A method, device and medium for managing accounts receivable based on a multi-layer perceptron

Abstract

The embodiment of the specification discloses a kind of based on multilayer perceptron accounts receivable management method, equipment and medium, it is related to prediction technical field, method includes: obtaining the multidimensional account data corresponding to accounts receivable, utilize the dimension correlation thermal data of pre-construction, multidimensional account data is preprocessed, determine multidimensional account characteristic vector;According to multidimensional account characteristic vector and pre-trained accounts receivable prediction model, determine the accounts receivable prediction data corresponding to accounts receivable, to determine the key risk factor set corresponding to each accounts receivable based on accounts receivable prediction data;Through key risk factor set, accounts receivable prediction data is causally chained reasoning, determine the risk causal chain data corresponding to accounts receivable;Based on risk causal chain data and accounts receivable prediction data, accounts receivable is visualized risk display, realize the risk management of accounts receivable, break through the limitation of only output prediction result of relying on single data and model.

Description

Technical Field

[0001] This specification relates to the field of forecasting technology, and in particular to an accounts receivable management method, device and medium based on a multilayer sensor. Background Technology

[0002] Accounts receivable risk management is a crucial aspect of a company's cash flow and financial soundness. Accounts receivable management is essential for financial health and cash flow forecasting. Traditional accounts receivable forecasting methods often rely on single historical accounts receivable data. However, accounts receivable involve various uncertainties and risk factors, such as market factors and customer dimensions. Traditional forecasting methods often struggle to fully mine and integrate multi-dimensional data, resulting in low forecast accuracy.

[0003] Furthermore, existing models are mostly "black box" structures, only outputting predicted results such as risk probabilities. Enterprise management users cannot pinpoint core risk drivers, such as the "contribution weight of customer payment delay rate to overdue payments," resulting in insufficiently targeted risk response measures. Currently, most risk scores are displayed through tables or two-dimensional charts, lacking visual and dynamic tracking of the causes of risks. This makes it difficult to intuitively identify the core risk causes and their transmission paths, leading to low efficiency in risk management.

[0004] Therefore, traditional accounts receivable forecasting methods often rely on single historical accounts receivable data, ignoring the influence of external factors, resulting in low forecast accuracy. Furthermore, they only output forecast results and lack the ability to visualize and dynamically track the causes of risks, which affects the efficiency of subsequent risk management. Summary of the Invention

[0005] This specification provides one or more embodiments of an accounts receivable management method, device, and medium based on a multilayer sensor, which addresses the following technical problem: Traditional accounts receivable forecasting methods often rely on single historical accounts receivable data, ignoring the influence of external factors, resulting in low forecast accuracy. Furthermore, they only output forecast results and lack visual dynamic tracking of the causes of risks, affecting the efficiency of subsequent risk management.

[0006] One or more embodiments of this specification employ the following technical solutions:

[0007] This specification provides one or more embodiments of an accounts receivable management method based on a multilayer perceptron. The method includes: acquiring multi-dimensional accounts receivable data; preprocessing the multi-dimensional accounts receivable data using pre-constructed dimensional correlation heatmap data to determine multi-dimensional accounts receivable feature vectors, wherein the multi-dimensional accounts receivable data includes accounts receivable customer dimension data, accounts receivable product dimension data, accounts receivable market dimension data, and internal enterprise management data; determining accounts receivable prediction data corresponding to the accounts receivable based on the multi-dimensional accounts receivable feature vectors and a pre-trained accounts receivable prediction model, and determining a set of key risk factors for each accounts receivable based on the accounts receivable prediction data; performing causal chain reasoning on the accounts receivable prediction data using the set of key risk factors to determine the risk causal chain data corresponding to the accounts receivable; and visually displaying the risks of the accounts receivable based on the risk causal chain data and the accounts receivable prediction data to achieve risk management of the accounts receivable.

[0008] This specification provides one or more embodiments of an accounts receivable management device based on a multilayer sensor, comprising:

[0009] At least one processor; and,

[0010] A memory communicatively connected to the at least one processor; wherein,

[0011] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the above-described method.

[0012] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0013] The above-mentioned at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: Through the embodiments of this specification, multi-dimensional accounts receivable data, including customer dimensions, product dimensions, market dimensions, and internal enterprise management data, are comprehensively collected. This changes the limitation of traditional methods that rely solely on single historical accounts receivable data. By using pre-constructed dimensional correlation heatmap data to preprocess multi-dimensional data, potential relationships between data are mined, and more representative multi-dimensional accounts receivable feature vectors are determined. Subsequent prediction models provide rich and high-quality data, enabling the model to better capture various factors affecting accounts receivable, thereby significantly improving the accuracy of accounts receivable prediction. Based on multi-dimensional accounts receivable feature vectors, accounts receivable prediction data is determined with the help of a pre-trained accounts receivable prediction model, and the set of key risk factors is further identified. This breaks through the limitation of existing "black box" models that only output prediction results, and can clearly identify the core risk drivers affecting accounts receivable. By analyzing a set of key risk factors, the contribution weight of factors such as customer payment delay rates to overdue payments can be accurately determined. Causal chain reasoning using the set of key risk factors on accounts receivable forecast data yields risk causal chain data. This clearly demonstrates the causes and transmission paths of risks, moving beyond an isolated view of risk factors and forecast results. It provides an intuitive understanding of how different risk factors interact and ultimately affect accounts receivable. This clear presentation of risk transmission paths helps to comprehensively grasp the risk formation mechanism and prevent and block the spread of risks in advance. Combining risk causal chain data and accounts receivable forecast data, a visualized risk display of accounts receivable is provided. Unlike traditional methods that only display risk scores through tables or two-dimensional charts, this method dynamically tracks the causes of risks, more intuitively identifies core risk causes and their transmission paths, quickly locates high-risk areas and key risk links, and improves the efficiency of risk management. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0015] Figure 1 A flowchart illustrating an accounts receivable management method based on a multilayer sensor, provided as an embodiment of this specification;

[0016] Figure 2 An example diagram of dimensional correlation thermodynamic data provided in the embodiments of this specification;

[0017] Figure 3This is a schematic diagram of the structure of an accounts receivable management device based on a multilayer sensor, provided as an embodiment of this specification. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] This specification provides an accounts receivable management method based on a multilayer sensor. It should be noted that the execution entity in this specification can be a server or any device with data processing capabilities. Figure 1 A flowchart illustrating an accounts receivable management method based on a multilayer sensor, as provided in this specification, is shown below. Figure 1 As shown, the main steps include the following:

[0020] Step S101: Obtain multi-dimensional accounts receivable data, and preprocess the multi-dimensional accounts receivable data using pre-constructed dimensional correlation heatmap data to determine the multi-dimensional accounts receivable feature vector.

[0021] In one embodiment of this specification, accounts receivable information within the real-time monitoring system is used to obtain multi-dimensional accounts receivable data. This multi-dimensional data includes customer-level data, product-level data, market-level data, and internal management data. It should be noted that the multi-dimensional accounts receivable data here refers to data relevant to accounts receivable. Customer-level data includes customer credit ratings, payment records, and industry trends; product-level data includes sales volume, product type, and sales cycle; market-level data includes macroeconomic indicators, industry prosperity index, and competitive landscape; and internal management data includes changes in sales policies, adjustments to credit policies, and collection efforts.

[0022] Using pre-constructed dimensional correlation heatmap data, the multi-dimensional accounts receivable data is preprocessed to determine the multi-dimensional accounts receivable feature vector. Specifically, this includes: identifying the historical multi-dimensional accounts receivable time series data; extracting features from the historical multi-dimensional accounts receivable time series data for each accounts receivable dimension to determine the historical dimensional feature data corresponding to each dimension, where the accounts receivable dimensions include customer, product, market, and internal management dimensions; performing feature correlation analysis on the historical dimensional feature data corresponding to each accounts receivable dimension to determine the feature correlation coefficient between the accounts receivable dimensions, and constructing dimensional correlation heatmap data based on this coefficient; identifying the missing dimensions in the multi-dimensional accounts receivable data; filling in the missing dimensions using the dimensional correlation heatmap data; determining the current multi-dimensional accounts receivable data; and extracting features from the current multi-dimensional accounts receivable data to determine the multi-dimensional accounts receivable feature vector.

[0023] In one embodiment of this specification, after collecting multi-dimensional accounts receivable data, preprocessing operations such as cleaning, missing value handling, outlier detection and correction, and time series alignment are required on the raw data. Historical multi-dimensional accounts receivable time series data constructed during model training is then used to analyze the historical multi-dimensional accounts receivable time series data to obtain the feature correlations between different accounts receivable dimensions. It should be noted that the accounts receivable dimensions here include customer, product, market, and internal management dimensions. Customer dimensions include secondary dimensions such as customer credit rating, payment records, and industry dynamics; product dimensions include secondary dimensions such as sales volume, product type, and sales cycle; market dimensions include secondary dimensions such as macroeconomic indicators, industry prosperity index, and competitive landscape; and internal management data includes secondary dimensions such as changes in sales policies, adjustments to credit policies, and collection efforts. These dimensions are only used as examples. Pearson correlation coefficients are used to analyze the correlation between features of each dimension, determining the feature correlation coefficients between accounts receivable dimensions. Based on these feature correlation coefficients, dimension correlation heatmaps are constructed. Figure 2 An example diagram of dimensional correlation thermodynamic data provided in the embodiments of this specification, such as... Figure 2 As shown, the correlation coefficient between a company's credit rating and its historical payment records is -0.25.

[0024] The missing dimensions in the multi-dimensional accounts receivable data are identified. Data is then used to fill in the missing dimensions using heatmap data related to those dimensions, thus determining the current multi-dimensional accounts receivable data. If a dimension is missing (e.g., industry trends), the missing value is estimated using a linear regression model based on the contemporaneous data of the dimension with the highest correlation in the heatmap (e.g., unit credit rating), and then filled in. Finally, features are extracted from the current multi-dimensional accounts receivable data to determine the multi-dimensional accounts receivable feature vector.

[0025] The above technical solutions comprehensively collect multi-dimensional accounts receivable data, covering aspects such as customers, products, markets, and internal management, ensuring the richness and comprehensiveness of the data and providing sufficient information for accurate analysis of accounts receivable. Using dimensional correlation heatmaps to fill missing dimensions, compared to simple data filling methods (such as mean or median filling), better preserves the inherent relationships between data, reduces information loss due to missing data, and improves data completeness and accuracy. Constructing dimensional correlation heatmaps by extracting features and performing correlation analysis on historical multi-dimensional accounts receivable time series data helps to uncover potential relationships between different dimensions of data, providing more valuable information for subsequent prediction models. This enables the models to more accurately capture the influencing factors of accounts receivable, thereby improving prediction accuracy and reducing prediction errors.

[0026] Step S102: Based on the multi-dimensional accounts receivable feature vector and the pre-trained accounts receivable prediction model, determine the accounts receivable prediction data corresponding to each accounts receivable, and determine the set of key risk factors corresponding to each accounts receivable based on the accounts receivable prediction data.

[0027] Before determining the accounts receivable prediction data corresponding to the accounts receivable based on the multi-dimensional accounts receivable feature vector and the pre-trained accounts receivable prediction model, the method further includes: collecting historical accounts receivable data and historical multi-dimensional accounts receivable time series data corresponding to the historical accounts receivable data, wherein the historical accounts receivable data includes accounts receivable amount data, aging distribution data, and collection cycle data; performing anomaly processing and missing value imputation on the historical accounts receivable data and the historical multi-dimensional accounts receivable time series data to determine the accounts receivable dataset; using the accounts receivable dataset, training a pre-constructed initial multilayer perceptron model through a preset optimization strategy to determine an accounts receivable prediction model that meets preset requirements, wherein the number of hidden layers in the initial multilayer perceptron model is determined based on a one-time neural architecture search combined with weight sharing.

[0028] In one embodiment of this specification, historical accounts receivable data from recent years is extracted from a financial database, including amount, aging, and collection cycle. This data is then correlated with concurrent multidimensional data (such as customer industry dynamics and records of changes in corporate sales policies) to collect historical accounts receivable data, including but not limited to core indicators such as accounts receivable amount, aging distribution, and collection cycle for each period. Historical multidimensional time-series data affecting accounts receivable is obtained, including but not limited to customer dimensions such as customer credit rating, historical payment records, and industry dynamics; product dimensions such as sales volume, product type, and sales cycle; market dimensions such as macroeconomic indicators, industry prosperity index, and competitive landscape; and internal management dimensions such as changes in sales policies, adjustments to credit policies, and collection efforts. The collected raw data undergoes preprocessing operations such as cleaning, missing value handling, outlier detection and correction, and time-series alignment. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) can identify regions with low density in the dataset; points in these regions may be outliers. For handling missing and outlier values, if the total data volume is large, outliers can be directly deleted. If the total data volume is small, the exponential smoothing method of time series can be used to estimate replacement values ​​for outliers.

[0029] Construct a multi-layer, fully connected neural network model, including an input layer, hidden layers, and an output layer. The input layer receives processed multi-dimensional features, the hidden layers process information at multiple levels through non-linear activation functions, and the output layer generates a prediction of future accounts receivable. MLP (Multilayer Perceptron) is a type of feedforward artificial neural network. It contains multiple hidden layers and one output layer, each layer consisting of multiple neurons, with neurons in adjacent layers fully connected. The basic working principle of MLP is that after receiving data at the input layer, information is propagated and processed layer by layer through non-linear transformations, finally generating a prediction result at the output layer.

[0030] The input layer receives raw feature data and uses it as input to the neurons in the first layer. Neurons in each hidden layer apply a non-linear activation function (such as Sigmoid, ReLU, Tanh, etc.) to the output of the previous layer, multiply it by the weight matrix of that layer, and add a bias term to produce the output of that layer. Setting appropriate hidden layers is crucial for model success. This specification uses One-Shot Neural Architecture Search (One-Shot NAS) combined with weight sharing techniques to determine the number of hidden layers in a multilayer perceptron (MLP). In traditional NAS, each candidate network architecture needs to be trained independently to evaluate its performance, which consumes significant computational resources and time. One-Shot NAS, however, constructs a supernet that contains all possible sub-network structures, and all sub-networks share weights, thus requiring only this single supernet to be trained. This approach significantly reduces repetitive training during the search process. Training the supernetwork allows for simultaneous updates of architecture parameters and weights. While this may introduce coupling issues between weights and architecture, proper design and optimization can accelerate the convergence of the entire search process. Weight sharing means that different network structures can learn from each other during the search, making training more efficient. Even if the final selected architecture is "pruned" from the supernetwork, it can inherit pre-trained weights and have a relatively high starting point, allowing for faster fine-tuning and better performance. Furthermore, the One-Shot NAS method does not require a separate training process for each potential architecture, making it easier to handle large-scale and complex search spaces. It can flexibly explore different levels of network components (e.g., kernel size, number of channels, layer type), and in many cases, the optimal architecture obtained can be directly extracted from the supernetwork and trained independently to achieve optimal performance. The core advantage of One-Shot NAS combined with weight sharing lies in its ability to effectively utilize limited computing resources to find high-performance neural network architectures in a short time. It is well-suited for applications in accounts receivable prediction scenarios.

[0031] The output of the last hidden layer enters the output layer, where the activation function is the Softmax function. The reason for using the Softmax function is that it alleviates the vanishing gradient problem, making it suitable for outputs based on multi-class classification with multiple neurons. The Softmax function takes an arbitrary real vector and transforms it into a probability distribution. For a given input vector Z = (z1, z2, ..., z_K), the Softmax function calculates the exponential function e^(zi) for each element and then divides it by the sum of the exponential functions of all elements, ensuring that the sum of all elements in the output vector is 1, satisfying the property of the probability distribution. The Softmax function is suitable for multi-class classification problems; the predicted score for each class is transformed into the probability of the corresponding class, with the highest probability corresponding to the most likely class. Due to the probability distribution characteristics of the output, the Softmax function guarantees that the output classes are mutually exclusive; that is, for a sample, it can only belong to one class, namely the class with the highest probability.

[0032] The output can be directly interpreted as the probability of each category, which helps in understanding and decision-making. Combined with the cross-entropy loss function, the Softmax function makes model optimization a clear task of minimizing the probability of misclassification. The Softmax function is smooth and continuous, which is beneficial for the convergence of optimization algorithms such as gradient descent. The Softmax function can provide relative comparisons between categories; the output value not only represents the probability of its own category but also reflects the relative strength of different categories relative to other categories. The derivative of the Softmax function is easy to calculate, which is beneficial for the backpropagation process of neural networks. Compared to simple linear output layers and the sigmoid function, it solves the problems of vanishing and exploding gradients in multi-class classification problems.

[0033] During model training, the AdamW optimizer with weight decay correction is used to adjust weights and biases to minimize the loss function between the predicted output and the actual label. AdamW modifies the weight decay of the original Adam algorithm to better align with L2 regularization. By modifying the implementation of weight decay, AdamW improves Adam's regularization handling while maintaining its original high-efficiency optimization characteristics. The Adabelief optimizer and a learning-slight decay strategy are used to update weights and biases, resulting in a gradual improvement in model performance on the training set. Adabelief overcomes the overconfidence in gradient assumptions by existing adaptive optimizers like Adam in certain situations. The Adabelief optimizer converges quickly and exhibits good stability, achieving performance comparable to or even better than Adam and other advanced optimizers in various environments. Because Adabelief reduces overly optimistic reliance on gradient estimates, it helps improve the model's generalization ability, especially in adversarial training and noise-sensitive environments. The Adabelief optimizer better adapts to gradient signals, improving training efficiency and model quality.

[0034] The obtained feature vectors are used as model input, and training is performed using the labels corresponding to historical accounts receivable data. Leave-One-Out Cross-Validation (LOOCV) is used to optimize model parameters, ensuring good generalization performance. LOOCV is more accurate than conventional k-fold cross-validation because only one sample is used for testing in each validation, with all other samples used for training, maximizing the use of limited data. It fully utilizes the data, ensuring that almost all data (N-1 observations) are used in each training iteration. When the sample size is relatively small and computational resources allow, LOOCV can provide accurate estimates, offering nearly N cross-validations. The historical accounts receivable dataset is not large, making this method very suitable. After training, the model's predictive performance on the validation set is evaluated, including mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). 2 Indicators such as these are used. Following the above method, an accounts receivable forecasting model is obtained.

[0035] In one embodiment of this specification, a trained accounts receivable prediction model is used, and multi-dimensional accounts receivable feature vectors are input into the model to predict accounts receivable for future time periods containing the latest multi-dimensional feature information, thereby determining the accounts receivable prediction data corresponding to the accounts receivable.

[0036] Based on the accounts receivable forecast data, a set of key risk factors corresponding to each accounts receivable is determined. Specifically, this includes: calculating the contribution weight of each account feature in the multi-dimensional account feature vector to the accounts receivable forecast data using a preset gradient approximation algorithm; and selecting at least one key risk feature from multiple account features based on the contribution weight of each account feature to the accounts receivable forecast data and a preset threshold, in order to construct the set of key risk factors.

[0037] The contribution weight of each account receivable feature in the multi-dimensional account receivable feature vector to the accounts receivable prediction data is calculated using a preset gradient approximation algorithm. Specifically, this includes: backpropagating the output of the accounts receivable prediction model to calculate the input feature gradient matrix corresponding to the multi-dimensional account receivable feature vector; and using the integral path method based on the input feature gradient matrix to calculate the contribution weight of each account receivable feature in the multi-dimensional account receivable feature vector to the accounts receivable prediction data.

[0038] In one embodiment of this specification, after model training is completed, backpropagation is performed on the model's output. During backpropagation, the input feature gradient matrix corresponding to the multi-dimensional accounts receivable feature vector is calculated according to the model's computational logic. The input feature gradient matrix reflects the gradient change of the model output corresponding to each input feature (i.e., accounts receivable feature), reflecting the degree of local influence of each accounts receivable feature on the prediction result. Based on the calculated input feature gradient matrix, the contribution weight of each accounts receivable feature to the predicted accounts receivable data is calculated using the integral path method. The integral path method selects an appropriate integral path in the feature space to integrate the gradient information, thereby obtaining a more comprehensive and accurate weight reflecting the overall contribution of accounts receivable features to the prediction result. In practice, the path from the point where all features take their mean to the target sample point can be selected for integration to comprehensively consider the impact of the feature's change from the initial state to the current state on the prediction result.

[0039] Set reasonable preset thresholds, taking into account factors such as the company's risk tolerance for accounts receivable, industry averages, and historical data analysis. Compare the contribution weight of each accounts receivable feature with the preset thresholds, and select those features whose contribution weight is greater than the preset thresholds. These selected features are the key risk features. Integrate all identified key risk features to construct a set of key risk factors. Each key risk feature in the set of key risk factors has a significant impact on accounts receivable forecast data and represents a risk factor that the company needs to focus on during accounts receivable management.

[0040] The above technical solution comprehensively covers customer, product, market, and internal management aspects by starting from multi-dimensional accounts receivable data. It utilizes gradient approximation algorithms and backpropagation to calculate contribution weights, comprehensively considering the impact of various accounts receivable characteristics on accounts receivable forecast data. This multi-dimensional analysis and precise calculation method avoids the limitations of single-dimensional analysis, accurately identifying key risk characteristics that significantly impact forecast data. This allows enterprises to clearly understand the core factors affecting accounts receivable, providing precise targets for subsequent risk prevention and control. Based on contribution weights and preset thresholds, key risk characteristics are screened, constructing a set of key risk factors. Based on this set, limited resources can be focused on key risk points to formulate targeted prevention and control strategies. For customers with significant credit rating impacts, credit monitoring is strengthened; for products significantly affected by sales cycle fluctuations, sales strategies are adjusted. Compared to traditional comprehensive but unfocused prevention and control methods, this significantly improves risk prevention and control efficiency, reduces management costs, and effectively reduces potential bad debt losses.

[0041] Step S103: Using the set of key risk factors, perform causal chain reasoning on the accounts receivable forecast data to determine the risk causal chain data corresponding to the accounts receivable.

[0042] The process involves collecting expert rule discourse texts and transforming them using natural language processing (NLP) technology to generate expert discourse rules. These rules include rule transmission data and preset rule strength coefficients. Historical accounts receivable data tables corresponding to historical risk cases are also collected. These tables include customer-level fields, product-level fields, market-level fields, internal management fields, and risk outcome fields. Based on these historical accounts receivable data tables, association rule mining is performed to obtain rule transmission data between specified feature combinations and risk outcomes. The regression relationship between feature changes and risk outcomes in the historical accounts receivable data tables is then used to determine the transmission strength coefficients, thereby identifying the association rules. Finally, a risk transmission rule library is constructed using these association rules and the expert discourse rules.

[0043] In one embodiment of this specification, expert rule texts were extensively collected through methods such as organizing expert seminars, interviewing industry veterans, and gathering relevant professional literature. These texts cover professional knowledge and experience in the field of accounts receivable risk, including but not limited to understanding the interaction of different risk factors, common risk transmission paths, and criteria for judging the severity of risks. For example, if a customer's industry prosperity index declines for three consecutive months and its accounts payable turnover rate is less than 20% of the industry average, then its probability of delinquency in the next six months increases by 40%. A pre-trained BERT model is used for dependency parsing to identify the conditional terms (such as "the industry prosperity index has declined for three consecutive months") and the result terms (such as "the probability of delinquency increases by 40%) in the rule text. Map the variables in the condition terms (e.g., "industry prosperity index") to preset dimension fields (e.g., the "industry prosperity index field" under the accounts receivable market dimension), and extract constraint parameters (e.g., "decline for 3 consecutive months" is converted into a time window parameter t_w = 3 and a trend determination function F (trend ≤ -0.1). Numerical extraction is performed on the probability variation values ​​in the result terms (e.g., "increase of 40%)" to generate a rule strength coefficient (denoted as R_s = 0.4), which is then converted into a preset rule strength base value (e.g., 0.4 × time decay factor) through regularization. The structured conditions and results are stored in triplet format.

[0044] {

[0045] "Preconditions":[

[0046] {"Dimension":"Market Dimension","Field":"Industry Prosperity Index","Constraint":"trend≤-0.1over t_w=3"},

[0047] {"Dimension":"Customer Dimension","Field":"Accounts Payable Turnover Ratio","Constraint":"value<Industry Average × 0.8"}

[0048] ],

[0049] Conclusion: "The probability of delinquency has increased."

[0050] "Strength coefficient": "R_s=0.4×exp(-0.1t)",

[0051] "Transmission Path": Market Risk → Customer liquidity risk → Overdue

[0052] }

[0053] In addition to the methods mentioned above, natural language processing (NLP) techniques can also be used to process the collected text. First, text cleaning is performed to remove noise, such as irrelevant punctuation, special characters, and redundant whitespace, improving readability and processing efficiency. Then, lexical analysis is used to break the text down into words or phrases, tagging parts of speech and identifying named entities, such as customer names, product types, and market metrics. Next, syntactic analysis and semantic understanding are used to parse sentence structure and understand the semantic logic of the text. Using pre-trained language models or custom rule templates, key information is extracted from the processed text to generate argumentative rule transmission data. This data describes the transmission relationships between risk factors, such as "If a customer's credit rating declines and market competition intensifies, the risk of overdue accounts receivable increases." Simultaneously, based on expert descriptions of the importance and impact of different rules, combined with industry experience and expert judgment, a preset rule strength coefficient is determined. This coefficient quantifies the relative importance of each rule in risk assessment; its value range can be set according to actual conditions, such as between 0 and 1, with higher values ​​indicating stronger rule impact.

[0054] Collect historical accounts receivable data tables corresponding to historical risk cases, ensuring data completeness and accuracy. The data tables include customer-level fields (such as customer credit rating, historical payment records, industry, etc.), product-level fields (sales volume, product type, sales cycle, etc.), market-level fields (macroeconomic indicators, industry prosperity index, market share, etc.), internal management fields (sales policy changes, credit policy adjustments, collection efforts, etc.), and risk outcome fields (whether overdue, overdue amount, bad debt status, etc.). Clean and preprocess the collected data, handling missing and outlier values, standardizing data format, and ensuring data quality meets mining requirements. Analyze the historical accounts receivable data tables using association rule mining algorithms, such as the Apriori algorithm or the FP-Growth algorithm. Set appropriate support (e.g., minimum support threshold of 0.05) and confidence thresholds. Support represents the frequency of a rule's occurrence in the dataset, and confidence represents the probability that a conclusion holds true under certain conditions. Use algorithms to mine the potential relationships between specified feature combinations and risk outcomes, obtaining data on the transmission of mining rules. For example, the mining results show: {Industry prosperity index is declining, accounts payable turnover rate <0.5}. → High risk (support = 0.08).

[0055] For the identified association rules, further analysis is conducted on the regression relationship between feature changes and risk outcomes in historical accounts receivable data tables to determine the transmission strength coefficient. Linear regression, logistic regression, or other suitable regression analysis methods can be used to quantify the impact of feature changes on risk outcomes. If a 10% increase in the probability of accounts receivable delinquency for each level decrease in a customer's credit rating, this percentage can be used as the transmission strength coefficient for that association rule. Based on support, confidence, and transmission strength coefficients, association rules with higher reliability and influence are selected to determine the identified association rules.

[0056] The generated expert-argumented rules and association-mining rules are integrated. During the integration process, duplicate or conflicting rules are coordinated and optimized. If both expert rules and association-mining rules relate to the impact of customer credit ratings on accounts receivable risk, but their wording differs slightly, the advantages and reasonableness of both are considered to form a unified rule expression. For example, if an expert rule states "customer overdue payments > 3 times..." → If there is a conflict between the "high risk" rule and the data rule "more than 2 overdue instances," the rule with higher strength will be retained based on the priority of the strength coefficient (expert rule R_s = 0.5 vs. data rule S_m = 0.7). The merged rules are stored in a graph database (Neo4j), where nodes represent rule conditions (e.g., "declining industry prosperity index"), edges represent transmission paths (e.g., "leading to overdue risk"), and edge weights are the strength coefficients. Each rule is assigned a unique identifier for easy querying and management. The rule base can be stored and managed using a database management system to ensure data security and scalability. A rule update mechanism is also established to regularly update and maintain the rule base as new expert knowledge, historical data, and business environment changes occur, ensuring the timeliness and accuracy of the rules. When the number of new risk cases exceeds a threshold, the FP-Growth algorithm is triggered to re-mine associated rules and update the corresponding rule strength.

[0057] Through the above technical solution, a multi-dimensional risk assessment system is constructed by integrating expert knowledge rules and data-driven association rules. Expert rules extract the experience and logic of seasoned practitioners (e.g., "a continuous decline in the industry index triggers the probability of overdue payments"), while association rule mining reveals potential statistical correlations (e.g., "accounts payable turnover rate is significantly correlated with risk"). A graph database is used to store rule nodes (e.g., "decline in the industry index") and transmission edges (e.g., "leading to deterioration of customer liquidity"), with edge weights and association strength coefficients. When a risk event is triggered, a visual transmission chain is automatically generated (e.g., raw material price increases). → Customer gross margin decreased → Overdue accounts →(Increased bad debt provisions), and marked the impact weight of key nodes (e.g., customer contribution accounts for 62%). Compared with traditional two-dimensional risk scoring, it effectively improves the efficiency of tracing the causes of risks and allows for the customization of targeted collection strategies based on the transmission path.

[0058] Using this set of key risk factors, causal chain reasoning is performed on the accounts receivable forecast data to determine the risk causal chain data corresponding to the accounts receivable. Specifically, this includes: pre-setting a risk transmission rule base, wherein the risk transmission rule base includes multiple preset rules, each of which includes a precondition; based on the multiple key risk factors in the set of key risk factors, traversing the preset rules in the risk transmission rule base, and determining at least one current triggering rule and the triggering risk factor corresponding to each current triggering rule by the matching status of the key risk factor and the rule precondition; generating the effect intensity corresponding to each current triggering rule by using the preset intensity coefficient of each current triggering rule and the contribution weight corresponding to the triggering risk factor; and sorting the current triggering rules according to the order of their effect intensity to generate a causal chain list data described in natural language.

[0059] In one embodiment of this specification, a distributed computing framework (Apache Spark) is used to perform parallel scanning of the rule base using multiple key risk factors from the set of key risk factors. The input data features are compared item by item with the rule antecedent conditions. For example, key risk factors include: customer dimension accounts payable turnover rate = 0.65 (industry average × 0.8 is the critical value); market dimension industry prosperity index has declined by 6.2% for four consecutive months; product dimension current sales gross profit margin has decreased by 12% compared to the previous quarter. Taking rule R_2023_001 as an example, it is determined whether all of the following conditions are met: Does the industry prosperity index meet the condition of "continuous"? ≥ A decline over three months and a single-month decline ≥ 5% (The current input meets the 6.2% decrease over 4 months, triggering successfully) Is the accounts payable turnover rate < industry average × 0.7 (The input value is 0.65; if the industry average is 0.9, then the threshold is 0.63, which is met here)? If all the antecedent conditions of a rule are matched, it is marked as "currently triggered rule". In the above case, rule R_2023_001 is triggered.

[0060] For each triggering rule, the strength of each triggering rule is generated based on the contribution weight of the key risk factors and the strength coefficient of the current triggering rule. For example, if the strength coefficient is 0.8, and the contribution weights of the two triggering risk factors are 0.7 and 0.3 respectively, then the strength of this rule is 0.8 × (0.7 + 0.3) = 0.8. All triggering rules are sorted in descending order of strength to ensure that the sorted list clearly shows the differences in the degree of impact of different rules on accounts receivable risk, with rules with greater strength appearing first. Natural language descriptions are generated, automatically producing readable text using an NLP template engine, with the transmission path indicated by arrows or other symbols. For example, "Due to customer A's credit rating being downgraded from A to C within one month and the industry issuing significant negative policies restricting industry development within the past three months---the customer's accounts receivable delinquency risk has significantly increased, with a strength of 0.8." The description details the triggering risk factors, risk consequences, and strength of effect, making the results easy to understand. The converted natural language descriptions are then integrated sequentially into a risk causal chain list data.

[0061] Through the aforementioned technical solution, and by using a pre-set risk transmission rule base and a matching mechanism for key risk factor sets, the risk factors affecting accounts receivable and their transmission paths can be comprehensively and accurately identified. Compared to traditional risk assessment methods, it is no longer limited to single factors or simple correlations, but can delve into complex causal relationships. By traversing the rule base, it accurately identifies the current triggering rules and triggering risk factors, clearly presenting the source and propagation chain of risks, accurately grasping the core of the risk, and providing precise targets for subsequent risk response. The generated causal chain list data, described in natural language, is intuitive and easy to understand, providing a clear risk panorama for corporate management and financial personnel. Based on the ranking of the impact intensity, the severity and priority of different risks can be quickly determined, enabling scientific decisions in formulating credit policies, sales strategies, and collection plans. For risks with high impact intensity, collection efforts are increased or credit limits are adjusted; for risks with low impact intensity, resources are rationally allocated for monitoring, improving resource utilization efficiency and balancing business development and risk control.

[0062] Step S104: Based on risk causal chain data and accounts receivable forecast data, visualize the risks of accounts receivable to achieve risk management of accounts receivable.

[0063] Based on the risk causal chain data and the accounts receivable forecast data, a visual risk display of the accounts receivable is performed. Specifically, this includes: obtaining a list of causal chains from the risk causal chain data, wherein the list includes multiple sequentially arranged risk causal chains; mapping each causal node in the risk causal chain to an entity node, mapping the transmission direction in the risk causal chain to directed edges, and setting the directed edge display parameters according to the list order corresponding to the risk causal chain; setting the entity display parameters for each entity node based on the risk forecast results in the accounts receivable forecast data using color-coded rules; and constructing a risk causal graph for the accounts receivable using the directed edge display parameters and the entity display parameters to visually display the risks of the accounts receivable.

[0064] In one embodiment of this specification, for each risk causal chain, causal nodes are mapped to entity nodes. Causal nodes may include various risk factors and events such as declining customer credit ratings, intensified market competition, and adjustments to internal sales policies. Each entity node represents a specific risk-related factor or event, presented graphically in a visualization, such as a circle, square, or other custom shape. Each entity node is assigned a unique identifier for accurate identification and management in subsequent operations. The transmission direction in the risk causal chain is mapped to directed edges connecting the entity nodes, clearly showing the risk propagation path. The directed edge display parameters, including color, thickness, and transparency, are set according to the list order corresponding to the risk causal chains. Directed edges for earlier risk causal chains can be set to brighter colors and thicker lines to highlight their importance; while directed edges for later causal chains are adjusted to lighter colors and thinner lines. In this way, users can intuitively understand the relative importance and order of different risk causal chains within the overall risk. Based on the risk prediction results in the accounts receivable forecast data, the entity display parameters for each entity node are set using color-coded rules. For example, if the risk prediction results indicate a high risk of overdue accounts receivable for a certain customer, the entity node representing the relevant risk factors for that customer will be set to a red color, with the shade adjusted according to the level of risk—the higher the risk, the darker the color; if the risk is low, it will be set to a green color. Simultaneously, the size of the entity nodes can be adjusted according to the risk prediction results—the greater the risk, the larger the node—further enhancing the visualization effect and allowing users to quickly identify high-risk areas.

[0065] Based on the pre-defined entity nodes and directed edge connections, a risk causal graph for accounts receivable is constructed using visualization tools (such as Graphviz and D3.js). During construction, the positions of entity nodes are strategically arranged to avoid overlap and clutter, ensuring the graph's readability. Hierarchical layout and force-directed layout algorithms can be employed to make the graph structure clearer and the risk transmission path readily apparent. The constructed risk causal graph is then displayed on a visualization interface, providing users with an intuitive view of risk. Interactive features are designed to enhance user experience. An interactive graph is generated on the front-end interface, supporting the expansion of secondary causal chains by clicking on nodes, and displaying the corresponding multi-dimensional accounts receivable feature vectors and gradient contribution weights. A WebGL dynamic rendering engine is invoked to synchronously update the real-time prediction data output by the accounts receivable prediction model with the node attributes of the visualization graph. Responding to user-input filtering conditions, nodes in the visualization graph are dynamically filtered, retaining only causal chain paths with overdue probabilities exceeding a preset threshold or influence strength greater than 0.2. A list of key rules for the causal chain paths is extracted to generate a risk report document, and log data is integrated with the financial audit system interface. In addition, a filtering function is provided, allowing users to filter and display partial risk causal chains based on specific conditions (such as risk type, risk level, etc.), focusing on key risks. Since accounts receivable risk is dynamic, the risk causal graph needs to be updated regularly. Based on newly acquired risk causal chain data and accounts receivable forecast data, the display parameters of entity nodes and directed edges are recalculated, and the graph content is updated.

[0066] The above technical solution transforms the risk causal chain into a visualized risk causal graph, graphically displaying risk factors (such as declining customer credit ratings and intensified market competition) and risk events and their transmission paths. This intuitive presentation allows enterprise personnel, especially non-technical finance and management personnel, to quickly understand complex risk relationships, eliminating the need to spend significant time analyzing and interpreting abstract data and textual reports, thus improving the efficiency of risk awareness. By setting directed edge display parameters, adjusting the color, thickness, and transparency of the directed edges according to the order of the risk causal chain list, users can intuitively understand the relative importance and sequence of different risk causal chains. This helps prioritize risk factors that have a greater impact on overall risk and transmit earlier, rationally allocate risk management resources, avoid neglecting key issues due to focusing on secondary risks, and improve the targeting and effectiveness of risk management.

[0067] This specification's embodiments comprehensively collect multi-dimensional accounts receivable data, including customer, product, market, and internal management data. This overcomes the limitations of traditional methods that rely solely on single historical accounts receivable data. By preprocessing multi-dimensional data using pre-built dimensional correlation heatmaps, it uncovers potential relationships between data points and determines more representative multi-dimensional accounts receivable feature vectors. Subsequent prediction models receive rich and high-quality data, enabling them to better capture various factors influencing accounts receivable and significantly improve prediction accuracy. Based on these multi-dimensional feature vectors, a pre-trained accounts receivable prediction model determines predicted data and further identifies a set of key risk factors. This overcomes the limitations of existing "black box" models that only output prediction results, clearly identifying the core risk drivers affecting accounts receivable. Through analysis of this set of key risk factors... It can accurately determine the contribution weight of factors such as customer payment delay rates to overdue payments; it uses a set of key risk factors to perform causal chain reasoning on accounts receivable forecast data to obtain risk causal chain data; it clearly shows the causes and transmission paths of risks, no longer viewing risk factors and forecast results in isolation, but intuitively understanding how different risk factors interact and ultimately affect accounts receivable. The clear presentation of risk transmission paths helps to fully grasp the risk formation mechanism and prevent and block the spread of risks in advance; combining risk causal chain data and accounts receivable forecast data, it provides a visualized risk display for accounts receivable. Unlike the traditional method of displaying risk scores only through tables or two-dimensional charts, it can dynamically track the causes of risks, more intuitively identify the core risk causes and their transmission paths, quickly locate high-risk areas and key risk links, and improve the efficiency of risk handling.

[0068] This specification also provides an accounts receivable management device based on a multilayer sensor, such as... Figure 3 As shown, the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method.

[0069] This specification also provides a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0070] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0071] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0072] The devices, media, and methods provided in the embodiments of this specification are one-to-one correspondences. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0073] Those skilled in the art will understand that embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may 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.

[0074] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. 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, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] 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.

[0076] 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.

[0077] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0078] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0079] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0080] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, the phrase "comprising one..." …… "The definition of a particular element does not preclude the presence of other identical elements in the process, method, product, or apparatus that includes the element."

[0081] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for managing accounts receivable based on a multi-layer perceptron, characterized by, The method comprises: acquiring multi-dimensional account data corresponding to the accounts receivable, preprocessing the multi-dimensional account data by using pre-constructed dimension correlation heat data, and determining a multi-dimensional account feature vector, wherein the multi-dimensional account data comprises account customer dimension data, account product dimension data, account market dimension data, and enterprise internal management data; determining accounts receivable prediction data corresponding to the accounts receivable according to the multi-dimensional account feature vector and a pre-trained accounts receivable prediction model, and determining a key risk factor set corresponding to each of the accounts receivable based on the accounts receivable prediction data; performing causal chain reasoning on the accounts receivable prediction data through the key risk factor set to determine risk causal chain data corresponding to the accounts receivable; performing visual risk display on the accounts receivable based on the risk causal chain data and the accounts receivable prediction data to realize risk management of the accounts receivable.

2. The method of claim 1, wherein, Before determining the accounts receivable prediction data corresponding to the accounts receivable according to the multi-dimensional account feature vector and the pre-trained accounts receivable prediction model, the method further comprises: collecting historical accounts receivable data and historical multi-dimensional account time series data corresponding to the historical accounts receivable data, wherein the historical accounts receivable data comprises accounts receivable amount data, account age distribution data, and repayment period data; performing abnormal data processing and missing value filling on the historical accounts receivable data and the historical multi-dimensional account time series data to determine an accounts receivable data set; training a pre-constructed initial multi-layer perception model by using the accounts receivable data set through a pre-set optimization strategy to determine an accounts receivable prediction model meeting pre-set requirements, wherein the number of hidden layers of the initial multi-layer perception model is determined based on one-time neural architecture search combined with weight sharing.

3. The accounts receivable management method based on a multilayer sensor according to claim 2, characterized in that, The preprocessing of the multi-dimensional account data by using the pre-constructed dimension correlation heat data to determine the multi-dimensional account feature vector specifically comprises: determining the historical multi-dimensional account time series data, performing feature extraction on the historical multi-dimensional account time series data according to each account dimension to determine historical dimension feature data corresponding to each account dimension, wherein the account dimensions comprise customer dimension, product dimension, market dimension, and internal management dimension; performing feature correlation analysis on the historical dimension feature data corresponding to each account dimension to determine feature correlation coefficients between the account dimensions, and constructing dimension correlation heat data based on the feature correlation coefficients; determining missing dimensions in the multi-dimensional account data, filling data of the missing dimensions by using the dimension correlation heat data to determine current multi-dimensional account data, performing feature extraction on the current multi-dimensional account data to determine a multi-dimensional account feature vector.

4. The method of claim 1, wherein, The determination of the key risk factor set corresponding to each of the accounts receivable based on the accounts receivable prediction data specifically comprises: Contribution weights of each account feature in the multi-dimensional account feature vector to the accounts receivable prediction data are calculated through a preset gradient approximation algorithm. At least one key risk feature is determined from the account features by filtering according to the contribution weights of each account feature to the accounts receivable prediction data and a preset threshold, so as to construct the key risk factor set.

5. The accounts receivable management method based on a multilayer sensor according to claim 4, characterized in that, The contribution weights of each account feature in the multi-dimensional account feature vector to the accounts receivable prediction data are calculated through a preset gradient approximation algorithm, specifically including: The input feature gradient matrix corresponding to the multi-dimensional account feature vector is calculated by back propagation of the output of the accounts receivable prediction model. The contribution weights of each account feature in the multi-dimensional account feature vector to the accounts receivable prediction data are calculated by integral path method based on the input feature gradient matrix.

6. The method of claim 1, wherein the method is based on a multi-layer perceptron. The risk causal chain data corresponding to the accounts receivable is determined by causal chain reasoning of the accounts receivable prediction data through the key risk factor set, specifically including: A risk transmission rule base is preset, wherein the risk transmission rule base includes a plurality of preset rules, and each rule includes a rule antecedent condition. At least one current trigger rule and a trigger risk factor corresponding to each current trigger rule are determined by matching the key risk factors in the key risk factor set with the rule antecedent conditions, so as to traverse the preset rules in the risk transmission rule base. The action strength of each current trigger rule is generated by a preset strength coefficient of each current trigger rule and a contribution weight corresponding to the trigger risk factor. The current trigger rules are sorted according to the size order of the action strength to generate a causal chain list data in natural language description.

7. The method of claim 1, wherein the method is based on a multi-layer perceptron. The method further includes: Expert rule discussion texts are collected to convert the expert rule discussion texts into expert discussion rules based on natural language processing technology, wherein the expert discussion rules include discussion rule transmission data and a preset rule strength coefficient. Historical accounts receivable data tables corresponding to historical risk cases are collected, wherein the historical accounts receivable data tables include customer dimension fields, account product dimension fields, account market dimension fields, enterprise internal management fields and risk result fields. Association rule mining is performed on the historical accounts receivable data tables to obtain mining rule transmission data of specified feature combinations and risk results, and a transmission strength coefficient is determined according to the regression relationship between feature changes and risk results in the historical accounts receivable data tables, so as to determine the mining association rules. A risk transmission rule base is constructed by the mining association rules and the expert discussion rules.

8. The method of claim 1, wherein the method is based on a multi-layer perceptron. The accounts receivable are visually displayed based on the risk causal chain data and the accounts receivable prediction data, specifically including: The causal chain list data in the risk causal chain data is obtained, wherein the causal chain list data includes a plurality of sequentially arranged risk causal chains. mapping a causal node in each of the risk causal chains to an entity node, mapping a conduction direction in the risk causal chains to a directed edge connection, and setting a directed edge display parameter of the directed edge connection through a list order corresponding to the risk causal chains; setting an entity display parameter of each of the entity nodes according to a risk prediction result in the accounts receivable prediction data by using a color scale coding rule; constructing a risk causal graph of the accounts receivable to perform visual risk display of the accounts receivable through the directed edge display parameter and the entity display parameter.

9. A multi-layer perceptron-based accounts receivable management device, comprising: The device comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

10. A non-transitory computer storage medium storing computer-executable instructions that, when executed, cause a computer to perform: The computer executable instructions are configured to perform the method of any one of claims 1-8.

Images