Coefficient value determination method and device, equipment and storage medium

Abstract

The invention provides a coefficient value determination method and device, equipment and a storage medium, and relates to the technical field of data processing. According to the specific technical scheme, time sequence feature data of a target replacement contract are acquired; the target replacement contract is used for appointing preset resources for converting the target object into the target value. The time sequence feature data comprises feature parameter sets of the target replacement contract at a plurality of historical moments. And constructing a feature matrix of the target replacement contract based on the time sequence feature data of the target replacement contract. Each row of the feature matrix represents a feature parameter set of the target replacement contract at a single historical moment, and each column of the feature matrix represents a feature parameter of the target replacement contract at different historical moments. And performing feature dimension reduction processing on the feature matrix to obtain a feature low-dimensional matrix of the target replacement contract. And processing the feature low-dimensional matrix through a preset coefficient model to obtain a target parameter value of the target replacement contract. The method is used for improving the accuracy of contract value conversion coefficient replacement.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular to a coefficient value determination method and device, equipment and a storage medium. BACKGROUND

[0002] In the application scenario of resource exchange, quantifying the conversion ratio between different resources is a basic technical problem. The determination of this ratio directly affects the fairness and security of the exchange process.

[0003] Currently, the calculation of the conversion ratio is usually obtained by directly calculating the value of the resources. For example, the value ratio of the two parties in the exchange of resources is used as the conversion ratio.

[0004] However, due to the time span of some resource exchange processes, the value of the related resources may fluctuate, causing the calculated conversion ratio to deviate from the actual value of the resources, thereby reducing its accuracy and applicability in actual application scenarios. SUMMARY

[0005] Embodiments of the present application provide a coefficient value determination method, device, equipment and storage medium, which can obtain a more accurate value conversion coefficient by performing time series analysis on the time series feature data of the replacement contract through a preset coefficient model.

[0006] To achieve the above purpose, embodiments of the present application adopt the following technical solutions: In a first aspect, a coefficient value determination method is provided, which comprises: First, the time series feature data of the target replacement contract is obtained, the target replacement contract being used to agree on a preset resource converted into a target value by a target object. The time series feature data includes a feature parameter set of the target replacement contract at a plurality of historical time points, and the feature parameter set at each historical time point includes: attribute feature parameters of the target replacement contract, feature parameters of the target object, feature parameters of the target object's owner, feature parameters of the target object's field, and breach disposal strategy feature parameters of the target replacement contract. Second, based on the time series feature data of the target replacement contract, a feature matrix of the target replacement contract is constructed. Each row of the feature matrix represents the feature parameter set of the target replacement contract at a single historical time point, and each column of the feature matrix represents a kind of feature parameter of the target replacement contract at different historical time points. Then, the feature matrix is processed by feature dimension reduction to obtain a feature low-dimensional matrix of the target replacement contract. Finally, the feature low-dimensional matrix is processed by a preset coefficient model to obtain a target parameter value of the target replacement contract, the target parameter value being a value conversion coefficient of the target object agreed by the target replacement contract, and the preset coefficient model having the function of determining the parameter value of the object agreed by the replacement contract based on the feature low-dimensional matrix.

[0007] The embodiment of the application provides a coefficient value determination method, which comprises the following steps: obtaining multi-dimensional feature parameters of a target replacement contract at multiple historical time points, and constructing a time sequence feature matrix, wherein the feature parameter system covers contract attributes, resource characteristics, the state of the owner, field dynamics, and default disposal strategies and other key influencing factors, and can comprehensively reflect the dynamic characteristics of the contract. On this basis, the feature low-dimensional matrix of the target replacement contract is obtained by dimension reduction processing of the feature matrix, and the feature combination most influencing the value conversion coefficient is effectively extracted. Compared with the prior art in the related art which directly calculates the conversion ratio based on the resource value, the method analyzes the comprehensive influence of the time sequence change of the multi-dimensional feature parameters on the value conversion coefficient, and uses a preset coefficient model to mine deep time sequence rules from the feature low-dimensional matrix, effectively overcomes the evaluation deviation caused by the traditional static calculation method when the value fluctuates due to the time span of the resource exchange process, and thus realizes accurate evaluation of the value conversion coefficient in a dynamic change environment, and significantly improves the accuracy and applicability of the conversion ratio in the actual application scenario.

[0008] In a possible implementation form of the first aspect, the attribute feature parameter of the target replacement contract comprises: a survival period of the target replacement contract, a use of the target replacement contract, and a third-party reserved resource of the target replacement contract. The feature parameter of the target object comprises: a type of the target object, a historical value change rate of the target object, and a depreciation rate of the target object. The feature parameter of the owner of the target object comprises: a performance feature parameter of the owner of the target object, and a credit identification of the owner of the target object. The feature parameter of the field to which the target object belongs comprises: a development trend index of the field to which the target object belongs, and a liquidity index of the field to which the target object belongs. The default disposal strategy feature parameter of the target replacement contract comprises: a type of the default disposal strategy, a default duration of the target replacement contract, and a default recovery preset resource of the target replacement contract.

[0009] It should be understood that the scheme establishes a comprehensive quantitative description framework for the replacement contract by constructing a multi-dimensional feature parameter system comprising a survival period, a resource type, an owner performance, a field development trend, and a default disposal strategy. Compared with the prior art which only considers a single value dimension, the multi-dimensional feature system can more comprehensively capture key factors influencing the value conversion coefficient, and provides a solid data foundation for subsequent accurate evaluation.

[0010] In a possible implementation manner of the first aspect, the feature matrix is subjected to feature dimension reduction processing to obtain a feature low-dimensional matrix of the target substitution contract, including: performing normalization processing on each column of feature parameter sets in the feature matrix to obtain a processed feature matrix. The processed feature matrix is subjected to covariance processing, and a covariance matrix obtained is taken as a correlation coefficient matrix. The correlation coefficient matrix is brought into a feature equation for solving to obtain a plurality of eigenvalues of the correlation coefficient matrix and a feature vector corresponding to each eigenvalue. The processed feature matrix is projected to a subspace formed by a feature vector corresponding to a first eigenvalue to obtain the feature low-dimensional matrix of the target substitution contract. The first eigenvalue refers to an eigenvalue selected from the front N eigenvalues in descending order of feature contribution rate, and a total of feature contribution rates of the N eigenvalues is greater than a preset feature contribution rate threshold, where N is a positive integer.

[0011] It should be understood that, by subjecting the feature matrix to feature dimension reduction processing, redundant information between features can be effectively eliminated, and a feature combination most influential to the value conversion coefficient can be extracted. This processing not only reduces the calculation complexity, but also improves the quality of feature data, so that the subsequent model can focus more on the learning of key features, thereby improving the accuracy and stability of the model.

[0012] In a possible implementation manner of the first aspect, the target parameter value includes a highest value conversion coefficient and a recommended value conversion coefficient. The highest value conversion coefficient is used as an upper limit for adjusting the value conversion coefficient of the target substitution contract, and the recommended value conversion coefficient is used as a reference for adjusting the value conversion coefficient of the target substitution contract. The method further includes: obtaining a to-be-evaluated value conversion coefficient of the target substitution contract. The to-be-evaluated value conversion coefficient is a value conversion coefficient of the target substitution contract determined at a historical time. In a case where the to-be-evaluated value conversion coefficient is greater than the highest value conversion coefficient, a first evaluation result of the to-be-evaluated value conversion coefficient is generated. The first evaluation result is used to indicate adjustment of the to-be-evaluated value conversion coefficient. In a case where the to-be-evaluated value conversion coefficient is less than the highest value conversion coefficient and greater than the recommended value conversion coefficient, a second evaluation result of the to-be-evaluated value conversion coefficient is generated. The second evaluation result is used to suggest adjustment of the to-be-evaluated value conversion coefficient.

[0013] It should be understood that, by setting the dual evaluation benchmarks of the highest value conversion coefficient and the recommended value conversion coefficient and establishing the corresponding evaluation result generation mechanism, the scheme realizes hierarchical early warning and dynamic adjustment of the value conversion coefficient. Compared with a single fixed ratio, this hierarchical evaluation mechanism can provide more refined safety control, which not only ensures the safety of the substitution process, but also maintains reasonable flexibility.

[0014] In another possible implementation of the first aspect, the preset coefficient model is obtained through the following method: First, obtain the temporal feature data corresponding to each of the multiple sample permutation contracts within a preset historical time period. Second, determine the feature dimensionality reduction matrix corresponding to each sample temporal feature data. Third, based on the credibility identifier of the sample object's owner for each sample permutation contract, construct a value conversion coefficient label matrix that corresponds one-to-one with each sample temporal feature matrix. Each row of the temporal feature matrix corresponds to each row of the corresponding value conversion coefficient label matrix. Fourth, train the initial model of the preset coefficient model based on each feature low-dimensional matrix and its corresponding value conversion coefficient label matrix to obtain the preset coefficient model.

[0015] It should be understood that this approach uses time-series data based on historical sample contracts as training data, enabling the model to learn complex patterns of multidimensional features changing over time and their intrinsic impact on value discounting factors. This deep learning-based method can effectively capture nonlinear time-series patterns that are difficult to discover using traditional statistical methods, thereby achieving more accurate evaluations in dynamic environments.

[0016] In another possible implementation of the first aspect, for any one of the first sample replacement contracts in each sample replacement contract: Based on the credibility identifier of the target object's owner in each sample replacement contract, a value conversion coefficient label matrix corresponding one-to-one with the temporal feature matrix of each sample is constructed, including: Performing a first operation: If the credibility identifier of the target object's owner corresponding to the first sample replacement contract is credible, the recommended value conversion coefficient label of the first sample replacement contract at the first historical moment is determined based on the ratio of the quantified value of the target object of the first sample replacement contract at the first historical moment to the quantified value of the preset resource. Performing a second operation: The preset highest value conversion coefficient corresponding to the target object type of the first sample replacement contract is used as the highest value conversion coefficient label of the first sample replacement contract at the first historical moment. Performing a third operation: The second historical moment is used as the new first historical moment. The second historical moment is later than the first historical moment. The first operation, the first operation, and the third operation are repeated until the first historical moment is null, obtaining the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and constructing the value conversion coefficient label matrix corresponding to the first sample replacement contract.

[0017] It should be understood that by using a label generation method based on the ratio of the target object to the pre-defined resource value for resource owners with credible identifiers, the decisive role of resource value in the conversion factor can be accurately reflected under stable performance conditions. This label construction mechanism enables the trained model to fully learn the inherent correlation between resource value and conversion factor under normal performance scenarios, thereby providing a value conversion assessment that matches the performance characteristics of replacement contracts with credible identifiers.

[0018] In another possible implementation of the first aspect, for any first sample replacement contract in each sample replacement contract, a value conversion coefficient label matrix corresponding one-to-one with the temporal feature matrix of each sample is constructed based on the credibility identifier of the target object's owner in each sample replacement contract. This includes: performing a fourth operation: if the credibility identifier of the sample object's owner corresponding to the first sample replacement contract is untrustworthy, the historical value conversion coefficient of the first sample replacement contract is used as the highest value conversion coefficient label of the first sample replacement contract at the first historical moment. Performing a fifth operation: based on the ratio of the cumulative quantified value of the pre-defined resources recovered by the first sample replacement contract at the first historical moment to the quantified value of the pre-defined resources in the first sample replacement contract, the recommended value conversion coefficient label of the first sample replacement contract at the first historical moment is determined. Performing a sixth operation: using the second historical moment as the new first historical moment. The second historical moment is later than the first historical moment. Repeating the fourth, fifth, and sixth operations until the first historical moment is null, the recommended value conversion coefficient labels and the highest value conversion coefficient labels of the first sample replacement contract at multiple historical moments are obtained, and the value conversion coefficient label matrix corresponding to the first sample replacement contract is constructed.

[0019] It should be understood that by employing a label generation method based on historical recycling data for resource owners with untrusted identifiers, the adjustment effect of resource recycling status on the conversion factor during actual performance can be effectively reflected. This differentiated label construction method enables the model to learn how to dynamically adjust the conversion factor based on actual recycling performance under different performance statuses, thereby providing a value assessment scheme adapted to the performance history of replacement contracts with different identifiers.

[0020] Secondly, a coefficient value determining device is provided, the device comprising: The acquisition module is used to acquire the time-series feature data of the target substitution contract. The target substitution contract is used to define the preset resources for converting the target object into the target value. The time-series feature data includes the feature parameter set of the target substitution contract at multiple historical moments. The feature parameter set at each historical moment includes: the attribute feature parameters of the target substitution contract, the feature parameters of the target object, the feature parameters of the party to which the target object belongs, the feature parameters of the domain to which the target object belongs, and the default handling strategy feature parameters of the target substitution contract.

[0021] The processing module constructs a feature matrix for the target substitution contract based on its time-series feature data. Each row of the feature matrix represents the set of feature parameters of the target substitution contract at a single historical moment, and each column represents a feature parameter of the target substitution contract at different historical moments. The feature matrix is ​​then subjected to dimensionality reduction processing to obtain a low-dimensional feature matrix of the target substitution contract. This low-dimensional feature matrix is ​​input into a preset coefficient model to obtain the target value conversion coefficient of the target substitution contract. The preset coefficient model has the function of determining the value conversion coefficient of the substitution contract based on the low-dimensional feature matrix.

[0022] Thirdly, a coefficient value determination device is provided, the method comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, the computer program code including computer instructions. When the processor executes the computer instructions, it causes the coefficient value determination device to perform the method as described in the first aspect and any possible implementation thereof.

[0023] Fourthly, a computer-readable storage medium is provided, wherein computer instructions are stored therein. When executed by a processor, the computer instructions are used to implement the method as described in the first aspect and any possible implementation thereof.

[0024] Fifthly, a computer program product is provided that, when run on a computer or executed by a processor of the computer, implements the method described in the first aspect and any possible design thereof. The computer may be the electronic device described in the second aspect and any possible implementation thereof.

[0025] It is understood that the beneficial effects achieved by the coefficient value determining device described in the second aspect, the coefficient value determining apparatus described in the third aspect, the computer-readable storage medium described in the fourth aspect, and the computer program product described in the fifth aspect can be referred to as the beneficial effects in the first aspect and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0026] Figure 1This application provides a schematic diagram of a data processing flow for a computing device. Figure 2 A flowchart illustrating a method for determining coefficient values ​​provided in an embodiment of this application; Figure 3 A flowchart illustrating another method for determining coefficient values ​​provided in an embodiment of this application; Figure 4 A flowchart illustrating yet another method for determining coefficient values ​​provided in this application embodiment; Figure 5 A flowchart illustrating a preset coefficient model training method provided in this application embodiment; Figure 6 This is a schematic diagram of the structure of an LSTM model provided in an embodiment of this application; Figure 7 A flowchart illustrating another preset coefficient model training method provided in this application embodiment; Figure 8 A flowchart illustrating another preset coefficient model training method provided in this application embodiment; Figure 9 This is a schematic diagram of a coefficient value determination device provided in an embodiment of this application; Figure 10 A schematic diagram of another coefficient value determination device provided in the embodiments of this application; Figure 11 This is a schematic diagram of a coefficient value determination device provided in an embodiment of this application. Detailed Implementation

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0028] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0029] The technical solutions provided in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data, comply with relevant laws and regulations and do not violate public order and good morals.

[0030] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0031] In resource exchange scenarios, the relevant value conversion methods typically determine the conversion ratio based on a static value assessment at the time of the exchange. This method only considers the resource value parameter at a single point in time and derives a fixed conversion factor through a simple value ratio calculation.

[0032] However, actual replacement contracts often have long performance periods, during which multi-dimensional parameters such as the target object's value attributes, owner status, and domain environment may dynamically change over time. This static evaluation method cannot perceive and respond to parameter changes, causing the output conversion factor to gradually deviate from the actual situation during contract performance, making it difficult to guarantee the stability and reliability of the replacement process.

[0033] Therefore, how to consider the impact of the dynamic changes of the replacement contract over time on the value conversion factor, so as to improve the accuracy and applicability of the conversion factor in practical applications, is an urgent problem to be solved.

[0034] In view of this, this application provides a method for determining coefficient values. This method collects multi-dimensional feature parameters of the replacement contract at multiple historical moments to construct a time-series feature matrix, comprehensively recording the dynamic changes of the contract. Key feature information is extracted by dimensionality reduction of the feature matrix. Finally, a trained preset coefficient model is used to learn the changing patterns of the value conversion coefficient from the time-series features, achieving accurate evaluation of the target replacement contract's value conversion coefficient. This dynamic evaluation method based on time-series feature analysis can effectively adapt to various changes in the resource replacement process, significantly improving the accuracy and practicality of the conversion coefficient.

[0035] First, the application scenarios of the coefficient value determination method provided in the embodiments of this application will be introduced.

[0036] The coefficient value determination method provided in this application can be applied to computing devices. Specifically, the computing device can be a single server or a server cluster composed of multiple servers, or a computer, or a processor or processing chip in a server or computer, etc. This application does not limit the specific device form of the computing device.

[0037] In an exemplary embodiment, Figure 1 This application provides a schematic diagram of a data processing flow for a computing device, which specifically includes: coefficient model acquisition, substitution contract coefficient determination, and substitution contract coefficient value security assessment.

[0038] The acquisition of coefficient models includes: data collection, data processing, and model training.

[0039] Data acquisition is used to obtain the temporal feature data of the sample permutation contracts required during model training.

[0040] In some embodiments, since the feature parameter set of each historical moment in the time series feature data comes from different data dimensions, the data acquisition specifically includes obtaining feature parameters of contract attribute dimension, resource characteristic dimension, owner status dimension, domain environment dimension and disposal strategy dimension from different data sources, and aligning and integrating them according to a unified time series.

[0041] Data processing is used to construct feature matrices and perform dimensionality reduction on the collected sample time-series feature data to obtain low-dimensional feature matrices and value conversion coefficient matrices of the sample time-series feature data.

[0042] Model training is used to train a model based on the low-dimensional feature matrix and value conversion coefficient matrix labels obtained after data processing, so as to obtain a model with preset coefficients.

[0043] Determining the coefficients of the permutation contract includes: obtaining the target permutation contract and model inference.

[0044] The target permutation contract acquisition is used to obtain the temporal feature data of the target permutation contract, and to construct and reduce the dimensionality of the temporal feature data of the target permutation contract to obtain the low-dimensional feature matrix corresponding to the target permutation contract.

[0045] Model inference is used to input the low-dimensional feature matrix corresponding to the target permutation contract into a preset coefficient model to obtain the target parameter value of the target permutation contract.

[0046] The safety assessment of the replacement contract coefficient value is used to compare and analyze the value conversion coefficient of the value to be assessed with the value conversion coefficient output by the model. Based on the comparison results, the assessment results are generated, providing a basis for decision-making on adjusting the value conversion coefficient of the replacement contract.

[0047] For a detailed description of the above data processing flow, please refer to the following embodiments, which will not be elaborated here.

[0048] This application can be applied to resource exchange processes in different fields.

[0049] For example, in a collateralized financing scenario, a company uses its liquid inventory, such as raw materials and finished products, as collateral (target assets) to apply for a short-term working capital loan from a financial institution (pre-set resources). Since the market value of collateral is easily affected by factors such as price fluctuations and product category iterations, this method can be used to dynamically assess its value conversion factor (i.e., value conversion factor) to improve asset utilization efficiency while ensuring financing security.

[0050] For example, in the scenario of network computing power leasing, service providers use their idle virtual computing resources such as cloud computing and rendering processing (target objects) as collateral to provide demanders with a credit limit (pre-set resources) for use first and payment later. Based on the real-time supply and demand of computing resources and the time-series characteristics such as hardware performance degradation, this method can dynamically determine the proportion of the credit limit that can be overdrawn (value conversion factor), thereby achieving optimal resource allocation and risk control.

[0051] For example, in industrial equipment financing lease scenarios, equipment manufacturers use their specialized production equipment (the target asset) as the leased asset, providing installment payment services (pre-set resources) to lessees with insufficient funds. Based on dynamic parameters such as the technological advancement of the equipment model, usage intensity and depreciation status, and the prosperity of the industry, this method can be used to determine the depreciation ratio between the installments during the lease period and the equipment value, effectively managing the risks arising from fluctuations in the equipment's residual value.

[0052] Therefore, this application does not impose specific limitations on the specific application fields and resource types.

[0053] When the coefficient value determination method provided in this application is applied to the aforementioned computing device, such as... Figure 2 As shown, it specifically includes the following: S101. Obtain the time-series characteristic data of the target replacement contract.

[0054] Among them, the target replacement contract is used to agree on the preset resources for converting the target object into the target value.

[0055] Specifically, a target replacement contract refers to a legal or electronic agreement that stipulates the rules for resource conversion; the target object refers to the resource entity with specific attributes used for conversion in the contract; and the preset resource refers to the predefined resource form that is ultimately obtained through the conversion process.

[0056] The time-series feature data includes feature parameter sets of the target replacement contract at multiple historical moments. Specifically, the feature parameter set at each historical moment includes: attribute feature parameters of the target replacement contract, feature parameters of the target object, feature parameters of the party to which the target object belongs, feature parameters of the domain to which the target object belongs, and default handling strategy feature parameters of the target replacement contract.

[0057] Historical moments refer to discrete time points determined according to a preset collection frequency, from the contract's inception to the current evaluation point. These time points constitute a continuous time series for observing the evolution of the contract's state, ensuring that the dynamic changes of various characteristic parameters during the contract's existence can be fully captured.

[0058] For example, the preset historical time is determined based on the preset collection frequency, which can be daily, weekly, or monthly, etc. This application embodiment does not limit the specific collection frequency of the preset historical time.

[0059] Specifically, time-series characteristic data can be understood as a sequence organized in chronological order that reflects the dynamic evolution of the contract state. Historical moments refer to discrete observation points selected at a preset frequency from the start of the contract to the current evaluation point.

[0060] The attribute characteristics of a target replacement contract refer to the inherent attributes of the contract itself, providing the basic framework and constraints for the value conversion factor. These include: duration, i.e., the total length of time from contract effectiveness to termination; contract purpose, referring to the scope and purpose of resource use stipulated in the contract; and third-party reserved resources, referring to the amount of guarantee resources agreed upon in the contract to be held in custody by an independent third party.

[0061] The characteristic parameters of the target object refer to the core attributes of the resource being converted, forming the objective basis for determining the value conversion factor. These include: type, i.e., the classification and characteristic description of the resource; historical value change rate, referring to the amplitude and frequency of value fluctuations of the resource in the past period; and depreciation rate, referring to the rate at which the value of the resource depreciates per unit time due to natural wear and tear or technological iteration.

[0062] The characteristic parameters of the target entity refer to the creditworthiness of the resource holder, providing a basis for adjusting the credit risk for the value conversion factor. These include: performance characteristic parameters, namely, the entity's performance ratings in historical contracts, such as the quality of performance and response speed; and credibility indicators, which refer to the reliability level rating based on the entity's historical credit records.

[0063] The characteristic parameters of the target object's industry refer to the environmental indicators of the industry in which the resource is located, providing a market-adaptive correction reference for the value conversion factor. These include: development trend indicators, which are a comprehensive development trend score calculated using data such as industry output growth rate, technology iteration speed, and market capacity changes; and liquidity indicators, which refer to the activity level of resource transactions within the industry, measured comprehensively using parameters such as transaction frequency, transaction size, and price elasticity.

[0064] The default handling strategy characteristic parameters of the target replacement contract refer to the default handling mechanism stipulated in the contract, which provides a safety margin for the value conversion factor. These include: handling type, i.e., the default handling method stipulated in the contract, such as negotiated transfer, public disposal, etc.; default duration, referring to the time span from the occurrence of the default event to the completion of the handling; and default recovery preset resources, referring to the quantity of standard resources or equivalents actually recovered through the default handling procedure.

[0065] In some embodiments, the specific parameter types of development trend indicators include the industry output growth rate, technology iteration rate, market capacity change rate, policy support index, and cost input growth rate.

[0066] Among them, the industry output growth rate is calculated by monitoring the output data of major players in the industry; the technology iteration rate is quantified based on the changes in the frequency of technology updates and the number of patents in the industry; the market capacity change rate is determined by analyzing the changing trends of market demand and the number of users; and the cost input growth rate is calculated by statistically analyzing the changes in the investment scale and frequency in the industry.

[0067] In some embodiments, the method for acquiring time-series feature data includes the following steps: first, establishing data connections with the contract management system, resource monitoring platform, and external data sources; then, automatically collecting feature parameters of each dimension according to a preset time frequency; next, cleaning and standardizing the collected data, including handling missing values, eliminating outliers, and unifying the data format; and finally, integrating the processed data into a structured feature parameter set in chronological order.

[0068] One possible implementation is that the computing device triggers the acquisition of time-series feature data under the following circumstances: when a request to calculate the value conversion factor for the target replacement contract is received; when a significant fluctuation in the value of the target object is detected; when the credibility identifier of the party to which the target object belongs changes; when the contract enters a specific lifecycle node; or when the system executes a batch evaluation task according to a preset cycle. This application embodiment does not limit the specific triggering method for acquiring time-series feature data.

[0069] S102. Based on the time-series feature data of the target permutation contract, construct the feature matrix of the target permutation contract.

[0070] In this matrix, each row represents the set of feature parameters of the target permutation contract at a single historical moment, while each column represents a feature parameter of the target permutation contract at different historical moments.

[0071] A feature matrix is ​​a two-dimensional data structure formed by arranging time-series feature data according to specific rules. The row dimension represents the time series, and the column dimension represents the feature space, forming a two-dimensional representation of time and features. The significance of constructing a feature matrix lies in transforming unstructured time-series data into a standardized input format that can be processed by machine learning. This provides a standardized data foundation for subsequent feature dimensionality reduction and model training, while preserving the dynamic information of feature changes over time.

[0072] One possible implementation involves constructing the feature matrix using a sliding time window mechanism. A fixed-length time window is set and slides across the time-series data with a specific step size; the data within each window constitutes a sub-matrix. This approach ensures the continuity of the time series while generating multiple training samples, improving data utilization.

[0073] For example, Table 1 shows a feature matrix example that illustrates the feature parameters of a permutation contract at three consecutive historical moments (T1 to T3).

[0074] Table 1

[0075] S103. Perform feature dimensionality reduction on the feature matrix to obtain the feature low-dimensional matrix of the target permutation contract.

[0076] A low-dimensional feature matrix refers to a matrix obtained by mapping the original high-dimensional feature matrix to a new space with significantly reduced dimensionality through mathematical transformation. The number of rows remains unchanged, still representing historical moments, but the number of columns is reduced, representing new, unrelated composite features. Feature dimensionality reduction serves two main purposes: first, to eliminate multicollinearity among the original features and reduce noise interference; second, to significantly reduce the data complexity of subsequent model calculations, improve processing efficiency, and avoid the "curse of dimensionality," thus helping the model better learn the essential patterns in the data.

[0077] In some embodiments, an autoencoder can be used for nonlinear feature dimensionality reduction. This method compresses the high-dimensional input features into a low-dimensional bottleneck layer using an encoder neural network; the output of this bottleneck layer is the dimensionality-reduced feature representation. During training, the autoencoder learns an efficient representation of the data by minimizing the reconstruction error, capturing complex nonlinear relationships in the original features, and is particularly effective for highly nonlinear permutation contract data.

[0078] In other embodiments, principal component analysis (PCA) is used for linear feature dimensionality reduction. This method first calculates the covariance matrix of the standardized feature matrix, then solves for the eigenvalues ​​and eigenvectors of this covariance matrix. Next, the eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the N largest eigenvalues ​​are selected to form a projection matrix. Finally, the original feature matrix is ​​projected onto the subspace spanned by these eigenvectors to obtain the dimensionality-reduced low-dimensional feature matrix. For a detailed description of this embodiment, please refer to the following text. Figure 3 As mentioned above, we will not go into details here.

[0079] It should be understood that the low-dimensional feature matrix obtained after dimensionality reduction has each new composite feature (i.e., principal component) that is a linear combination of the original features. By analyzing the weight coefficients of the original features in this linear combination, these composite features can be interpreted in terms of business implications. For example, the first principal component may have a high load on "value change rate" and "development trend indicators," which can be interpreted as a "value growth factor"; the second principal component may have a high load on "depreciation rate" and "performance score," which can be interpreted as a "state decay factor." This interpretability helps to understand the fundamental drivers of changes in value discount factors.

[0080] S104. Process the feature low-dimensional matrix through a preset coefficient model to obtain the target parameter values ​​of the target permutation contract.

[0081] The target parameter value is the value conversion factor of the target object as stipulated in the target replacement contract.

[0082] Among them, the preset coefficient model has the function of determining the parameter values ​​of the objects agreed upon in the permutation contract based on the feature low-dimensional matrix.

[0083] The preset coefficient model is a machine learning model trained on time-series data. It learns from a large amount of historical swap contract data to establish a mapping relationship between time-series features and value conversion coefficients. The model receives a low-dimensional feature matrix as input, processes it through its internal computational logic, and outputs the corresponding target value conversion coefficient.

[0084] In some embodiments, the model employs an LSTM architecture. This network iteratively processes the data at each time step in the low-dimensional feature matrix through its unique gating mechanism (including input gate, forget gate, and output gate), effectively capturing long-term dependencies in temporal features. The hidden state at the last time step undergoes a non-linear transformation through a fully connected layer, ultimately outputting the target value discount factor. This architecture is particularly suitable for handling the long-term dependency problem prevalent in permutation contracts.

[0085] In other embodiments, the model may employ an encoder-decoder architecture based on an attention mechanism. The encoder encodes the low-dimensional feature matrix of the input to extract temporal features; the attention mechanism calculates the importance weights of features at different time steps to the output; and the decoder generates the final target value conversion coefficient based on the weighted temporal features. This architecture can dynamically focus on the key time points that have the greatest impact on value conversion, improving the interpretability of the model.

[0086] In some other embodiments, the model may employ a temporal convolutional network. Multiple layers of causal convolutions capture local temporal patterns in the input sequence, while dilated convolutions increase the receptive field to capture long-term dependencies. Finally, a global pooling layer integrates information from all time steps to output the target value discount factor. This architecture offers advantages such as high parallel computation efficiency and stable training.

[0087] This application does not impose any restrictions on the specific implementation architecture of the preset coefficient model.

[0088] In one possible implementation, the pre-defined coefficient model determines the value conversion factor through the following process: First, the model receives a low-dimensional feature matrix as input, where each column represents the time series of a principal component feature. The model continuously tracks the dynamic changes of each principal component feature using a sliding time window mechanism. Specifically, for each principal component feature, the model calculates its trend characteristics, fluctuation characteristics, and its correlation with other principal components.

[0089] Next, the model constructs a comprehensive evaluation index based on these dynamic features. When significant changes are detected in multiple principal component features, the model initiates a risk reassessment mechanism, making a comprehensive judgment based on the current state of each feature to determine the direction and magnitude of the adjustment.

[0090] Then, the model dynamically adjusts the contribution of each principal component to the final result through a feature importance weighting mechanism. This dynamic weight allocation ensures that the model can flexibly adjust the evaluation focus according to different stages of the contract.

[0091] Finally, the model fuses the weighted principal component features and generates the final value conversion coefficient through a nonlinear transformation. This process fully considers the current state, trends, and interactions of each principal component feature, ensuring that the output conversion coefficient accurately reflects the real-time risk status of the contract. In particular, when certain principal component features exhibit abnormal fluctuations, the model initiates sensitivity analysis to assess the potential impact of these fluctuations on the conversion coefficient, ensuring the robustness of the output results.

[0092] This application provides a method for determining coefficient values. By acquiring multi-dimensional feature parameters of the target exchange contract at multiple historical moments, a time-series feature matrix is ​​constructed. The feature parameter system encompasses key influencing factors such as contract attributes, resource characteristics, owner status, domain dynamics, and default handling strategies, comprehensively reflecting the dynamic characteristics of the contract. Based on this, a low-dimensional feature matrix of the target exchange contract is obtained by dimensionality reduction of the feature matrix, effectively extracting the feature combinations most influential on the value conversion coefficient. Compared with existing technologies that directly calculate the conversion ratio based solely on resource value, this method analyzes the comprehensive impact of the time-series changes of multi-dimensional feature parameters on the value conversion coefficient. It utilizes a preset coefficient model to mine deep time-series patterns from the low-dimensional feature matrix, effectively overcoming the evaluation bias caused by traditional static calculation methods when value fluctuates due to the time span of the resource exchange process. This allows for accurate evaluation of the value conversion coefficient in a dynamically changing environment, significantly improving the accuracy and applicability of the conversion ratio in practical applications.

[0093] In some embodiments, the process of performing feature dimensionality reduction on the feature matrix using the principal component analysis method in S103 is as follows: Figure 3 As shown, S103 specifically includes the following steps: S201. Normalize the feature parameter set of each column in the feature matrix to obtain the processed feature matrix.

[0094] Normalization is a data preprocessing technique designed to eliminate biases caused by differences in the dimensions and numerical ranges of different characteristic parameters.

[0095] In some embodiments, the normalization method for each column of the feature parameter set in the feature matrix is ​​as follows: the Z-score normalization method is used to process each column of the feature matrix (i.e., the value of the same feature at different times) independently.

[0096] For example, a computing device can normalize the set of feature parameters in each column of the feature matrix using the following formula:

[0097] in, It is the first The characteristic parameter at the th ... The original value at a historical moment, It is the first The average value of each feature parameter across all historical time points. It is the first The labeling difference of each feature parameter across all historical moments.

[0098] For example, regarding the "historical value change rate" (assuming it is the j=2th feature) in the "feature parameters of the target object", its original value sequence for 5 consecutive historical moments (i=1 to 5) is: [0.02, 0.015, -0.005, 0.008, 0.012].

[0099] First, the average value of this feature parameter is calculated as: (0.02 + 0.015 - 0.005 + 0.008 + 0.012) / 5 = 0.01.

[0100] Secondly, calculate the standard deviation of this characteristic parameter: sqrt([(0.02-0.01)² + (0.015-0.01)² + (-0.005-0.01)² + (0.008-0.01)² + (0.012-0.01)²] / 5) ≈ 0.00806 Finally, the standardized value for each historical moment is calculated according to formula (1): (0.02 - 0.01) / 0.00806 ≈ 1.241, (0.015 - 0.01) / 0.00806 ≈ 0.620, (-0.005 - 0.01) / 0.00806 ≈ -1.861, (0.008 - 0.01) / 0.00806 ≈ -0.248, (0.012 - 0.01) / 0.00806 ≈ 0.248. Through this processing, the original historical value change rate sequence [0.02, 0.015, -0.005, 0.008, 0.012] is converted into a standardized sequence [1.241, 0.620, -1.861, -0.248, 0.248], eliminating the influence of dimensions and facilitating subsequent feature dimensionality reduction processing.

[0101] It should be understood that this processing transforms the distribution of all features into a standard normal distribution with a mean of 0 and a standard deviation of 1. This allows features with vastly different dimensions, such as "disposal time" and "cumulative recovery amount," to be compared and calculated on the same scale. This ensures that in subsequent analysis, the contribution of each feature to the results is determined solely by its inherent variation pattern, rather than by its original numerical value, thus laying a solid foundation for extracting truly representative principal components.

[0102] S202. Perform covariance processing on the processed feature matrix and use the resulting covariance matrix as the correlation coefficient matrix.

[0103] Covariance is a statistic used to measure the similarity of the changing trends of two characteristic parameters. Specifically, covariance determines their correlation by calculating the degree to which each characteristic parameter deviates from its mean. When the two characteristic parameters change in the same direction, the covariance is positive; when they change in opposite directions, the covariance is negative; and when the changes of the two characteristic parameters lack correlation, the covariance approaches zero.

[0104] Covariance processing refers to the systematic calculation of the covariance values ​​between all pairwise features in a standardized feature matrix by a computing device. For a feature matrix containing n feature parameters, covariance processing will generate an n×n square matrix. The diagonal elements of this matrix are the covariances of each feature parameter with itself, and the off-diagonal elements are the covariances between different feature parameters.

[0105] The correlation coefficient matrix is ​​a standardized covariance matrix. Because the original covariance values ​​are affected by the dimensions of the feature parameters themselves, it is difficult to directly compare the correlation strength between different feature pairs. The correlation coefficient, through mathematical transformation, standardizes the covariance to the [-1, 1] interval, making the correlation between different feature pairs comparable.

[0106] The reason why the covariance processing result can be used as the correlation coefficient matrix is ​​that in S201, the computing device has standardized the feature matrix so that the mean of each feature parameter is zero and the standard deviation is one. Under this specific condition, the value range of the covariance naturally falls within the interval [-1, 1], and its value directly reflects the degree of linear correlation between the feature parameters. Therefore, it can be directly used as the correlation coefficient.

[0107] In some embodiments, the covariance processing is performed by the computing device calculating the covariance of all feature parameter pairs in batches using matrix operations. Specifically, the covariance matrix is ​​obtained by multiplying the standardized feature matrix by its transpose and then dividing by the number of samples minus one. This matrix operation method is more computationally efficient than pairwise calculation and is particularly suitable for handling situations involving a large number of feature parameters and sample data.

[0108] For example, the process can be represented as:

[0109] in, It is the processed feature matrix. yes The transpose of the matrix, Represents the sample degrees of freedom. This represents the final correlation coefficient matrix.

[0110] For example, suppose the standardized feature matrix Z contains 3 historical moments and 3 feature parameters with the following values: first row 0.8, 0.2, -0.5, second row 0.1, -0.3, 0.6, and third row -0.9, 0.1, -0.1.

[0111] First, calculate the transpose matrix of Z: the first row is 0.8, 0.1, -0.9, the second row is 0.2, -0.3, 0.1, and the third row is -0.5, 0.6, -0.1.

[0112] Then calculate the transpose of Z multiplied by Z: the first row of the new matrix is ​​1.46, -0.01, -0.25, the second row is -0.01, 0.14, -0.23, and the third row is -0.25, -0.23, 0.62.

[0113] Finally, calculate the correlation coefficient matrix R: Divide each element of the above matrix by 2 (because m-1=2) to get the first row of the R matrix as 0.73, -0.005, -0.125, the second row as -0.005, 0.07, -0.115, and the third row as -0.125, -0.115, 0.31.

[0114] The correlation coefficient matrix R shows that the diagonal elements are the variances of each feature, and the off-diagonal elements are the correlation coefficients between features. For example, -0.005 in the first row and second column indicates a very weak negative correlation between feature one and feature two.

[0115] S203. Substitute the correlation coefficient matrix into the characteristic equation to solve for multiple eigenvalues ​​of the correlation coefficient matrix and the eigenvector corresponding to each eigenvalue.

[0116] The characteristic equation is used to find the main direction of change in the correlation coefficient matrix. The computing device obtains multiple eigenvalues ​​by solving the characteristic equation. Since the dimension of the correlation coefficient matrix determines the number of solutions to the characteristic equation, for a correlation coefficient matrix containing n characteristic parameters, solving the characteristic equation yields n eigenvalues. The magnitude of each eigenvalue reflects the importance of the original data in the corresponding direction of change; a larger eigenvalue indicates richer original information contained in that direction.

[0117] For example, it can be represented as a characteristic equation:

[0118] in, Represents eigenvalues, It is an identity matrix, with diagonal elements all being 1 and off-diagonal elements all being 0. It is the correlation coefficient matrix. The identity matrix is ​​a square matrix with all elements on its main diagonal being 1 and all other elements being 0.

[0119] For example, suppose the correlation coefficient matrix R is: first row 0.73, -0.005, -0.125, second row -0.005, 0.07, -0.115, third row -0.125, -0.115, 0.31.

[0120] The computing device substitutes it into formula (3), and after expansion, obtains a cubic equation in λ: The equation was solved using numerical methods, yielding three eigenvalues: 0.85, 0.22, and 0.04.

[0121] Next, the computing device calculates the eigenvector corresponding to each eigenvalue: For 0.85, the equation (0.85ER)v=0 is solved. The coefficient matrix is ​​as follows: first row: -0.12, 0.005, 0.125; second row: 0.005, 0.78, 0.115; third row: 0.125, 0.115, 0.54. This yields the first component of eigenvector v1 as 0.65, the second component as -0.42, and the third component as 0.63. Similarly, for 0.22, the eigenvector v2 is found to have the first component as 0.53, the second component as 0.75, and the third component as -0.39. For 0.04, the eigenvector v3 has the first component as 0.54, the second component as -0.51, and the third component as -0.67.

[0122] Finally, the computing device solves the above system of linear equations to obtain the eigenvectors corresponding to each eigenvalue: eigenvalue 0.85 corresponds to eigenvector [0.65, -0.42, 0.63], eigenvalue 0.22 corresponds to eigenvector [0.53, 0.75, -0.39], and eigenvalue 0.04 corresponds to eigenvector [0.54, -0.51, -0.67]. These eigenvectors are orthogonal to each other, and their corresponding eigenvalues ​​are arranged from largest to smallest, representing the three principal directions with the largest variance in the original data.

[0123] S204. Project the processed feature matrix onto the subspace formed by the eigenvectors corresponding to the first eigenvalues ​​to obtain the feature low-dimensional matrix.

[0124] The first feature value refers to the feature value whose sum of the contribution rates of the first N selected features is greater than the preset feature contribution rate threshold after the feature values ​​are arranged in descending order of feature contribution rate, where N is a positive integer.

[0125] The core of this step is to transform the standardized feature matrix from the original high-dimensional space to a new coordinate system composed of the top N most important eigenvectors. Each eigenvector represents a major direction of data variation, and the magnitude of its corresponding eigenvalue characterizes the importance of that direction. Through the projection operation, the core variation patterns of the original data are preserved in the low-dimensional space, while noise and redundant information are effectively removed.

[0126] The feature contribution rate is an indicator that measures the importance of each principal component, specifically defined as the ratio of a single eigenvalue to the sum of all eigenvalues. A higher feature contribution rate indicates that the principal component contains more original information. The cumulative feature contribution rate is the sum of the contribution rates of the first k features, reflecting the degree to which the corresponding principal component subspace retains original information.

[0127] In some embodiments, the computing device obtains the feature contribution rate by the following steps: first, calculating the sum of all feature values, and then dividing each feature value by the sum to obtain the contribution rate of each feature value.

[0128] The preset feature contribution rate threshold represents the amount of original information that the low-dimensional feature matrix after dimensionality reduction needs to retain.

[0129] For example, when the preset feature contribution rate threshold is set to 85%, the computing device will accumulate the feature contribution rate from small to large until the cumulative value exceeds 85%. The number of feature values ​​selected at this point is the N value. In addition, the preset feature contribution rate threshold can also be set to 80%, 90%, etc. The specific value of the preset feature contribution rate threshold is not limited in this embodiment.

[0130] One possible implementation: The computing device first calculates and sorts the contribution rates of all eigenvalues, then accumulates the contribution rates until they exceed a preset threshold, determining the number N principal components to be retained. Next, the eigenvectors corresponding to the top N eigenvalues ​​are selected to form a projection matrix P. Finally, the standardized feature matrix Z is multiplied by the projection matrix P to obtain a low-dimensional feature matrix Z' = Z × P. This matrix retains the same number of rows but reduces the number of columns to N, thus preserving the main information while achieving dimensionality reduction.

[0131] Furthermore, when selecting the N value, the computing device also considers the actual distribution of the feature values. When there is a significant discontinuity in the feature values, the N value may be determined at the discontinuity even if the preset threshold is not reached. Simultaneously, the new features obtained after projection are linear combinations of the original features; these new features are uncorrelated, providing better input conditions for subsequent modeling.

[0132] It should be understood that by performing feature dimensionality reduction on the feature matrix, redundant information between features can be effectively eliminated, and the feature combinations that have the greatest impact on the value conversion factor can be extracted. This process not only reduces computational complexity but also improves the quality of the feature data, allowing subsequent models to focus more on learning key features, thereby improving the accuracy and stability of the model.

[0133] In some embodiments, the target parameter value includes a maximum value conversion factor and a recommended value conversion factor. The maximum value conversion factor is used as the upper limit for adjusting the value conversion factor of the target swap contract, and the recommended value conversion factor is used as a reference for adjusting the value conversion factor of the target swap contract.

[0134] The maximum value conversion factor represents the upper limit of parameter adjustment dynamically calculated based on the current time-series state characteristics under the target replacement resource type. This factor provides an important stability reference benchmark for setting the value conversion factor, and its calculation comprehensively considers the characteristic fluctuations of the replacement resource, changes in the status of the owner, and changes in environmental factors. When the actual value conversion factor used exceeds this benchmark, the computing device will identify it as a parameter state that requires special attention and initiate the corresponding analysis process.

[0135] The recommended value conversion factor is an optimized reference value derived from a comprehensive analysis of multi-dimensional parameters, including the quantitative relationship between the replacement resources and the preset resources, the characteristics of the contracting parties, and environmental factors. This factor reflects the balance point between system stability and efficiency calculated by the model under the current time-series environment, providing data support for optimizing and adjusting the value conversion factor. Its core function is to provide a refined reference for contract management; when the actual value conversion factor deviates from this reference value, the computing device will generate corresponding optimization suggestions.

[0136] In this case, the computing device can also perform real-time evaluation and adjustment of the value conversion factor of the target replacement contract based on the target value conversion factor output by the model. For example... Figure 4 As shown, after S104, the following steps are specifically included: S301. Obtain the valuation factor for the target replacement contract.

[0137] The valuation factor to be evaluated is the valuation factor of the target replacement contract determined at a historical point in time.

[0138] The valuation factor refers to the target replacement contract generated before the current valuation point in time.

[0139] In some embodiments, the valuation factor can specifically be: an initial valuation factor set by the reviewer during the target replacement contract approval stage, an execution valuation factor updated according to the real-time status during the preset resource distribution stage, and an actual valuation factor calculated based on the recovery status during the resource recovery stage.

[0140] The approval stage refers to the period during which key terms are reviewed and confirmed during the contract signing process; the pre-determined resource disbursement stage refers to the period during which resources are allocated to the recipient in accordance with the contract; and the resource recovery stage refers to the period during which assets are liquidated when the contract is terminated or in breach.

[0141] In some embodiments, the computing device obtains the coefficient by querying a contract management database or by receiving manual input. This coefficient is typically associated with a specific effective timestamp to ensure that comparative analyses are performed on the same time-series benchmark.

[0142] S302. If the conversion factor of the value to be assessed is greater than the highest conversion factor, generate the first assessment result of the conversion factor of the value to be assessed.

[0143] The first assessment result is used to indicate the adjustment factor for the value to be assessed.

[0144] In one possible implementation, the computing device generates the first evaluation result through the following process: First, it calculates the relative deviation between the coefficient to be evaluated and the highest coefficient, using the formula (coefficient to be evaluated - highest coefficient) / highest coefficient × 100%; second, it determines the handling priority based on the deviation range, with deviations within 5% being medium priority, 5%-10% being high priority, and above 10% being the highest priority; then, it generates differentiated handling plans based on the priorities, including measures such as immediately suspending contract execution, adjusting coefficients within a specified period, or strengthening monitoring; finally, it generates a complete evaluation report containing deviation analysis, priority assessment, and handling recommendations.

[0145] For example, the specific form of the first assessment result is as follows: when it is detected that the valuation factor of a certain inventory pledge contract is 0.78, which exceeds the highest valuation factor of 0.70, an assessment report is generated stating that "the factor deviation is 11.4%, which is the highest priority. It is recommended to adjust the valuation factor to below 0.70 within 8 hours and suspend the issuance of new quotas before the adjustment is completed".

[0146] In some embodiments, the computing device also initiates a tracking mechanism to periodically check the adjustment progress until the coefficients to be evaluated return to a reasonable range. Simultaneously, the system records relevant data about this anomaly, including the duration of the anomaly, the adjustment process, and the final result, for subsequent analysis and model optimization.

[0147] S303. If the conversion factor of the value to be assessed is less than the highest conversion factor but greater than the recommended conversion factor, a second assessment result of the conversion factor of the value to be assessed is generated.

[0148] The second assessment result is used to recommend adjustments to the conversion factor of the value to be assessed.

[0149] In one possible implementation, the computing device generates the second evaluation result through the following process: First, it analyzes the specific reasons why the current coefficient deviates from the recommended coefficient, including influencing factors such as fluctuations in resource value and changes in the status of the owner; second, it assesses the potential impact of maintaining the status quo, including the changing trends of resource utilization efficiency and contract stability; then, it provides multiple adjustment options, including options such as immediate adjustment, phased adjustment, and continued observation; finally, it provides an analysis of the expected effects of each option and implementation suggestions.

[0150] For example, the specific form of the second assessment result is as follows: when the conversion factor of a certain equipment leasing contract is 0.68, which is between the recommended factor of 0.65 and the maximum factor of 0.70, an assessment report is generated stating that "the current factor deviates from the recommended value by 4.4%, and it is recommended to adjust it to the range of 0.65-0.67 before the next billing cycle, which is expected to improve resource utilization efficiency by about 6% while maintaining contract stability."

[0151] In some embodiments, the computing device can also comprehensively analyze the evaluation scenarios in S302 and S303 using a preset large language model. For the out-of-limit situation in S302, the large language model can generate detailed anomaly analysis reports and handling guidance based on historical handling cases and industry knowledge; for the optimization scenario in S303, the large language model can provide multi-faceted optimization solutions by combining contract characteristics and environmental factors. Through natural language generation technology, the large language model transforms complex parameter relationships and evaluation logic into easily understandable business descriptions while maintaining consistency with structured data, thereby providing users with more operable and interpretable comprehensive evaluation results.

[0152] It should be understood that this scheme achieves tiered early warning and dynamic adjustment of the value conversion factor by setting a dual evaluation benchmark of a maximum value conversion factor and a recommended value conversion factor, and establishing a corresponding evaluation result generation mechanism. Compared with a single fixed ratio, this tiered evaluation mechanism can provide more refined security control, ensuring both the safety of the replacement process and maintaining reasonable flexibility.

[0153] The following is a detailed explanation of the preset coefficient model mentioned above.

[0154] In some embodiments, the computing device can acquire and deploy a pre-trained coefficient model from an external device to obtain a preset coefficient model.

[0155] For example, a computing device downloads a pre-trained coefficient model and its configuration file from a model server. This model contains a fixed network structure and optimized parameter weights, which can be used directly for inference tasks.

[0156] In other embodiments, the computing device can also train an initial model of a preset coefficient model locally using training data to obtain a preset coefficient model. In this case, the specific process of training the preset coefficient model is as follows: Figure 5 As shown, it specifically includes the following: S401. Obtain the sample time series feature data corresponding to each of the multiple sample permutation contracts within a preset historical time period.

[0157] The sample time-series feature data includes a set of feature parameters for each sample replacement contract at multiple historical moments. Each set of feature parameters includes the attribute feature parameters of the sample contract, the feature parameters of the replacement resource, the feature parameters of the resource owner, the feature parameters of the domain, and the feature parameters of default handling.

[0158] For an introduction to the sample time-series feature data, please refer to the relevant content in S101, which will not be elaborated here.

[0159] The preset historical duration can be configured according to the actual application scenario, such as being set to 10 years, 20 years or 30 years, etc., and this application embodiment does not limit this.

[0160] In addition, the sample time-series feature data includes a set of feature parameters for each sample replacement contract at multiple historical moments within its complete lifespan or a preset observation window. The time-series length (i.e., the number of historical moments) of each sample contract can be variable, which truly reflects the different business characteristics of different contract lifespans.

[0161] S402. Determine the feature dimensionality reduction matrix corresponding to each sample's time-series feature data.

[0162] This step involves constructing a feature matrix and a low-dimensional feature matrix for each sample's time-series feature data. Refer to steps S102 (constructing the feature matrix) and S201-204 (constructing the low-dimensional feature matrix) above. Further details are omitted here.

[0163] By reducing the dimensionality of the high-dimensional temporal features of each sample contract, a low-dimensional feature matrix is ​​obtained, achieving data standardization and redundancy removal. This provides a unified and clean input format for subsequent models, ensuring that the model can focus on learning the most informative feature patterns, while significantly reducing the computational complexity of the model and improving training efficiency. The low-dimensional feature matrices of all sample contracts will be used together as the training input set for the initial model of the preset coefficient model.

[0164] In addition, this process ensures that the patterns learned by the model during the training phase are consistent with the data processing flow faced in subsequent practical applications (inference phase), thus guaranteeing the effectiveness and reliability of the model's prediction results.

[0165] S403. Based on the credibility identifier of the sample object belonging to each sample replacement contract, construct a value conversion coefficient label matrix that corresponds one-to-one with the temporal feature matrix of each sample.

[0166] In this model, each row of the temporal feature matrix corresponds to each row of the corresponding value discount coefficient label matrix. Specifically, for each set of feature parameters in the feature matrix at each historical moment, a corresponding target value for the value discount coefficient is set in the same row of the label matrix. This row-column aligned structure ensures that the model can learn the accurate mapping relationship from the feature state at a specific moment to the corresponding value discount coefficient.

[0167] A credibility rating is a reliability level assessment based on the historical credit records of the party involved. It is used to quantitatively measure the party's willingness and ability to fulfill its contractual obligations. This credibility rating is usually a discrete variable, such as a binary category of "credible" and "unreliable," or a multi-level category such as "excellent / good / average / poor."

[0168] A label matrix is ​​a two-dimensional data structure containing the values ​​of the supervised learning target. In this application, each element of the matrix represents a reference value, such as the value reduction factor, that a corresponding temporal feature should have at a specific historical moment. The label matrix matches the feature matrix in dimensionality, providing a standard answer for model training.

[0169] Different value conversion factor determination rules apply to the parties associated with different credibility identifiers. This means the calculation logic and stability adjustment parameters for the value conversion factor also differ, reflecting the differentiated measurement principle for different levels of performance guarantee during resource exchange. Triggering different label generation strategies through credibility identifiers ensures that training data conforms to business logic, enabling the model to learn a reasonable balance between stability and efficiency.

[0170] For detailed implementation information of this step, please refer to the following text. Figure 5 or Figure 7 This will not be discussed in detail here.

[0171] S404. Based on the low-dimensional matrix of each feature and its corresponding value conversion coefficient label matrix, train the initial model of the preset coefficient model to obtain the preset coefficient model.

[0172] That is, each feature low-dimensional matrix and its corresponding value conversion coefficient label matrix can be regarded as a training sample, and multiple training samples constitute the training sample set for training the initial model of the preset coefficient model.

[0173] For example, each feature low-dimensional matrix and its corresponding value conversion factor label matrix can be represented as: a feature low-dimensional matrix of dimension [T, M] and a label matrix of dimension [T, 2], where T represents the time step and M represents the number of features after dimensionality reduction. During training, the computing device groups N such training samples into a batch, forming an input tensor of [N, T, M] and a label tensor of [N, T, 2], which are then input into the LSTM for batch training. The hidden layer of the LSTM network processes the input at each time step through 64 hidden units, and finally maps the temporal features into two output values ​​through a fully connected layer, corresponding to the highest value conversion factor and the recommended value conversion factor, respectively.

[0174] In some embodiments, the initial model of the preset coefficient model is an LSTM. The training process iteratively optimizes the model parameters through multiple rounds: in each iteration, the model receives a preset number of low-dimensional feature matrices as input, and outputs a predicted value-reduced coefficient sequence through the collaborative processing of the forget gate, input gate, and output gate; then, the predicted sequence is compared with the value-reduced coefficient label matrix, the loss function is calculated, and the parameters of the gating units are adjusted through the backpropagation algorithm. During training, the computing device divides the sample dataset into a training set and a validation set, where the validation set is used to monitor the model's generalization ability. Through continuous iteration, the model significantly improves the recognition accuracy of feature fluctuation patterns and temporal correlations by adjusting hyperparameters such as the learning rate and batch size, and optimizing gating parameters such as the forget gate threshold and input gate weights. Finally, when the model's prediction accuracy on the validation set reaches a preset standard, the training process terminates, resulting in a preset coefficient model with accurate temporal feature parsing capabilities.

[0175] For example, such as Figure 6 The diagram shows the structure of an LSTM model, which includes an input layer, a hidden layer, a fully connected layer, and an output layer.

[0176] The input layer receives a low-dimensional feature matrix, sequentially feeding the feature values ​​of each row into the hidden layer. The hidden layer comprises multiple LSTM units, each processing the input for one time step. For example, the first unit processes input x1 and produces output y1, the second unit receives x2 and the hidden state from the previous unit and produces output y2, and so on. By connecting multiple LSTM units sequentially over time, the model can progressively extract and memorize long-term dependencies between rows and utilize internal control mechanisms to filter and transmit important information. The hidden layer ultimately outputs 64 features, representing the 64-dimensional hidden state features extracted at the last time step. A fully connected layer performs a non-linear transformation on these 64 features, mapping them to the target output dimension through a weight matrix. The final output layer includes two independent output heads, generating the highest value discount factor and the recommended value discount factor, respectively, providing a complete value discount reference for subsequent contract evaluation.

[0177] Furthermore, a multi-temporal granularity joint training framework can be introduced during LSTM training. This framework enables the model to simultaneously process feature sequences sampled at different frequencies such as daily, weekly, and monthly, and automatically learns the importance weights of features at each granularity through a gating mechanism. This design allows the model to simultaneously capture short-term abnormal fluctuations and long-term trend changes, significantly improving the prediction accuracy of value discount factors in complex market environments.

[0178] In other embodiments, the initial model of the preset coefficient model is a temporal convolutional network. The training process iteratively optimizes the model parameters through multiple rounds: in each iteration, the model receives a preset number of low-dimensional feature matrices as input, processes them through multiple layers of causal convolution and dilated convolution, and outputs a sequence of predicted value-reduced coefficients; then, the predicted sequence is compared with the value-reduced coefficient label matrix, the loss function is calculated, and the convolution kernel parameters are adjusted through backpropagation. During training, the computing device divides the sample dataset into a training set and a validation set, where the validation set is used to evaluate the model's temporal feature extraction capability. Through continuous iteration, the model significantly improves its ability to capture multi-scale temporal patterns by adjusting hyperparameters such as dilation rate and convolution kernel size, and optimizing the weight parameters of each convolutional layer. Finally, when the model's prediction accuracy on the validation set reaches a preset standard, the training process terminates, resulting in a preset coefficient model with efficient temporal feature extraction capabilities.

[0179] Furthermore, during the training of temporal convolutional network models, computing devices can incorporate a permutation resource sensitivity weight allocation mechanism. This mechanism dynamically adjusts the receptive field of the convolutional kernel at different time scales based on the liquidity characteristics of different types of permutation resources (such as equipment, inventory, and computing power). For example, for inventory resources with frequent value fluctuations, the model assigns higher weights to short-term features; while for equipment resources with longer depreciation periods, it focuses more on long-term trend features. This targeted design allows the model to better adapt to the characteristics of different types of permutation resources, significantly improving the accuracy of value discount factor prediction.

[0180] In some embodiments, the initial model of the preset coefficient model is an encoder-decoder architecture based on an attention mechanism. The training process iteratively optimizes the model parameters through multiple iterations: in each iteration, the model receives a preset number of low-dimensional feature matrices as input; the encoder encodes the sequence into a hidden representation; the attention mechanism calculates the weight distribution at different time steps; and the decoder generates a predicted value discount coefficient sequence based on a weighted context vector. Subsequently, the predicted sequence is compared with the value discount coefficient label matrix, the loss function is calculated, and the attention weights and network parameters are adjusted using a backpropagation algorithm. During training, the computing device divides the sample dataset into a training set and a validation set, where the validation set is used to monitor the model's ability to focus on key features. Through continuous iteration, the model significantly improves its ability to identify features at key time steps by adjusting hyperparameters such as the number of attention heads and the dimension of hidden layers, and optimizing the attention distribution mechanism. Finally, when the model's prediction accuracy on the validation set reaches a preset standard, the training process terminates, resulting in a preset coefficient model with accurate feature selection capabilities.

[0181] In some embodiments, during the training process of the different model architectures described above, the computing device also needs to set iteration stopping conditions, that is, the termination criteria for the model training process, including: performance convergence conditions, iteration number thresholds, early stopping protection mechanisms, and time budget constraints.

[0182] Specifically, the performance convergence condition refers to the criterion for determining when the model's evaluation metrics on the validation set no longer show significant improvement. This condition is achieved by monitoring the trend of the loss function on the validation set. When the improvement of the loss function value over multiple consecutive training epochs (e.g., 10 epochs) is less than a preset threshold (e.g., 0.1%), the model is considered to have reached convergence.

[0183] The iteration threshold refers to the preset maximum number of training epochs. This condition serves as a fundamental guarantee for the training process, ensuring that training does not proceed indefinitely. When the number of training epochs reaches the preset maximum value (e.g., 1000 epochs), the computing device will forcibly terminate the training process regardless of model performance.

[0184] The early termination protection mechanism refers to a strategy that terminates training prematurely based on the decline in validation set performance. This mechanism is implemented by continuously monitoring the validation set performance. When the validation set loss increases significantly (e.g., more than 5%) over several consecutive periods (e.g., 5 periods), it indicates that the model may be overfitting. The computing device will immediately stop training and automatically roll back to the model parameter state where the validation set performance is optimal.

[0185] Time budget constraints refer to the duration limits of the training process. This condition is mainly based on engineering implementation considerations. When the total training time exceeds the preset time limit (such as 24 hours), in order to ensure project progress and the rational use of computing resources, the computing device will automatically save the current optimal model and terminate the training process.

[0186] It should be understood that the embodiments of this application realize end-to-end automated construction from raw data to a usable model. The process first obtains sufficient historical sample data in S401 to ensure the representativeness and diversity of the training set; then, in S402, it performs dimensionality reduction on the original features, eliminating redundancy while retaining key information; next, in S403, it constructs a label matrix based on business rules, providing accurate target guidance for supervised learning; finally, in S404, it uses an optimized training strategy to perform targeted training on different model architectures. This organic combination of steps not only ensures that the final model with preset coefficients possesses accurate predictive capabilities but also improves the model's adaptability and stability in complex business scenarios through an innovative training mechanism, providing reliable technical support for determining the value conversion coefficient in the resource exchange process.

[0187] In some embodiments, when constructing the value conversion coefficient label matrix corresponding to each sample's time-series feature matrix, the computing device employs a differentiated label generation strategy based on credibility identifiers to ensure that the label data conforms to business logic. This process is as follows: Figure 7 As shown, S403 may include the following steps: S501. Perform the first operation: If the credibility identifier of the party to which the sample object belongs corresponding to the first sample replacement contract is credible, determine the recommendation value conversion coefficient label of the first sample replacement contract at the first historical moment based on the ratio of the quantitative value of the target object of the first sample replacement contract at the first historical moment to the quantitative value of the preset resource.

[0188] For parties identified as trustworthy, the certainty of fulfilling their replacement contracts is relatively high. Therefore, the determination of the recommended value conversion factor can focus more on reflecting the intrinsic value correspondence between the replacement resources and the pre-set resources. By calculating the ratio of the quantified value of the target object to the quantified value of the pre-set resources, the intrinsic equivalence of the two types of resources at the first historical moment can be directly reflected. One possible implementation method for assessing the quantified value of a target object includes: for physical resource objects, using a comparative method; for virtual resource objects, quantifying them based on their service capabilities and use value; and converting the quantified value of pre-defined resources according to the standard unit of measurement agreed upon in the contract.

[0189] In some embodiments, the ratio of the quantitative value of the target object to the quantitative value of the preset resource based on the first sample replacement contract at the first historical moment can be understood as the minimum value in the ratio sequence formed by all time points within the time period from the previous historical moment to the first historical moment.

[0190] In this embodiment, the computing device determines the recommended value conversion factor label for the current first historical moment by iterating through and comparing the target object value / preset resource value ratios at all points in time within a certain time period (e.g., 30 days prior to the first historical moment) and selecting the minimum value among them. This method aims to capture the worst-case scenario of recent value relationships, providing the model with a dynamic and prudent learning objective.

[0191] S502, Perform the second operation: Use the preset highest value conversion factor corresponding to the target object type of the first sample replacement contract as the highest value conversion factor label of the first sample replacement contract at the first historical moment.

[0192] For parties identified as trustworthy, the certainty of fulfilling their replacement contracts is relatively high. Therefore, the determination of the maximum value conversion factor is primarily based on the value characteristics of the target object type itself, rather than the real-time credit status of the party. The preset maximum value conversion factor corresponding to the target object type of the first sample replacement contract is, for different types of target objects (such as equipment, inventory, intellectual property, etc.), a preset maximum value conversion factor is established based on their inherent attributes such as liquidity, value stability, and disposal difficulty.

[0193] In some embodiments, the computing device can query a pre-configured target object type-maximum conversion factor mapping table to obtain the corresponding preset maximum value conversion factor based on the target object type. This mapping table is established based on historical data analysis and business experience, setting a unified and standardized security boundary for various target objects, ensuring that the model always adheres to basic security principles while pursuing efficiency.

[0194] It should be noted that the "first operation" and "second operation" in the above steps are executed in parallel, generating the recommended value conversion factor label and the highest value conversion factor label, respectively. This temporal traversal method allows for the construction of a complete label matrix for each sample contract, ensuring that each historical moment has a corresponding value conversion factor label, thus providing accurate supervision signals for subsequent model training.

[0195] S503, Perform the third operation, using the second historical moment as the new first historical moment.

[0196] The second historical moment is later than the first historical moment.

[0197] The computing device establishes a progressive temporal relationship by updating the second historical moment to the new first historical moment, thereby enabling traversal of all historical moments within the contract's lifespan. This iterative mechanism ensures that each historical moment can sequentially serve as the current processing node, receiving the same label generation logic processing.

[0198] S504. Repeat the first operation, the second operation, and the third operation until the first historical moment is empty. Obtain the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

[0199] In one possible implementation, the computing device implements this process through a loop control structure: during initialization, the contract start time is taken as the first historical moment. In each loop iteration, a first operation is performed to generate a recommended value conversion factor label, a second operation is performed to generate a highest value conversion factor label, and then a third operation updates the next moment to the new first historical moment. When all historical moments have been processed, the first historical moment becomes null, and the loop terminates. At this point, the computing device arranges the label values ​​generated by each historical moment in chronological order, forming a recommended value conversion factor label sequence and a highest value conversion factor label sequence, respectively.

[0200] Next, the computing device constructs a value conversion factor label matrix using the recommended value conversion factor label sequence and the highest value conversion factor label sequence. This involves aligning and merging the two label sequences along the time dimension to form a two-dimensional matrix structure. The rows of this matrix correspond to historical time points, and the columns correspond to different types of value conversion factor labels. The first column stores the recommended value conversion factor label sequence, and the second column stores the highest value conversion factor label sequence. This matrix structure maintains a strict correspondence between the temporal feature matrix and the label matrix along the row dimension, ensuring that the feature data at each historical time point can be mapped to the corresponding value conversion factor label, providing complete supervised learning data pairs for subsequent model training.

[0201] It should be understood that by using a label generation method based on the ratio of the target object to the pre-defined resource value for resource owners with credible identifiers, the decisive role of resource value in the conversion factor can be accurately reflected under stable performance conditions. This label construction mechanism enables the trained model to fully learn the inherent correlation between resource value and conversion factor under normal performance scenarios, thereby providing a value conversion assessment that matches the performance characteristics of replacement contracts with credible identifiers.

[0202] In other embodiments, when the credibility identifier of the party to which the sample object belongs corresponding to the first sample replacement contract is untrustworthy, such as Figure 8 As shown, S403 may also include the following steps: S601. Perform the fourth operation: If the credibility of the party to which the sample object belongs corresponding to the first sample replacement contract is identified as untrustworthy, use the historical value conversion factor of the first sample replacement contract as the highest value conversion factor label of the first sample replacement contract at the first historical moment.

[0203] The historical value conversion factor refers to the value conversion factor determined during the contract approval stage of the first sample replacement contract. Because the credibility of the party owning the sample object is marked as untrustworthy, a relatively prudent evaluation standard was adopted during the initial contract approval, representing the highest guarantee requirement that can be set for this type of contract. Using this historical value as the label of the highest value conversion factor for the first sample replacement contract at the first historical moment ensures that the model can follow the original guarantee standard for this type of scenario during the learning process, establishing a reliable guarantee benchmark for the prediction output.

[0204] S602. Perform the fifth operation: Based on the ratio of the cumulative quantified value of the preset resources recovered by the first sample replacement contract at the first historical moment to the quantified value of the preset resources in the first sample replacement contract, determine the recommended value conversion factor label of the first sample replacement contract at the first historical moment.

[0205] As mentioned above, the recommended value conversion coefficient is an optimized reference value derived from a comprehensive analysis of multi-dimensional parameters, including the quantitative relationship between the replacement resources and the preset resources, the characteristics of the contracting parties, and environmental factors. Therefore, for untrustworthy sample parties, the logic for determining the recommendation coefficient shifts from theoretical value matching to actual performance guarantee. By using the ratio of the cumulative recovered preset resources to the total preset resources as the label for the recommended value conversion coefficient, the adjustment principle based on actual recovery effect is reflected in low-confidence scenarios. This label construction method ensures that the model learns the decision-making rules for reasonable conversion under the premise of ensuring the smooth execution of the contract, providing a more practical value conversion reference for specific contract scenarios.

[0206] S603, Execute the sixth operation, using the second historical moment as the new first historical moment.

[0207] The second historical moment is later than the first historical moment.

[0208] This step can be referred to in S503 above, and will not be described in detail here.

[0209] S604. Repeat the fourth, fifth and sixth operations until the first historical moment is empty, and obtain the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

[0210] This step can be referred to in S504 above, and will not be described in detail here.

[0211] It should be understood that by employing a label generation method based on historical recycling data for resource owners with untrusted identifiers, the adjustment effect of resource recycling status on the conversion factor during actual performance can be effectively reflected. This differentiated label construction method enables the model to learn how to dynamically adjust the conversion factor based on actual recycling performance under different performance statuses, thereby providing a value assessment scheme adapted to the performance history of replacement contracts with different identifiers.

[0212] Figure 9 This is a schematic diagram of a coefficient value determination device provided in an embodiment of this application. Figure 9 As shown, the coefficient value determination device includes: an acquisition module 901 and a processing module 902.

[0213] The acquisition module 901 is used to acquire the time-series characteristic data of the target substitution contract. The target substitution contract is used to define the preset resources for converting the target object into the target value. The time-series characteristic data includes the characteristic parameter set of the target substitution contract at multiple historical moments. The characteristic parameter set at each historical moment includes: the attribute characteristic parameters of the target substitution contract, the characteristic parameters of the target object, the characteristic parameters of the party to which the target object belongs, the characteristic parameters of the domain to which the target object belongs, and the default handling strategy characteristic parameters of the target substitution contract.

[0214] Processing module 902 is used to construct a feature matrix of the target substitution contract based on its time-series feature data. Each row of the feature matrix represents the set of feature parameters of the target substitution contract at a single historical moment, and each column represents a feature parameter of the target substitution contract at different historical moments. The feature matrix is ​​then subjected to dimensionality reduction processing to obtain a low-dimensional feature matrix of the target substitution contract. This low-dimensional feature matrix is ​​then processed by a preset coefficient model to obtain the target parameter values ​​of the target substitution contract. These target parameter values ​​are the value conversion coefficients of the target object stipulated in the target substitution contract. The preset coefficient model has the function of determining the parameter values ​​of the object stipulated in the substitution contract based on the low-dimensional feature matrix.

[0215] In other embodiments, the attribute characteristics of the target replacement contract include: the duration of the target replacement contract, the purpose of the target replacement contract, and the third-party reserved resources of the target replacement contract. The characteristic parameters of the target object include: the type of the target object, the historical value change rate of the target object, and the depreciation rate of the target object. The characteristic parameters of the party to which the target object belongs include: the performance characteristic parameters of the party to which the target object belongs, and the credibility identifier of the party to which the target object belongs. The characteristic parameters of the field to which the target object belongs include: the development trend indicators of the field to which the target object belongs, and the liquidity indicators of the field to which the target object belongs. The characteristic parameters of the default handling strategy of the target replacement contract include: the type of default handling strategy, the default duration of the target replacement contract, and the preset resources for default recovery of the target replacement contract.

[0216] In other embodiments, the processing module 902 is specifically used to normalize the feature parameter set of each column in the feature matrix to obtain a processed feature matrix. The processed feature matrix is ​​then subjected to covariance processing, and the resulting covariance matrix is ​​used as the correlation coefficient matrix. The correlation coefficient matrix is ​​substituted into the characteristic equation for solving to obtain multiple eigenvalues ​​of the correlation coefficient matrix and the eigenvector corresponding to each eigenvalue. The processed feature matrix is ​​then projected onto the subspace formed by the eigenvectors corresponding to the first eigenvalues ​​to obtain the low-dimensional feature matrix of the target permutation contract. The first eigenvalue refers to the eigenvalue whose sum of contribution rates is greater than a preset feature contribution rate threshold after arranging the eigenvalues ​​in descending order of their contribution rates, where N is a positive integer.

[0217] In other embodiments, the target parameter value includes a maximum value conversion factor and a recommended value conversion factor. The maximum value conversion factor is used as the upper limit for adjusting the value conversion factor of the target replacement contract, and the recommended value conversion factor is used as a reference for adjusting the value conversion factor of the target replacement contract. The processing module 902 described above is further configured to: obtain the value conversion factor to be evaluated for the target replacement contract. The value conversion factor to be evaluated is the value conversion factor of the target replacement contract determined at a historical time. If the value conversion factor to be evaluated is greater than the maximum value conversion factor, a first evaluation result of the value conversion factor to be evaluated is generated. The first evaluation result is used to indicate the adjustment of the value conversion factor to be evaluated. If the value conversion factor to be evaluated is less than the maximum value conversion factor but greater than the recommended value conversion factor, a second evaluation result of the value conversion factor to be evaluated is generated. The second evaluation result is used to suggest the adjustment of the value conversion factor to be evaluated.

[0218] In other embodiments, such as Figure 10 As shown, the above-mentioned coefficient value determination device may further include a training module 903.

[0219] This training module 903 is used to acquire the temporal feature data of multiple sample permutation contracts within a preset historical time period. It determines the feature dimensionality reduction matrix corresponding to each sample's temporal feature data. Based on the credibility identifier of the sample object's owner for each sample permutation contract, it constructs a value conversion coefficient label matrix that corresponds one-to-one with each sample's temporal feature matrix. Each row of the temporal feature matrix corresponds to each row of the corresponding value conversion coefficient label matrix. Based on each feature low-dimensional matrix and its corresponding value conversion coefficient label matrix, it trains the initial model of the preset coefficient model to obtain the preset coefficient model.

[0220] In other embodiments, the training module 903 targets any one of the first sample replacement contracts in each sample replacement contract. Based on the credibility identifier of the target object's owner in each sample replacement contract, a value conversion coefficient label matrix corresponding one-to-one with the temporal feature matrix of each sample is constructed. Specifically, it is used to: perform a first operation: when the credibility identifier of the target object's owner corresponding to the first sample replacement contract is credible, determine the recommended value conversion coefficient label of the first sample replacement contract at the first historical moment based on the ratio of the quantified value of the target object of the first sample replacement contract at the first historical moment to the quantified value of the preset resource. Perform a second operation: use the preset highest value conversion coefficient corresponding to the target object type of the first sample replacement contract as the highest value conversion coefficient label of the first sample replacement contract at the first historical moment. Perform a third operation: use the second historical moment as the new first historical moment. The second historical moment is later than the first historical moment. Repeat the first operation, the first operation, and the third operation until the first historical moment is null, to obtain the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

[0221] In other embodiments, the training module 903 targets any one of the first sample replacement contracts in each sample replacement contract. Based on the credibility identifier of the target object's owner in each sample replacement contract, a value conversion coefficient label matrix corresponding one-to-one with the temporal feature matrix of each sample is constructed. Specifically, it is used to: perform a fourth operation: when the credibility identifier of the sample object's owner corresponding to the first sample replacement contract is untrustworthy, the historical value conversion coefficient of the first sample replacement contract is used as the highest value conversion coefficient label of the first sample replacement contract at the first historical moment. Perform a fifth operation: based on the ratio of the cumulative quantified value of the pre-defined resources recovered by the first sample replacement contract at the first historical moment to the quantified value of the pre-defined resources in the first sample replacement contract, the recommended value conversion coefficient label of the first sample replacement contract at the first historical moment is determined. Perform a sixth operation: using the second historical moment as the new first historical moment. The second historical moment is later than the first historical moment. Repeat the fourth, fifth, and sixth operations until the first historical moment is null, to obtain the recommended value conversion coefficient labels and the highest value conversion coefficient labels of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

[0222] The coefficient value determination device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0223] Figure 11 This is a schematic diagram of a coefficient value determination device provided in an embodiment of this application. Figure 11 As shown, the device for determining the coefficient value includes: a memory 1101, a transceiver 1102, and at least one processor 1103.

[0224] The transceiver 1102 is used to interact with other devices to send and receive data. For example, in this embodiment, the transceiver 1102 can specifically be used to obtain the time-series characteristic data of the target substitution contract.

[0225] The memory 1101 is used to store computer program code, which includes computer instructions. These computer instructions run in the aforementioned coefficient value determining device to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk, or optical disc, etc.

[0226] Processor 1103 can be a general-purpose processor, including a Central Processing Unit (CPU), a network processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 1103 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0227] The memory 1101, transceiver 1102, and processor 1103 are communicatively connected. For example, the memory 1101 and transceiver 1102 can be connected to the processor 1103 via a system bus to complete mutual communication. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0228] Optionally, the memory 1101 can be either standalone or integrated with the processor 1103. When the memory 1101 is set up independently, it is connected to the processor 1103 via the system bus.

[0229] This application also provides a chip for executing instructions, which is used to execute the technical solution of the coefficient value determination method in the above embodiments.

[0230] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solution of the coefficient value determination method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the coefficient value determination device can perform the technical solution of the coefficient value determination method described in the above embodiments.

[0231] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and when the at least one processor executes the computer program, it can implement the technical solution of the coefficient value determination method in the above embodiments.

[0232] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage media can be any available medium accessible to a general-purpose or special-purpose computer. An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0233] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0234] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0235] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0236] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0237] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0238] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0239] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for determining coefficient values, characterized in that, The method includes: Obtain the time-series feature data of the target replacement contract; the target replacement contract is used to agree on the preset resources for converting the target object into the target value; the time-series feature data includes the feature parameter set of the target replacement contract at multiple historical moments, and the feature parameter set at each historical moment includes: the attribute feature parameters of the target replacement contract, the feature parameters of the target object, the feature parameters of the party to which the target object belongs, the feature parameters of the domain to which the target object belongs, and the default handling strategy feature parameters of the target replacement contract; Based on the time-series feature data of the target permutation contract, a feature matrix of the target permutation contract is constructed; each row of the feature matrix represents the set of feature parameters of the target permutation contract at a single historical moment, and each column of the feature matrix represents a feature parameter of the target permutation contract at different historical moments; The feature matrix is ​​subjected to feature dimensionality reduction processing to obtain the low-dimensional feature matrix of the target permutation contract; The target parameter value of the target substitution contract is obtained by processing the feature low-dimensional matrix through a preset coefficient model. The target parameter value is the value conversion coefficient of the target object agreed upon by the target substitution contract. The preset coefficient model has the function of determining the parameter value of the object agreed upon by the substitution contract based on the feature low-dimensional matrix.

2. The method according to claim 1, characterized in that, The attribute characteristics parameters of the target replacement contract include: the duration of the target replacement contract, the purpose of the target replacement contract, and the third-party reserved resources of the target replacement contract; The characteristic parameters of the target object include: the type of the target object, the historical value change rate of the target object, and the depreciation rate of the target object; The characteristic parameters of the party to which the target object belongs include: the performance characteristic parameters of the party to which the target object belongs, and the credibility identifier of the party to which the target object belongs; The characteristic parameters of the target object's domain include: the development trend index of the target object's domain and the liquidity index of the target object's domain. The default handling strategy characteristic parameters of the target replacement contract include: the type of the default handling strategy, the default duration of the target replacement contract, and the preset resource recovery for the default of the target replacement contract.

3. The method according to claim 1, characterized in that, The step of performing feature dimensionality reduction on the feature matrix to obtain the low-dimensional feature matrix of the target permutation contract includes: The feature parameter set in each column of the feature matrix is ​​normalized to obtain the processed feature matrix; The processed feature matrix is ​​subjected to covariance processing, and the resulting covariance matrix is ​​used as the correlation coefficient matrix. The correlation coefficient matrix is ​​substituted into the characteristic equation and solved to obtain multiple eigenvalues ​​of the correlation coefficient matrix and the eigenvector corresponding to each eigenvalue; The processed feature matrix is ​​projected onto the subspace formed by the feature vectors corresponding to the first feature values ​​to obtain the low-dimensional feature matrix; the first feature value refers to the feature value whose sum of the feature contribution rates of the first N selected features is greater than a preset feature contribution rate threshold after the feature values ​​are arranged in descending order of feature contribution rate, where N is a positive integer.

4. The method according to claim 1, characterized in that, The target parameter value includes a maximum value conversion factor and a recommended value conversion factor; the maximum value conversion factor is used as the upper limit for adjusting the value conversion factor of the target replacement contract, and the recommended value conversion factor is used as a reference for adjusting the value conversion factor of the target replacement contract; the method further includes: Obtain the valuation factor of the target replacement contract; the valuation factor is the valuation factor of the target replacement contract determined at a historical time. If the conversion factor for the value to be assessed is greater than the highest conversion factor for the value, a first assessment result for the conversion factor for the value to be assessed is generated; the first assessment result is used to indicate an adjustment to the conversion factor for the value to be assessed. If the conversion factor of the value to be evaluated is less than the highest conversion factor but greater than the recommended conversion factor, a second evaluation result of the conversion factor of the value to be evaluated is generated; the second evaluation result is used to suggest adjusting the conversion factor of the value to be evaluated.

5. The method according to any one of claims 1-4, characterized in that, The preset coefficient model is obtained through the following method: Obtain the time-series feature data of each sample permutation contract within a preset historical time period; Determine the feature dimensionality reduction matrix corresponding to each of the time-series feature data of the samples; Based on the credibility identifier of the party to which the sample object belongs in each sample replacement contract, a value discount coefficient label matrix is ​​constructed that corresponds one-to-one with the temporal feature matrix of each sample; each row of the temporal feature matrix corresponds to each row of the corresponding value discount coefficient label matrix; The initial model of the preset coefficient model is trained based on each of the feature low-dimensional matrices and its corresponding value conversion coefficient label matrix to obtain the preset coefficient model.

6. The method according to claim 5, characterized in that, For any one of the first sample replacement contracts in each of the sample replacement contracts; the step of constructing a value discount coefficient label matrix corresponding one-to-one with each of the sample time-series feature matrices based on the credibility identifier of the target object belonging to each of the sample replacement contracts includes: Perform the first operation: If the credibility identifier of the party to which the sample object belongs to the first sample replacement contract is credible, determine the recommendation value conversion coefficient label of the first sample replacement contract at the first historical moment based on the ratio of the quantitative value of the target object of the first sample replacement contract to the quantitative value of the preset resource at the first historical moment. Perform the second operation: use the preset highest value conversion factor corresponding to the target object type of the first sample replacement contract as the highest value conversion factor label of the first sample replacement contract at the first historical moment; Perform the third operation, using the second historical moment as the new first historical moment; the second historical moment is later than the first historical moment. Repeat the first operation, the third operation, and the first operation until the first historical moment is empty, to obtain the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

7. The method according to claim 5, characterized in that, For any one of the first sample replacement contracts in each of the sample replacement contracts; the step of constructing a value discount coefficient label matrix corresponding one-to-one with the temporal feature matrix of each sample based on the credibility identifier of the target object belonging to each sample replacement contract; includes: Perform the fourth operation: If the credibility of the party to which the sample object belongs corresponding to the first sample replacement contract is identified as untrustworthy, use the historical value conversion factor of the first sample replacement contract as the highest value conversion factor label of the first sample replacement contract at the first historical moment. Perform the fifth operation: Based on the ratio of the cumulative quantified value of the preset resources recovered by the first sample replacement contract at the first historical moment to the quantified value of the preset resources in the first sample replacement contract, determine the recommended value conversion factor label of the first sample replacement contract at the first historical moment; Perform the sixth operation, using the second historical moment as the new first historical moment; the second historical moment is later than the first historical moment. Repeat the fourth, fifth, and sixth operations until the first historical moment is empty, to obtain the recommended value conversion coefficient label and the highest value conversion coefficient label of the first sample replacement contract at multiple historical moments, and construct the value conversion coefficient label matrix corresponding to the first sample replacement contract.

8. A device for determining coefficient values, characterized in that, include: The acquisition module is used to acquire the time-series characteristic data of the target replacement contract; The target replacement contract is used to agree on preset resources for converting a target object into a target value; The time-series feature data includes feature parameter sets of the target replacement contract at multiple historical moments. The feature parameter set at each historical moment includes: attribute feature parameters of the target replacement contract, feature parameters of the target object, feature parameters of the party to which the target object belongs, feature parameters of the domain to which the target object belongs, and default handling strategy feature parameters of the target replacement contract. The processing module is used to construct a feature matrix of the target substitution contract based on the time-series feature data of the target substitution contract; each row of the feature matrix represents the feature parameter set of the target substitution contract at a single historical moment, and each column of the feature matrix represents a feature parameter of the target substitution contract at different historical moments; the feature matrix is ​​subjected to feature dimensionality reduction processing to obtain a low-dimensional feature matrix of the target substitution contract; the low-dimensional feature matrix is ​​processed by a preset coefficient model to obtain the target parameter value of the target substitution contract, and the target parameter value is the value conversion coefficient of the target object agreed upon by the target substitution contract; the preset coefficient model has the function of determining the parameter value of the object agreed upon by the substitution contract based on the low-dimensional feature matrix.

9. A device for determining coefficient values, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the coefficient value determining device performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-7.

Images