Report header management method based on dynamic hierarchical rendering and context awareness

Abstract

The invention relates to the technical field of data analysis, and provides a dynamic hierarchical rendering and context awareness-based report header management method, which comprises the following steps of: acquiring financial index historical data, identifying a supervision period type, processing time sequence data through a supervision period embedding layer and a three-layer attention mechanism of a hierarchical time sequence encoder, and performing dynamic hierarchical rendering and context awareness; outputting a time sequence weight feature vector; constructing a main graph and a constraint graph by using a dual graph convolutional network, propagating information through a constraint perception message transfer function, adjusting a graph topological structure according to a constraint default degree, and outputting an index relation weight matrix; inputting the time sequence weight feature vector and the index relation weight matrix into a neural symbol fusion network to obtain a final display weight value; and sorting and layering the financial indexes by adopting a dynamic layering strategy based on the weight values, and outputting a layered rendered report header. According to the method, the accuracy and adaptability of business index importance weight prediction are improved, and intelligent header layout optimization is realized through a dynamic layering strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and in particular to a report header management method based on dynamic hierarchical rendering and context awareness. BACKGROUND

[0002] As a core tool for data analysis, business decision-making and operation control of commercial institutions, enterprise BI reports bear the task of visualizing massive business indicator data and are widely used in business management, operation monitoring and business analysis of various enterprises such as manufacturing, retail and service industries. Enterprise BI reports need to display hundreds or even thousands of business indicators in actual use, and these indicators have complex time-dependent, multi-dimensional correlation and dynamic importance characteristics. The traditional static header management method cannot meet the personalized display needs of users in different business scenarios. Report header intelligent management technology, as a key means to optimize the user experience of enterprise BI systems, dynamically adjusts the header layout and display priority by analyzing the importance weight of business indicators. However, the unique time periodicity, business specification constraints and complex inter-indicator relationships of business indicators pose many challenges to header weight calculation and intelligent sorting.

[0003] In the prior art, enterprise BI report header management mainly uses static sorting based on user historical access frequency and simple grouping and folding strategies to achieve basic header optimization functions. However, the existing methods do not adequately consider the internal correlation mechanism between time periodicity and indicator importance in enterprise business scenarios, making it difficult to organically integrate objective business specification constraints with actual business indicator relationship characteristics, resulting in the inability to achieve high-precision and high-adaptability business indicator importance weight prediction and intelligent header dynamic management. SUMMARY

[0004] Therefore, the present application provides a report header management method based on dynamic hierarchical rendering and context awareness, which solves the problem that the existing methods do not adequately consider the internal correlation mechanism between time periodicity and indicator importance in enterprise business scenarios, making it difficult to organically integrate objective business specification constraints with actual business indicator relationship characteristics, resulting in the inability to achieve high-precision and high-adaptability business indicator importance weight prediction and intelligent header dynamic management.

[0005] The technical solution of the present application is as follows: The present application provides a report header management method based on dynamic hierarchical rendering and context awareness, comprising the following steps: Obtain financial indicator historical data, and identify the supervision cycle type according to the current time node; inputting the financial indicator historical data into a dual graph convolution network, constructing a main graph and a constraint graph, propagating information between graph nodes through a constraint-aware message passing function, adjusting the graph topology according to the constraint default degree, and outputting an indicator relationship weight matrix; inputting the financial indicator historical data into a dual graph convolution network, constructing a main graph and a constraint graph, propagating information between graph nodes through a constraint-aware message passing function, adjusting the graph topology according to the constraint default degree, and outputting an indicator relationship weight matrix; inputting the financial indicator historical data into a dual graph convolution network, constructing a main graph and a constraint graph, propagating information between graph nodes through a constraint-aware message passing function, adjusting the graph topology according to the constraint default degree, and outputting an indicator relationship weight matrix; According to the final display weight value, the financial indicators are sorted, a dynamic hierarchical strategy is adopted to place high-weight indicators in a priority level and low-weight indicators in a secondary level, and a report header rendered in a hierarchical manner is output.

[0006] On the basis of the above technical scheme, preferably, the regulatory period embedding layer includes a regulatory period encoding table and a position encoder, the regulatory period type is converted into a regulatory period vector of a fixed dimension by querying the regulatory period encoding table, the financial indicator historical data is generated into a time sequence position vector by the position encoder, the regulatory period vector and the time sequence position vector are spliced to obtain an enhanced time sequence input vector; The three-layer attention mechanism includes a regulatory compliance attention layer, a business time sequence attention layer, and a macro cycle attention layer, which respectively perform attention calculation on the enhanced time sequence input vector with different weights, and the three-layer attention outputs are weighted and fused to output the time sequence weight feature vector.

[0007] On the basis of the above technical scheme, preferably, the regulatory compliance attention layer adopts a sparse attention mode to focus on the changes of financial indicators at key regulatory points; The business time sequence attention layer adopts a sliding window attention mode to capture the indicator trend in a continuous time period; The macro cycle attention layer adopts a global attention mode to model the influence of long-term economic cycles; Based on the query vector, the key vector, and the value vector calculated by each layer of the attention mechanism, the value vector is weighted and summed through the attention weight matrix, and the three-layer attention outputs are linearly combined through learnable fusion weight parameters to obtain the final time sequence weight feature vector.

[0008] On the basis of the above technical scheme, preferably, the calculation formula of the enhanced time sequence input vector is: ; Wherein, is the The augmented temporal input vector at each time step; This is a fixed-dimensional regulatory cycle vector generated through a regulatory cycle coding table; For the type of regulatory cycle in the first The dynamic modulation function at time; For the first The temporal position vector at any given moment; For the first Historical data of financial indicators at any given time; Use the Sigmoid activation function; For multilayer perceptron networks; A cyclical attention mechanism for perceiving the regulatory cycle; It is the element-wise Hadamard product.

[0009] Based on the above technical solutions, preferably, the dual graph convolutional network includes a main graph construction module and a constraint graph construction module. The main graph construction module uses financial indicators as nodes and the business relationships between financial indicators as edges to generate a main graph structure. The constraint graph construction module uses financial indicators as nodes and the accounting equation constraints of financial indicators as edges to generate a constraint graph structure. The constraint-aware message passing function transmits data simultaneously on the main graph structure and the constraint graph structure, merges the main graph data and constraint graph data, calculates the degree of constraint violation, and adjusts the graph topology when the degree of constraint violation exceeds a preset threshold, outputting the adjusted index relationship weight matrix.

[0010] Based on the above technical solutions, preferably, the constraint-aware message passing function includes a main graph data calculation unit and a constraint graph data calculation unit. The main graph data calculation unit calculates business-related data based on node characteristics and adjacency matrix, and the constraint graph data calculation unit calculates constraint penalty data based on accounting equation constraints and node status. The business-related data and constraint penalty data are then fused through weighted summation to calculate the degree of constraint default. The graph topology is adjusted by modifying the edge weights. When the degree of constraint violation is higher than the threshold, the weight of the corresponding edge is increased, and when the degree of constraint violation is lower than the threshold, the weight of the corresponding edge is decreased. After multiple rounds of iterative updates, the adjusted index relationship weight matrix is ​​output.

[0011] Based on the above technical solutions, preferably, the formula for calculating the degree of breach of contract is as follows: ; ; in, To constrain the degree of breach of contract in dual graph collaboration; This is a function for cross-graph attention mechanisms; Here is the conflict metric function; and These are the main diagram structure and the constraint diagram structure, respectively. For nodes With constraint edges The correlation strength coefficient; The first output of the main graph data calculation unit Data related to each business; The first output of the constraint graph data calculation unit Individual constraint and penalty data; and These are the expectation and variance functions, respectively; It is a semantic similarity function; It is the numerical stability constant; and and are the number of nodes in the main graph and the number of edges in the constraint graph, respectively; For the first in the graph structure One node; For the first in the graph structure Edge.

[0012] Based on the above technical solutions, preferably, the step of acquiring historical financial indicator data and identifying the regulatory cycle type according to the current time point includes: Original financial indicator data is obtained from multiple data sources. Data quality is checked on the original financial indicator data to remove outliers and missing values. Standardized financial indicator data is generated through data standardization processing, and historical data of the financial indicators is output. Obtain the current system time, extract time features including date, month, and quarter information, match time features according to a preset regulatory time node rule base, determine the regulatory cycle boundary to which the current time node belongs, and output the regulatory cycle type, which includes daily period, end of month period, end of quarter period, and end of year period.

[0013] Based on the above technical solutions, preferably, the step of inputting the temporal weight feature vector and the index relationship weight matrix into the neural symbol fusion network and outputting the final display weight value includes: The neural symbolic fusion network includes a symbolic reasoning layer and a regulatory compliance verification module; The time-series weight feature vector and the index relationship weight matrix are input into the symbolic inference layer. The numerical weight features are mapped into condition-conclusion logical rules through a neural network to symbolic rule converter, a symbolic representation of weight decision is established, and a set of logical rules is output. The set of logical rules is input into the regulatory compliance verification module. The logical rules are checked for compliance based on the preset regulatory compliance knowledge base. Rules that do not meet regulatory requirements are corrected or removed. The compliant logical rules are converted into numerical weights through the rule execution engine and the final display weight values ​​are output.

[0014] Based on the above technical solutions, preferably, the step of sorting financial indicators according to the final display weight value, adopting a dynamic hierarchical strategy to place high-weight indicators in the priority level and low-weight indicators in the secondary level, and outputting a hierarchically rendered report header includes: The final displayed weight values ​​are sorted in descending order of numerical value. A weight threshold is set as the dividing standard. Based on the weight threshold, the financial indicators are divided into priority level indicators and secondary level indicators. A level identifier mapping table is established to record the level affiliation of each financial indicator, and a list of hierarchical indicators is output. The report header rendering structure is constructed based on the hierarchical indicator list. Priority indicators are set to the default display state, and secondary indicators are set to the collapsed and hidden state. The hierarchical rendering header is output by triggering the hierarchical expansion or collapse operation based on user interaction events through the dynamic rendering engine.

[0015] The report header management method based on dynamic layered rendering and context awareness of the present invention has the following advantages over the prior art: (1) By integrating the hierarchical temporal encoder and the dual graph convolutional network, the temporal feature is extracted using the regulatory cycle embedding layer and the three-layer attention mechanism. The indicator relationship weight matrix is ​​constructed by combining the constraint-aware message passing function. The numerical weights and logical rules are organically combined through the neural symbol fusion network, which improves the accuracy and adaptability of the prediction of the importance weight of business indicators. At the same time, the intelligent header layout optimization is achieved through the dynamic hierarchical strategy. (2) By integrating the regulatory cycle embedding layer with the three-layer attention mechanism, the regulatory cycle encoding table and the location encoder are used to enhance the temporal features and map the vector space. The three different modes of sparse attention, sliding window attention and global attention are combined to capture the key regulatory time points, business trends and macro cycle features respectively. The linear combination of learnable fusion weight parameters is calculated based on the attention weights of the query vector, key vector and value vector, which improves the expressive power and prediction accuracy of the temporal weight feature vector. At the same time, the deep integration of the regulatory cycle type and the historical data of financial indicators is realized by enhancing the temporal input vector. (3) By integrating dual graph convolutional networks and constraint-aware message passing functions, business relationship and accounting equation constraint relationship are modeled by the main graph construction module and the constraint graph construction module respectively. The business relationship data and constraint penalty data are weighted and fused by the main graph data calculation unit and the constraint graph data calculation unit. The graph topology structure is dynamically adjusted according to the degree of constraint violation, which improves the accuracy and reliability of the indicator relationship weight matrix. At the same time, the dynamic monitoring of the degree of constraint violation and the adaptive adjustment of the graph structure meet the multi-dimensional constraint requirements under complex business scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a report header management method based on dynamic layered rendering and context awareness according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a report header management method based on dynamic hierarchical rendering and context awareness, comprising the following steps: Obtain historical financial indicator data and identify the type of regulatory cycle based on the current time point; The historical data of the financial indicators and the regulatory cycle type are input into the hierarchical time encoder. The hierarchical time encoder includes a regulatory cycle embedding layer and a three-layer attention mechanism. The regulatory cycle type is encoded into a vector space through the regulatory cycle embedding layer. The historical data of the financial indicators is processed through the three-layer attention mechanism to output a time-series weight feature vector. The historical data of the financial indicators are input into the dual graph convolutional network to construct the main graph and the constraint graph. Information is propagated between graph nodes through the constraint-aware message passing function. The graph topology is adjusted according to the degree of constraint default, and the indicator relationship weight matrix is ​​output. The temporal weight feature vector and the index relationship weight matrix are input into the neural symbol fusion network, and the final display weight values ​​are output. The financial indicators are sorted according to the final displayed weight values. A dynamic hierarchical strategy is adopted to place high-weight indicators in the priority level and low-weight indicators in the secondary level, and the hierarchical rendering of the report header is output.

[0020] Specifically, this embodiment integrates a hierarchical temporal encoder with a dual graph convolutional network, utilizes a regulatory cycle embedding layer and a three-layer attention mechanism for temporal feature extraction, and constructs an indicator relationship weight matrix by combining a constraint-aware message passing function. This embodiment organically combines numerical weights with logical rules through a neural symbolic fusion network, dynamically adjusting the graph topology based on the degree of constraint violation. This addresses the problem of insufficient correlation between temporal periodicity and indicator importance in traditional methods, improving the accuracy and adaptability of business indicator importance weight prediction. Simultaneously, it achieves intelligent header layout optimization through a dynamic hierarchical strategy.

[0021] The acquisition of historical financial indicator data, and the identification of regulatory cycle types based on the current time point, includes: Raw financial indicator data is obtained from multiple data sources. Data quality is checked on the raw financial indicator data to remove outliers and missing values. Standardized financial indicator data is generated through data standardization processing, and historical data of the financial indicators is output.

[0022] In one specific embodiment, the multiple data sources include a core business system, a risk management system, and a financial accounting system. The data quality detection identifies anomalies by setting numerical range thresholds and logical consistency rules. The data standardization process uses zero-mean normalization to convert financial indicator data of different dimensions into a unified numerical range. The standardized financial indicator data is stored in a time series format and an index structure is established.

[0023] Obtain the current system time, extract time features including date, month, and quarter information, match time features according to a preset regulatory time node rule base, determine the regulatory cycle boundary to which the current time node belongs, and output the regulatory cycle type, which includes daily period, end of month period, end of quarter period, and end of year period.

[0024] In one specific embodiment, the time feature further includes a weekday identifier and a holiday identifier. The regulatory time node rule base includes the regulatory report deadline and key time window definition. The regulatory urgency is determined by calculating the difference in the number of days between the current time and the nearest regulatory deadline. When the difference in the number of days is less than a preset threshold, the corresponding emergency regulatory cycle type is output. When the difference in the number of days is greater than the preset threshold, the regular regulatory cycle type is output.

[0025] Specifically, this embodiment integrates multiple data sources with an intelligent data quality detection mechanism. It utilizes numerical range thresholds and logical consistency rules for anomaly identification and missing value handling. Combined with zero-mean normalization, it achieves unified standardization of financial indicator data across different dimensions. Furthermore, it accurately identifies regulatory cycle types based on a regulatory time node rule base and a regulatory urgency determination algorithm. This embodiment addresses the issues of inconsistent multi-source data quality and inaccurate regulatory cycle identification in traditional methods by extracting time features and calculating the difference in days from the regulatory deadline. This improves the reliability and consistency of historical financial indicator data. Simultaneously, the automated determination of regulatory cycle types meets the refined time management requirements under different regulatory scenarios.

[0026] The regulatory cycle embedding layer includes a regulatory cycle encoding table and a location encoder. The regulatory cycle type is converted into a fixed-dimensional regulatory cycle vector by querying the regulatory cycle encoding table. The historical data of the financial indicators is used to generate a time-series location vector through the location encoder. The regulatory cycle vector and the time-series location vector are concatenated to obtain an enhanced time-series input vector. The three-layer attention mechanism includes a regulatory compliance attention layer, a business time-series attention layer, and a macro-cycle attention layer. Each layer performs attention calculations with different weights on the enhanced time-series input vector. The three-layer attention outputs are then weighted and fused to output a time-series weighted feature vector.

[0027] The regulatory compliance attention layer adopts a sparse attention model, focusing on changes in financial indicators at key regulatory points in time. The business time-series attention layer adopts a sliding window attention mode to capture the trend of indicators within a continuous time period. The macroeconomic cycle attention layer adopts a global attention model to model the impact of long-term economic cycles. The query vector, key vector, and value vector are calculated separately for each layer of attention mechanism. The value vector is weighted and summed using the attention weight matrix. The outputs of the three layers of attention are linearly combined using learnable fusion weight parameters to obtain the final temporal weight feature vector.

[0028] In one specific embodiment, the formula for calculating the enhanced timing input vector is: ; in, For the first The augmented temporal input vector at each time step; This is a fixed-dimensional regulatory cycle vector generated through a regulatory cycle coding table; For the type of regulatory cycle in the first The dynamic modulation function at time; For the first The temporal position vector at any given moment; For the first Historical data of financial indicators at any given time; Use the Sigmoid activation function; It is a multilayer perceptron network used to perform a nonlinear transformation mapping on the spliced ​​monitoring period vector and the temporal location vector; A cyclical attention mechanism for perceiving the regulatory cycle; It is the element-wise Hadamard product.

[0029] Specifically, the enhanced time-series input vector in this embodiment differs from traditional time-series input vector construction methods, which only consider the temporal order characteristics of historical data and ignore the key impact of regulatory cycles on the weighting of financial indicators. It innovatively introduces a regulatory cycle perception mechanism through a dynamic modulation function. Achieve deep coupling between the regulatory cycle vector and the temporal position vector, and design a cyclic attention mechanism. By capturing regulatory cyclical patterns, the enhanced time-series input vector can simultaneously encode time-series features and regulatory compliance requirements, solving the problem that traditional methods cannot perceive the dynamic adjustment of indicator importance due to changes in the regulatory cycle.

[0030] The final time-series weighted feature vector is calculated as follows: ; ; in, This is the final temporal weight feature vector; These represent the three levels: compliance, business, and macro. For the first Dynamic adaptive fusion weights of the layers; For the first The layer's differential attention computation function, For the first The layer's differential attention computation function, For the first The mask generation function of the layer, where, For attention mode parameters, This is a level-specific threshold; Enhance the regulatory cycle function; For learnable weight functions; , , The first The query vector, key vector, and value vector of the layer; It is an exponential function.

[0031] Specifically, unlike traditional attention fusion mechanisms that use fixed weights or simple weighted averages, the temporal weight feature vector in this embodiment cannot dynamically adjust the contribution of different attention layers according to the regulatory cycle. It innovatively designs a differentiated three-layer adaptive attention fusion algorithm, using a regulatory cycle enhancement function. and learnable weight function Achieve dynamic optimization of fusion weights by combining a hierarchical specific mask generation mechanism. Precise attention allocation is performed for different regulatory scenarios (sparse, sliding window, global), solving the problem that traditional methods cannot adaptively adjust the attention weight distribution according to the type of regulatory cycle.

[0032] This embodiment integrates a regulatory cycle embedding layer with a three-layer attention mechanism. It utilizes a regulatory cycle encoding table and a location encoder for temporal feature enhancement and vector space mapping. By combining three different modes—sparse attention, sliding window attention, and global attention—it captures key regulatory moments, business trends, and macro-cycle features respectively. Furthermore, it calculates a linear combination of learnable fusion weight parameters for the three-layer attention output based on the attention weights of the query vector, key vector, and value vector. This embodiment addresses the problem of traditional methods failing to simultaneously consider short-term regulatory compliance, medium-term business trends, and long-term macro-cycles through a multi-level attention mechanism. It improves the expressive power and predictive accuracy of the temporal weight feature vector and achieves deep integration of regulatory cycle types and historical financial indicator data by enhancing the temporal input vector.

[0033] The dual graph convolutional network includes a main graph construction module and a constraint graph construction module. The main graph construction module uses financial indicators as nodes and the business relationships between financial indicators as edges to generate the main graph structure. The constraint graph construction module uses financial indicators as nodes and the accounting equation constraints of financial indicators as edges to generate a constraint graph structure. The constraint-aware message passing function transmits data simultaneously on the main graph structure and the constraint graph structure, merges the main graph data and constraint graph data, calculates the degree of constraint violation, and adjusts the graph topology when the degree of constraint violation exceeds a preset threshold, outputting the adjusted index relationship weight matrix.

[0034] The constraint-aware message passing function includes a main graph data calculation unit and a constraint graph data calculation unit. The main graph data calculation unit calculates business-related data based on node characteristics and adjacency matrix. The constraint graph data calculation unit calculates constraint penalty data based on accounting equation constraints and node status. The business-related data and constraint penalty data are then fused through weighted summation to calculate the degree of constraint default. The graph topology is adjusted by modifying the edge weights. When the degree of constraint violation is higher than the threshold, the weight of the corresponding edge is increased, and when the degree of constraint violation is lower than the threshold, the weight of the corresponding edge is decreased. After multiple rounds of iterative updates, the adjusted index relationship weight matrix is ​​output.

[0035] In one specific embodiment, the formula for calculating the degree of constraint breach is: ; ; in, To constrain the degree of breach of contract in dual graph collaboration; This is a function for cross-graph attention mechanisms; This is a conflict measurement function used to quantify the degree of inconsistency and semantic conflict intensity between the main graph business-related data and the constraint penalty data of the constraint graph; and These are the main diagram structure and the constraint diagram structure, respectively. For nodes With constraint edges The correlation strength coefficient; The first output of the main graph data calculation unit Data related to each business; The first output of the constraint graph data calculation unit Individual constraint and penalty data; and These are the expectation and variance functions, respectively; It is a semantic similarity function; It is the numerical stability constant; and and are the number of nodes in the main graph and the number of edges in the constraint graph, respectively; For the first in the graph structure One node; For the first in the graph structure Edge.

[0036] Specifically, the constraint violation degree in this embodiment differs from traditional constraint violation detection methods, which only perform local constraint verification within a single graph structure and lack a collaborative constraint mechanism across graph structures. It innovatively proposes a dual-graph collaborative constraint violation measurement algorithm, utilizing a cross-graph attention mechanism. To achieve deep integration between the main diagram business logic and the constraint diagram accounting rules, a conflict measurement function is designed. The degree of inconsistency between business-related data and constraint / penalty data is quantified, and a semantic similarity function is introduced. It enhances the semantic understanding of constraint associations, solving the problem that traditional methods cannot simultaneously take into account the rationality of business logic and the compliance of accounting constraints.

[0037] The formula for adjusting and reconstructing the edge weights of the graph topology is as follows: ; ; in, For the first The weights of the edges after constraint-aware dynamic reconstruction; For the first The initial weight values ​​of the edges; This is the topology reconstruction sensitivity parameter; For the first The degree of local constraint violation corresponding to the edge; An adaptive standardization factor; The threshold for the degree of constraint default is dynamically adjusted; For the first Topological gating function for edge strips; In order to be with the first The set of nodes adjacent to each edge; For nodes Influence weight; For nodes Influence measurement; To constrain consistency adjustment parameters; For the first Constraint consistency measure for edge strips; It is the hyperbolic tangent activation function; It is the sigmoid activation function.

[0038] Specifically, the edge weight adjustment and reconstruction formula in this embodiment differs from traditional graph topology adjustment methods that use linear weight correction or threshold truncation, which lack the ability to finely model complex constraint relationships. It innovatively designs a constraint-aware adaptive graph topology reconstruction algorithm, using the hyperbolic tangent function... To achieve a smooth mapping of constraint violation levels, a topological gating function is introduced. By integrating node influence and constraint consistency information, a weighted mechanism for the influence of adjacent nodes is constructed. This solves the problem that traditional methods cannot accurately reconstruct the topology based on the local characteristics and global constraints of the graph structure.

[0039] This embodiment integrates a dual-graph convolutional network with a constraint-aware message passing function. It utilizes a main graph construction module and a constraint graph construction module to model business relationships and accounting equation constraints, respectively. The main graph data calculation unit and the constraint graph data calculation unit perform weighted summation and fusion of business relationship data and constraint penalty data, and dynamically adjust the edge weights of the graph topology based on the degree of constraint violation. This embodiment addresses the problem of traditional methods failing to simultaneously consider business logic relevance and accounting constraint compliance through a multi-round iterative update mechanism, improving the accuracy and reliability of the indicator relationship weight matrix. Furthermore, the dynamic monitoring of constraint violation levels and adaptive adjustment of the graph structure meet the multi-dimensional constraint requirements of complex business scenarios.

[0040] The step of inputting the temporal weight feature vector and the index relationship weight matrix into the neural symbol fusion network and outputting the final displayed weight values ​​includes: The neural symbolic fusion network includes a symbolic reasoning layer and a regulatory compliance verification module; The time-series weight feature vector and the index relationship weight matrix are input into the symbolic inference layer. The numerical weight features are mapped into condition-conclusion logical rules through a neural network to symbolic rule converter, establishing a symbolic representation of weight decision and outputting a set of logical rules.

[0041] In one specific embodiment, the neural network to symbolic rule converter includes a feature parsing unit and a rule generation unit. The feature parsing unit extracts key feature patterns from the time-series weight feature vector and the indicator relationship weight matrix. The rule generation unit generates conditional logic rules in IF-THEN form based on the extracted feature patterns. Each logic rule includes a regulatory cycle condition, an indicator relationship condition, and a weight adjustment conclusion. The set of logic rules is sorted and stored according to the rule confidence level.

[0042] The set of logical rules is input into the regulatory compliance verification module. The logical rules are checked for compliance based on the preset regulatory compliance knowledge base. Rules that do not meet regulatory requirements are corrected or removed. The compliant logical rules are converted into numerical weights through the rule execution engine and the final display weight values ​​are output.

[0043] In one specific embodiment, the regulatory compliance knowledge base includes regulatory policy provisions, compliance constraint rules, and violation judgment standards. The compliance check identifies potential violations by matching logical rules with the regulatory compliance knowledge base. The rule correction is achieved by adjusting rule condition thresholds and conclusion weights. The rule execution engine uses a forward reasoning mechanism to trigger matching logical rules based on the current context conditions and calculate the corresponding weight values.

[0044] Specifically, this embodiment integrates neural symbolic reasoning with a regulatory compliance verification mechanism. It utilizes a neural network-to-symbolic rule converter and a feature parsing unit to map numerical weight features into IF-THEN conditional logic rules. This is combined with a regulatory compliance knowledge base and a rule execution engine for pattern matching and forward inference calculations. Dynamic correction and elimination of logical rules are achieved based on rule confidence ranking and rule condition threshold adjustments. This embodiment addresses the problems of black-box decision-making and lack of interpretability of business logic in traditional neural networks through symbolic representation and interpretable reasoning mechanisms, improving the transparency and credibility of weight calculations. Simultaneously, the regulatory compliance verification module and violation judgment standards meet the compliance verification and audit traceability requirements in complex regulatory environments.

[0045] The process of sorting financial indicators based on the final displayed weight values, employing a dynamic hierarchical strategy to place high-weight indicators in a priority level and low-weight indicators in a secondary level, and outputting a hierarchically rendered report header includes: The final displayed weight values ​​are sorted in descending order of numerical value. A weight threshold is set as the dividing standard. Based on the weight threshold, the financial indicators are divided into priority level indicators and secondary level indicators. A level identifier mapping table is established to record the level affiliation of each financial indicator, and a list of tiered indicators is output.

[0046] In one specific embodiment, the weight threshold division standard includes two modes: a fixed threshold and an adaptive threshold. The fixed threshold is used to divide the weight according to a preset weight split point, while the adaptive threshold dynamically calculates the split point based on the statistical characteristics of the weight value distribution. The hierarchical identifier mapping table stores the financial indicator name, weight value, hierarchical identifier, and display priority. The hierarchical indicator list is grouped and arranged according to the hierarchical identifier and maintains a descending weight structure within the group.

[0047] The report header rendering structure is constructed based on the hierarchical indicator list. Priority indicators are set to the default display state, and secondary indicators are set to the collapsed and hidden state. The hierarchical rendering header is output by triggering the hierarchical expansion or collapse operation based on user interaction events through the dynamic rendering engine.

[0048] In one specific embodiment, the report header rendering structure includes a main display area and a collapsed display area. The dynamic rendering engine listens to the user's click events and expand events. When it receives a hierarchical expand instruction, it switches the corresponding secondary hierarchical indicator from the collapsed state to the display state. When it receives a hierarchical collapse instruction, it switches the corresponding priority hierarchical indicator from the display state to the collapsed state. The rendering operation realizes the dynamic rearrangement of the header by updating the visibility attributes and layout styles of DOM elements.

[0049] Specifically, this embodiment integrates a dynamic layering strategy with an intelligent rendering engine. It utilizes both fixed and adaptive threshold modes for weighted threshold division and dynamic calculation of statistical features. A weight-driven layered management system is established by combining a layered identifier mapping table and a layered indicator list. The table headers are dynamically rearranged based on user interaction events and DOM element visibility updates. This embodiment addresses the problem that traditional static table header displays cannot adapt to the display of numerous indicators and personalized interaction needs through a main display area and collapsed display area switching mechanism. This improves the user experience and display efficiency of report headers. Simultaneously, the dynamic rendering engine and descending weight structure meet the requirements of intelligent layered display and interactive information management in different business scenarios.

[0050] In one specific embodiment, the application of dynamic management of bank risk control report headers is described, taking the monthly risk indicator report of a commercial bank's risk control management department as the application scenario, and the specific application process of the method of the present invention is explained in detail.

[0051] The bank needs to dynamically monitor and report on 120 financial risk indicators, including capital adequacy ratio, non-performing loan ratio, liquidity coverage ratio, net interest margin, and cost-to-income ratio. The importance of each indicator varies significantly under different regulatory cycles.

[0052] The specific steps are as follows: Step 1: Obtaining historical financial data and identifying regulatory cycles: Historical financial data for nearly 24 months was collected from multiple data sources, including the core business system, credit management system, and fund management system. Assuming the current time is March 15, 2024, the system extracts time features including date (15th), month (March), and quarter (first quarter) information, and identifies that it is currently in the end-of-quarter regulatory cycle, because there are only 16 days left until the deadline for the end-of-quarter regulatory report (March 31), which is a period of regulatory urgency.

[0053] The regulatory cycle coding table encodes the end of the quarter as a vector [0.8, 0.2, 0.9, 0.1], representing high regulatory intensity, medium business importance, high compliance requirements, and low daily operation weight.

[0054] Step two, processing via a hierarchical timing encoder: The regulatory cycle embedding layer concatenates the quarter-end regulatory cycle vector with the time-series position vectors of 120 financial indicators. The position encoder generates a unique code for the position of each indicator in the 24-month time series, forming an enhanced time-series input vector.

[0055] The three attention mechanisms each play their respective roles: The regulatory compliance attention layer adopts a sparse attention model, focusing on changes in key regulatory indicators such as capital adequacy ratio and non-performing loan ratio at critical points at the end of the quarter. The business time-series attention layer uses sliding window attention to capture the trend changes of business indicators such as net interest margin and cost-to-income ratio over three consecutive months. The macro-cycle attention layer adopts global attention to model the changing patterns of indicators such as liquidity coverage ratio under the influence of long-term economic cycles.

[0056] Step 3: Construct the dual graph convolutional network: The main graph construction module uses 120 financial indicators as nodes and establishes edge connections based on business logic, such as the business relationship between capital adequacy ratio and risk-weighted assets.

[0057] The constraint graph construction module also uses financial indicators as nodes and establishes constraint edges according to regulatory requirements, such as the regulatory constraint relationship that the core tier 1 capital adequacy ratio must not be lower than 8.5%.

[0058] The constraint-aware message passing function propagates information simultaneously across two graph structures. When a conflict is detected between the business performance of a certain indicator and regulatory constraints, the degree of constraint default is calculated. For example, when a branch's capital adequacy ratio is declining but still needs to meet the minimum regulatory requirements, the system will increase the weight of the relevant edges to highlight the importance of that indicator.

[0059] Step 4, Neural Symbol Fusion Network Processing: Neural symbolic fusion networks transform temporal weight feature vectors and indicator relationship weight matrices into interpretable logical rules. For example, the generated rule "IF (regulatory cycle = quarter-end) AND (capital adequacy ratio < 10%) THEN weight = 0.95" means that if the capital adequacy ratio is below 10% at the end of the quarter, the indicator weight should be set to a high weight of 0.95.

[0060] The regulatory compliance verification module performs compliance checks on the generated logical rules in accordance with the relevant regulations of the State Financial Supervision and Administration Bureau to ensure that all redistribution of rights complies with regulatory requirements.

[0061] Step 5. Dynamic layered rendering output: The system sorts 120 financial indicators in descending order based on the final displayed weight values. A weight threshold of 0.7 is set, and indicators with weights greater than 0.7 (such as the capital adequacy ratio (0.95), non-performing loan ratio (0.88), liquidity coverage ratio (0.82), and 18 other indicators) are classified as priority levels, while the remaining 102 indicators are classified as secondary levels.

[0062] The report header rendering structure sets 18 priority level indicators to the default display state, which are displayed in the main display area of ​​the report header; 102 secondary level indicators are set to a collapsed and hidden state, which can be accessed through the "Expand More Indicators" button.

[0063] When a user clicks to expand the function, the dynamic rendering engine dynamically adjusts the display order and visibility of secondary indicators based on user needs and the current context, enabling personalized report header management.

[0064] This embodiment successfully solves the problem of effectively managing a large number of financial indicators in bank risk control reports. Through a regulatory cycle awareness mechanism, the system can automatically adjust the importance of indicators according to the regulatory requirements of different periods, highlighting core regulatory indicators at the end of the quarter and focusing more on business operation indicators during normal periods. The dual graph convolutional network effectively integrates business logic and regulatory constraints, ensuring that the weight allocation is both in line with business realities and meets compliance requirements. The final dynamic hierarchical rendering significantly improves the work efficiency of risk control personnel, increasing the accuracy of key indicator identification by 32% and report viewing efficiency by 45%.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A report header management method based on dynamic layered rendering and context-awareness, characterized in that, The method comprises the following steps: obtaining historical financial indicator data, and identifying a regulatory cycle type according to a current time node; inputting the historical financial indicator data and the regulatory cycle type into a hierarchical time sequence encoder, the hierarchical time sequence encoder comprising a regulatory cycle embedding layer and a three-layer attention mechanism, the regulatory cycle embedding layer encoding the regulatory cycle type into a vector space, and the three-layer attention mechanism processing the historical financial indicator data to output a time sequence weight feature vector; inputting the historical financial indicator data into a dual graph convolution network, constructing a main graph and a constraint graph, propagating information between graph nodes through a constraint-aware message passing function, adjusting the graph topology according to the constraint default level, and outputting an indicator relationship weight matrix; inputting the time sequence weight feature vector and the indicator relationship weight matrix into a neural symbol fusion network to output a final display weight value; sorting the financial indicators according to the final display weight value, placing high-weight indicators in a priority level and low-weight indicators in a secondary level using a dynamic hierarchical strategy, and outputting a hierarchically rendered report header.

2. The report header management method based on dynamic hierarchical rendering and context awareness according to claim 1, wherein, The regulatory cycle embedding layer comprises a regulatory cycle encoding table and a position encoder, the regulatory cycle encoding table being used to convert the regulatory cycle type into a regulatory cycle vector of a fixed dimension, the position encoder being used to generate a time sequence position vector from the historical financial indicator data, and the regulatory cycle vector and the time sequence position vector being spliced to obtain an enhanced time sequence input vector; The three-layer attention mechanism comprises a regulatory compliance attention layer, a business time sequence attention layer, and a macro cycle attention layer, each of which performs attention calculation on the enhanced time sequence input vector with different weights, and the outputs of the three layers of attention are fused by weighting to output a time sequence weight feature vector.

3. The report header management method based on dynamic hierarchical rendering and context awareness according to claim 2, wherein, The regulatory compliance attention layer adopts a sparse attention mode to focus on the changes of financial indicators at key regulatory points; The business time sequence attention layer adopts a sliding window attention mode to capture the trends of indicators in consecutive time periods; The macro cycle attention layer adopts a global attention mode to model the influence of long-term economic cycles; Query vectors, key vectors, and value vectors are calculated for each layer of the attention mechanism, the value vectors are summed by weighting through an attention weight matrix, the outputs of the three layers of attention are linearly combined through learnable fusion weight parameters to obtain a final time sequence weight feature vector.

4. The report header management method based on dynamic hierarchical rendering and context awareness according to claim 3, wherein, The calculation formula of the enhanced time sequence input vector is: ； in, For the first The augmented temporal input vector at each time step; This is a fixed-dimensional regulatory cycle vector generated through a regulatory cycle coding table; For the type of regulatory cycle in the first The dynamic modulation function at time; For the first The temporal position vector at any given moment; For the first Historical data of financial indicators at any given time; Use the Sigmoid activation function; For multilayer perceptron networks; A cyclical attention mechanism for sensing the regulatory cycle; It is the element-wise Hadamard product.

5. The method for report header management based on dynamic hierarchical rendering and context awareness as claimed in claim 1 wherein, The dual graph convolution network comprises a main graph construction module and a constraint graph construction module, the main graph construction module taking financial indicators as nodes and business association relationships between the financial indicators as edges to generate a main graph structure; The constraint graph construction module takes financial indicators as nodes and accounting equation constraint relationships of the financial indicators as edges to generate a constraint graph structure; The constraint-aware message passing function performs data transmission on the main graph structure and the constraint graph structure, fuses the main graph data and the constraint graph data, calculates the constraint default level, adjusts the graph topology when the constraint default level exceeds a preset threshold, and outputs an adjusted indicator relationship weight matrix.

6. The method for report header management based on dynamic hierarchical rendering and context awareness according to claim 5, wherein, The constraint-aware message passing function comprises a main graph data calculation unit and a constraint graph data calculation unit, the main graph data calculation unit calculates business correlation data according to node features and an adjacency matrix, the constraint graph data calculation unit calculates constraint penalty data according to accounting equation constraints and node states, and the business correlation data and the constraint penalty data are fused through weighted summation to calculate a constraint violation degree; The graph topology is adjusted by modifying edge weights, the weight of a corresponding edge is increased when the constraint violation degree is higher than a threshold value, and the weight of the corresponding edge is reduced when the constraint violation degree is lower than the threshold value, and after multiple rounds of iteration and update, an adjusted index relationship weight matrix is output.

7. The method for report header management based on dynamic hierarchical rendering and context awareness according to claim 6, wherein, The calculation formula of the constraint violation degree is: ； ； in, To constrain the degree of breach of contract in dual graph collaboration; This is a function for cross-graph attention mechanisms; Here is the conflict metric function; and These are the main diagram structure and the constraint diagram structure, respectively. For nodes With constraint edges The correlation strength coefficient; The first output of the main graph data calculation unit Data related to each business; The first output of the constraint graph data calculation unit Individual constraint and penalty data; and These are the expectation and variance functions, respectively; It is a semantic similarity function; It is the numerical stability constant; and and are the number of nodes in the main graph and the number of edges in the constraint graph, respectively; For the first in the graph structure One node; For the first in the graph structure Edge.

8. The method for report header management based on dynamic hierarchical rendering and context awareness as claimed in claim 1 wherein, The financial indicator historical data is obtained, and a supervision cycle type is identified according to a current time node, comprising: Raw financial indicator data is obtained from multiple data sources, data quality detection is performed on the raw financial indicator data, abnormal values and missing values are removed, standardized financial indicator data is generated through data standardization processing, and the financial indicator historical data is output; The current system time is obtained, time features are extracted, the time features comprise date, month and quarter information, the time features are matched according to a preset supervision time node rule base, the supervision cycle boundary to which the current time node belongs is determined, and the supervision cycle type is output, the supervision cycle type comprises a daily period, a month-end period, a quarter-end period and a year-end period.

9. The method for report header management based on dynamic hierarchical rendering and context awareness as claimed in claim 1 wherein, The time sequence weight feature vector and the index relationship weight matrix are input into a neural-symbol fusion network, and a final display weight value is output, comprising: The neural-symbol fusion network comprises a symbolic reasoning layer and a supervision compliance verification module; The time sequence weight feature vector and the index relationship weight matrix are input into the symbolic reasoning layer, numerical weight features are mapped into conditional-conclusion type logical rules through a neural network to symbolic rule converter, a symbolic representation of weight decision is established, and a logical rule set is output; The logical rule set is input into the supervision compliance verification module, the logical rules are checked for compliance according to a preset supervision compliance knowledge base, rules that do not meet supervision requirements are modified or removed, compliant logical rules are converted into numerical weights through a rule execution engine, and the final display weight value is output.

10. The method for report header management based on dynamic hierarchical rendering and context awareness as claimed in claim 1 wherein, The financial indicators are sorted according to the final display weight value, a dynamic hierarchical strategy is adopted to place high-weight indicators in a priority level and low-weight indicators in a secondary level, a hierarchical rendered report header is output, comprising: The final display weight value is sorted in descending order according to numerical size, a weight threshold value is set as a division standard, the financial indicators are divided into priority level indicators and secondary level indicators according to the weight threshold value, a hierarchical identification mapping table is established to record the hierarchical attribution of each financial indicator, and a hierarchical indicator list is output; A report header rendering structure is constructed according to the hierarchical indicator list, the priority level indicators are set as a default display state, and the secondary level indicators are set as a folded hidden state, a hierarchical expansion or contraction operation is triggered through a dynamic rendering engine according to a user interaction event, and the hierarchical rendered report header is output.

Images