can be generated using cross-operations: C = G⊙(GW) c ); where C: cross feature matrix; ⊙: element-wise product (Hadamard); W c : Cross-projection matrix, used to learn linear combinations to construct higher-order interactions. Parameter explanation: G: Input comprehensive feature matrix; W c : Trainable projection matrix of the cross layer. Weighted fusion and prediction head: The attention fusion representation Z and the cross feature C are concatenated and input into several fully connected layers to obtain the score vector s. Softmax is applied to the score vector to obtain the performance level probability distribution Π. Parameter explanation: Π a The predicted probability that the target company belongs to the a-th performance level (e.g., high / medium / low); s k : The original score output by the neural network for the a-th category. Attention dynamic adjustment strategy (weight / threshold determination method): Adaptive calibration based on historical prediction error feedback is adopted: When the mean of the short-term prediction error sequence exceeds the threshold ζ, a decay factor is applied to the key mapping learning rate of the attention or to the attention weights of low-contribution dimensions, thereby reducing the impact of overfitting noise features. The threshold ζ is determined by the error cutoff point on the validation set that achieves the optimal combination of prediction stability and immediacy (this can be determined using the inflection point method or ROC-based threshold selection method). Parameter description: ζ: Error trigger threshold, determined by parameter tuning on the validation set.

[0038] Specifically, the process of generating an interactive dynamic information graph that simultaneously includes fund flow paths, supply chain network status, and risk prediction results is as follows: A layered visualization architecture is constructed, comprising three main views: a basic network layer, a risk identification layer, and a prediction display layer. In the basic network layer, a force-directed layout algorithm is used to display the supply chain network topology, using node size to represent enterprise scale, node color depth to represent collaborative efficiency coefficients, and edge thickness to represent transaction intensity. In the risk identification layer, identified key vulnerable nodes are marked with highlighted borders and pulse animations, while their vulnerability scores and impact ranges are displayed. In the prediction display layer, a circular dashboard displays a comprehensive performance capability score, along with a trend curve showing historical score changes, and color coding is used to distinguish risk levels. A multi-layered interactive mechanism is designed so that when a user clicks on any enterprise node, all its fund flow paths, related transaction parties, and risk transmission paths are simultaneously highlighted.

[0039] This implementation plan constructs a three-layer view: a basic network layer (displaying topology and element mapping), a risk identification layer (highlighting key vulnerable nodes, displaying vulnerability scores and impact range), and a prediction display layer (displaying compliance probability, trend curves, and risk level color codes). Data is linked between layers: selecting any node triggers a linkage event, synchronously highlighting the corresponding information in each layer and opening a detailed panel. Vulnerability score calculation: to display the node risk intensity in the visual layers, the node vulnerability index Ω is defined as a weighted linear combination: Ω p =γ a ·U p +γ b ·V p +γ c ·W p ; where Ω p Vulnerability score of the p-th node; U p Node centrality volatility, used to reflect the volatility of connectivity relationships; V p The decline in collaborative efficiency reflects the degree of deterioration in the capital-business coupling; W p γ: The relative impact of node failure on network connectivity (simulation results); a ,γ b ,γ c : Weighting coefficient for the scoring combination. Parameter explanation: U p ; Nodal centrality volatility; V p The extent of the decline in collaborative efficiency; W p The degree of impact of node failure on network connectivity; γ a Central component weights; γ b ; Cooperative efficiency component weights; γ cConnectivity component weights. Weight determination method: Aiming at the accuracy of risk identification on historical samples (or historical defaults / event backtracking), the optimal γ combination is searched on the validation set by minimizing the sum of squared residuals or through Bayesian optimization. Visual mapping rules: Basic network layer: Force-oriented layout displays the topology; node size maps to enterprise size; node color depth maps to collaborative efficiency; edge thickness maps to transaction strength. Risk identification layer: For Ω... p Nodes exceeding a set visual threshold are highlighted and animated with pulses, and the affected area (a list of affected nodes) is displayed in a floating panel. The visual threshold is determined by visualization design and user experience testing. Predictive display layer: Displays the fulfillment probability of an entity using a circular dashboard or scale bar, along with a time-series curve to replay historical scores. When a user clicks on any node in the network graph, the system simultaneously displays: the node's fund flow path (source / destination link), related trading parties directly involved and their current vulnerability scores, and the risk transmission path (based on the affected subgraph obtained from simulation), supporting a time slider to replay historical evolution.

[0040] Please see Figure 2 A system for generating user financial information graphs through a financing platform includes the following modules: a data parsing and feature extraction module, used to perform multi-dimensional analysis of user bank statement data based on transaction semantic recognition, extract transaction scenario features from statement summaries using natural language processing technology, identify the capital flow path related to core enterprises in the supply chain, and extract the periodic fluctuation characteristics of operating cash flow based on time series analysis; and a collaborative efficiency analysis module, used to analyze the matching relationship between fund collection and payment time and order delivery cycle based on capital flow path and periodic fluctuation characteristics, combined with order payment period information in supply chain transaction data, construct a cash flow-supply chain collaborative efficiency coefficient, and quantify the cash cycle of enterprise operations. The system includes several modules: a supply chain network analysis module, a risk prediction and visualization module, and a risk prediction and visualization module. The supply chain network analysis module uses the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network to construct a dynamically weighted supply chain network topology. It learns the embedded representation of network nodes through a graph neural network and identifies key vulnerable nodes in the supply chain network by analyzing the changes in the topology of the weighted network within a continuous time window. The risk prediction and visualization module integrates the identification of key vulnerable nodes with the collaboration efficiency coefficient across dimensions. It evaluates the contribution weight of different risk factors through a dynamic attention mechanism, predicts the comprehensive performance capability of user enterprises, and generates an interactive dynamic information graph that simultaneously includes the cash flow path, supply chain network status, and risk prediction results.

[0041] In this implementation plan, the data parsing and feature extraction module is used to achieve the underlying data understanding and structured processing of the system. This module performs semantic parsing on bank transaction data to identify the purpose of funds and business background information contained in the transaction summary, thereby reconstructing the actual fund flow relationship of the enterprise within the supply chain system. Furthermore, by combining time series analysis methods, it extracts the periodic fluctuation characteristics of operating cash flow, providing a foundation for subsequent analysis of matching funds with business rhythms. The role of this module is to transform raw financial data into structured fund behavior characteristics, forming a quantifiable and traceable foundational layer of fund flow information. The collaborative efficiency analysis module undertakes the task of time-series collaborative modeling between fund behavior and business cycles. This module uses the fund flow path as the main line, aligning the cash flow fluctuation characteristics with the payment terms and delivery cycles in the supply chain transaction data. By matching the fund receipt and payment time with the business operation rhythm, it calculates the cash flow-supply chain collaborative efficiency coefficient. This coefficient is used to quantify the efficiency of fund use and the coordination of liquidity in the enterprise's operating cycle, providing a unified and comparable operational efficiency indicator for each node enterprise in the supply chain network. The core function of this module is to realize the dynamic correlation and quantitative description between fund behavior and business logic. The supply chain network analysis module identifies systemic risk nodes at the supply chain level. This module embeds collaborative efficiency coefficients into the supply chain network structure as node weights, constructs a dynamically weighted topology, and uses graph neural network algorithms to learn and analyze the temporal evolution of node embeddings, thereby identifying key vulnerable nodes that exhibit prominent characteristics when collaborative efficiency declines or structural stability weakens. This module extends the financial behavior characteristics of individual enterprises to the network level, revealing the propagation impact of funds and transaction relationships at the group level and potential risk transmission paths. The risk prediction and visualization module, based on the risk identification results, enables comprehensive performance capability prediction and information visualization. This module integrates key vulnerable node characteristics with collaborative efficiency characteristics across dimensions, assigns adaptive weights to different risk factors through a dynamic attention mechanism, and generates predictions of the user enterprise's performance capability for a specific future period. Simultaneously, the prediction results, fund flow paths, and supply chain network status are interactively displayed on the same interface, enabling the integrated perception and visual exploration of multi-dimensional risk information. This module transforms complex financial behavior models into interpretable decision-making criteria, helping financing platforms achieve integrated applications of risk monitoring, enterprise credit assessment, and visualized decision support.

[0042] In summary, this application has at least the following effects:

[0043] A method and system for generating financial information graphs for users through a financing platform are proposed. By semantically parsing user bank transaction data and integrating supply chain transaction data, a deep correlation modeling of cash flow and supply chain structure is achieved. By constructing a cash flow-supply chain collaborative efficiency coefficient and introducing a graph neural network to analyze the dynamic changes in the supply chain network topology, intelligent identification of enterprise capital turnover efficiency and network vulnerability nodes is realized. Furthermore, by combining a dynamic attention mechanism to comprehensively evaluate multi-source risk factors, a forward-looking prediction of enterprise performance capabilities can be made, and the results are visualized in the form of an interactive dynamic information graph, thereby improving the financing platform's quantitative analysis capabilities and risk identification accuracy of user operating conditions.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process.Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0049] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for generating a user's financial information graph through a financing platform, characterized in that, Includes the following steps: S1. Perform multi-dimensional analysis on user bank statement data based on transaction semantic recognition, extract transaction scenario features from statement summaries through natural language processing technology, identify the capital flow path related to core enterprises in the supply chain, and extract the periodic fluctuation characteristics of operating cash flow based on time series analysis; S2. Based on the characteristics of capital flow paths and cyclical fluctuations, and combined with order payment period information in supply chain transaction data, analyze the matching relationship between capital receipt and payment time and order delivery cycle, construct a cash flow-supply chain collaboration efficiency coefficient, and quantify the capital turnover efficiency of the enterprise's operating cycle; S3. Using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network, a dynamically weighted supply chain network topology is constructed. The embedded representation of network nodes is learned through graph neural networks. By analyzing the changes in the topology of the weighted network within a continuous time window, key vulnerable nodes in the supply chain network are identified. S4. Cross-dimensional feature fusion of key vulnerable node identification and collaborative efficiency coefficient, evaluation of the contribution weight of different risk factors through dynamic attention mechanism, prediction of the comprehensive performance capability of user enterprises, and generation of an interactive dynamic information graph that simultaneously includes capital flow path, supply chain network status and risk prediction results.

2. The method for generating a user's financial information graph through a financing platform according to claim 1, characterized in that: The process of performing multi-dimensional analysis on user bank statement data based on transaction semantic recognition, extracting transaction scenario features from the statement summary using natural language processing technology, and identifying the specific process of fund transfer paths related to core enterprises in the supply chain is as follows: Key elements such as counterparty name, type of goods / services, and purpose of transaction are extracted from the transaction summary based on named entity recognition. The semantic similarity between transaction summaries and typical supply chain scenarios is calculated using a pre-trained domain word vector model, identifying transaction scenarios including procurement payments, sales receipts, and logistics settlements. By analyzing the patterns of fund flows among trading counterparties using association rules, a fund flow network graph with the core enterprise as the hub is constructed. Based on the connectivity analysis of the graph, key paths with direct or indirect fund transactions with the core enterprise are identified.

3. The method for generating a user's financial information graph through a financing platform according to claim 2, characterized in that: The specific process for extracting the cyclical fluctuation characteristics of operating cash flow based on time series analysis is as follows: The operating cash flow series is decomposed into multiple scales to obtain periodic terms representing different time scales. The significant period lengths corresponding to the business cycle in the cash flow series are identified by autocorrelation function analysis. Based on the periodic term and the identified period length, the amplitude characteristics of cash flow fluctuations are calculated, including the range and standard deviation of fluctuations within the period. Based on the temporal distribution of the periodic term, phase features are extracted, including the start time and peak position of the periodic fluctuation; Based on the stability analysis of continuous periodic terms, a periodic consistency index is calculated to characterize the sustainable characteristics of a company's business cycle.

4. The method for generating a user's financial information graph through a financing platform according to claim 1, characterized in that: Step S2 includes the following steps: Based on the characteristics of fund flow paths and cyclical fluctuations, and combined with order payment period information in supply chain transaction data, a mapping relationship between fund receipt and payment time and order delivery cycle is established. Calculate the dynamic matching degree between the capital turnover cycle and the business operation cycle, quantify the collaborative efficiency in the time dimension, and evaluate the stability characteristics of collaborative efficiency through sliding window analysis at multiple time scales. By integrating cycle consistency indicators and time-series matching characteristics, a multi-dimensional cash flow-supply chain collaboration efficiency coefficient is constructed.

5. The method for generating a user's financial information graph through a financing platform according to claim 1, characterized in that: The specific process of constructing a dynamically weighted supply chain network topology by using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network is as follows: An initial network structure is constructed based on the upstream and downstream transaction relationships of the supply chain, where nodes represent enterprise entities and edges represent transaction relationships. The cash flow-supply chain collaboration efficiency coefficient of each enterprise entity is used as the node weight to reflect the enterprise's capital turnover efficiency in the supply chain. The weight of the edges is dynamically adjusted based on the transaction amount and the frequency of the transaction, with higher edge weights corresponding to high-frequency, large-value transactions. A network adjacency matrix containing node weights and edge weights is constructed, where diagonal elements store node weights and off-diagonal elements store edge weights. By introducing a time dimension, a dynamic network topology sequence under a continuous time window is constructed to capture the evolution of the network structure.

6. The method for generating a user's financial information graph through a financing platform according to claim 5, characterized in that: The specific process of learning the embedded representations of network nodes through graph neural networks and identifying key vulnerable nodes in the supply chain network by analyzing the topological changes of the weighted network within continuous time windows is as follows: Based on the dynamically weighted supply chain network topology, the network structure features of each time slice are extracted, the degree centrality change trend of nodes in continuous time windows is calculated, and nodes with unstable connection relationships are identified. The propagation influence of nodes is analyzed, and their importance is assessed by simulating the impact of node failure on network connectivity. By combining the time series changes of node weights, nodes with continuously declining collaborative efficiency coefficients and critical network positions are identified. By considering changes in the network topology and the evolution of node weights, nodes with high network importance and low collaborative efficiency are identified as critical and vulnerable nodes.

7. The method for generating a user's financial information graph through a financing platform according to claim 1, characterized in that: The specific process of fusing key vulnerability node identification with collaborative efficiency coefficients across dimensions, and using a dynamic attention mechanism to evaluate the contribution weight of different risk factors to predict the comprehensive performance capability of user enterprises is as follows: Construct a feature vector for key vulnerable nodes, including the node’s centrality index in the network, vulnerability duration, and impact range characteristics. Combine the cash flow-supply chain coordination efficiency coefficient with the key vulnerable node characteristics in multiple dimensions to form a comprehensive feature matrix. The importance weights of different dimensions in the feature matrix are calculated using a multi-head attention mechanism, where the query vector is dynamically adjusted based on historical performance. By learning the deep interaction relationship between the features of key vulnerable nodes and the collaborative efficiency coefficient through the feature cross layer, a cross feature representation is generated; The comprehensive feature matrix is ​​weighted and fused based on attention weights to highlight the feature dimensions that contribute highly to the prediction of performance capability. The weighted feature representation is input into a fully connected neural network, and the Softmax function outputs the probability distribution of the user company's comprehensive performance capability in a specific future period.

8. The method for generating a user's financial information graph through a financing platform according to claim 7, characterized in that: The specific process for generating an interactive, dynamic infographic that simultaneously includes fund flow paths, supply chain network status, and risk prediction results is as follows: Construct a layered visualization architecture, including three main views: the basic network layer, the risk identification layer, and the prediction display layer; In the basic network layer, the supply chain network topology is displayed through a force-directed layout algorithm, with node size representing enterprise scale, node color depth representing collaboration efficiency coefficient, and edge thickness representing transaction intensity. In the risk identification layer, key vulnerable nodes are marked with highlighted borders and pulse animations, while their vulnerability scores and impact ranges are displayed. In the forecasting display layer, a circular dashboard displays the comprehensive performance capability score, along with a trend curve showing historical score changes, and color coding is used to distinguish risk levels. The design incorporates a multi-layered interactive mechanism. When a user clicks on any enterprise node, all its fund transfer paths, related transaction parties, and risk transmission paths are simultaneously highlighted.

9. A system for generating a user's financial information map through a financing platform, applied to the method for generating a user's financial information map through a financing platform as described in any one of claims 1-9, characterized in that, Includes the following modules: The data parsing and feature extraction module is used to perform multi-dimensional analysis of user bank transaction data based on transaction semantic recognition. It extracts transaction scenario features from the transaction summary through natural language processing technology, identifies the capital flow path related to the core enterprise in the supply chain, and extracts the periodic fluctuation characteristics of operating cash flow based on time series analysis. The collaborative efficiency analysis module is used to analyze the matching relationship between the time of payment and the order delivery cycle based on the characteristics of the capital flow path and cyclical fluctuations, combined with the order payment period information in the supply chain transaction data, to construct the cash flow-supply chain collaborative efficiency coefficient and quantify the capital turnover efficiency of the enterprise's operating cycle. The supply chain network analysis module is used to construct a dynamically weighted supply chain network topology by using the cash flow-supply chain collaboration efficiency coefficient as the weight of each enterprise entity node in the supply chain network. It learns the embedded representation of network nodes through graph neural networks and identifies key vulnerable nodes in the supply chain network by analyzing the changes in the topology of the weighted network within a continuous time window. The risk prediction and visualization module is used to integrate key vulnerable node identification and collaborative efficiency coefficient across dimensions, evaluate the contribution weight of different risk factors through a dynamic attention mechanism, predict the comprehensive performance capability of user enterprises, and generate an interactive dynamic information map that simultaneously includes the capital flow path, supply chain network status, and risk prediction results.

Images