Personalized report configuration platform system

Abstract

This invention relates to the field of computer data processing and data visualization technology, specifically a personalized report configuration platform system, comprising: a data virtualization module: establishing physical data source mapping and constructing atomic data pools; an indicator graph construction module: generating atomic indicator dependency graphs, storing logical operation relationships rather than calculation results; a context-aware module: capturing environmental parameters such as tenant roles and time windows; a dynamic compilation engine: injecting environmental parameters into the graph and instantiating and parsing logical nodes to generate structured query instructions; and a rendering and compositing module: executing instructions to obtain result data and mapping it to view components. This invention achieves dynamic instantiation of logic defined once in multiple places, avoiding hard-coding development and improving the maintainability and response speed of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of computer data processing and data visualization, and specifically to a personalized report configuration platform system. Background Art

[0002] In the current medical informatization and multi-tenant report application scenarios, heterogeneous physical data sources such as the Hospital Information System (HIS) and Electronic Medical Record System (EMR) within a hospital have the characteristics of multi-modal, large differences in schema definitions, and decentralized storage; at the same time, different departments have highly personalized requirements for report indicators, and query requests are often in a high-concurrency situation; to meet such requirements, existing technologies generally adopt traditional ETL architectures or hard-coded models, that is, pre-develop fixed reports and migrate all heterogeneous data to an intermediate database.

[0003] However, the above solutions have significant drawbacks: they rely on heavy physical data handling and redundant storage, and the indicator logic is strongly coupled with the underlying database dialects, resulting in difficulties in cross-source data alignment for the system; in the face of interactive filtering or high-frequency queries, existing mechanisms usually perform undifferentiated full recalculations, causing serious waste of computing resources and page response delays, and it is difficult to support real-time dynamic analysis and agile changes of a large number of indicators. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a personalized report configuration platform system, which is applied to the real-time linkage monitoring and clinical decision-making assistance scenarios of hospital HIS and EMR data. Specifically, the technical solutions of the present invention include: A data virtualization module for establishing a mapping relationship between medical heterogeneous physical data sources and the logical layer, mapping physical fields to standardized atomic data elements, and constructing an atomic data pool; wherein, the atomic data elements include admission time, discharge time, diagnosis code, and expense details; An indicator graph construction module that interacts with the data virtualization module to obtain the atomic data elements, defines business indicators through declarative logic, and generates an atomic indicator dependency graph; wherein, the atomic indicator dependency graph is a directed acyclic graph structure that stores the logical operation relationships between business indicators rather than calculation results; A context awareness module for capturing the access environment parameters of the current session in real time during runtime, and the access environment parameters at least include a tenant role identifier corresponding to a specific department of the hospital and a time window constraint; The dynamic compilation engine is integrated with the indicator graph construction module and the context-aware module through the instruction stream. It is used to inject the access environment parameters as environment variables into the atomic indicator dependency graph and to instantiate and parse the logical nodes in the atomic indicator dependency graph. This allows the structured query instruction to automatically carry the permission filtering conditions based on the tenant role identifier and the range retrieval conditions based on the time window constraint when the underlying data source is executed, thereby generating a structured query instruction for the underlying data source. The rendering and compositing module is configured at the output of the dynamic compilation engine to execute the structured query instructions to obtain result data and map the result data to front-end view components based on preset visualization rules.

[0005] Preferably, the indicator map construction module includes: An atomic definition unit is used to define the attribute type and source path of the atomic data element. A logical arrangement unit is used to combine multiple atomic data elements into a composite index node through preset operators; The dependency resolution unit is used to resolve the dependency path of the composite index node to the atomic data element and establish the parent-child node topology relationship in the atomic index dependency graph.

[0006] Preferred dynamic compilation engines include: The subgraph extraction unit is used to extract relevant logical subgraphs from the atomic indicator dependency graph based on the report view requested by the user. The context injection unit is used to load the corresponding differentiated filtering conditions or calculation formulas in the logical subgraph according to the tenant role identifier, so as to realize the polymorphic logical calculation of the same indicator under different tenants. The query optimization unit is used to combine the time window constraints to convert the logical subgraph after injection context into a database-executable SQL query statement.

[0007] Preferably, the system also includes: The differential calculation module is communicatively associated with the rendering and compositing module and is used to respond to changes in the filtering conditions of the front-end view component. The differential calculation module is specifically used to identify downstream nodes affected by the change in the screening conditions based on the topological sorting of the atomic index dependency graph, and to trigger recalculation instructions only for the affected nodes, while maintaining the cached state of the unaffected nodes.

[0008] Preferably, the system also includes: The intelligent attribution module, integrated with the dynamic compilation engine, is used to perform path traversal analysis on the atomic indicator dependency graph to generate trend interpretation text in natural language form. The intelligent attribution module is configured to backtrack along the atomic index dependency graph to locate the leaf node or intermediate calculation node that caused the numerical fluctuation when a numerical fluctuation is detected.

[0009] Preferably, the intelligent attribution module includes: The fluctuation detection unit is used to calculate the month-on-month change rate of the target indicator in the current time window and the historical time window; The path traversal unit is used to identify the child nodes constituting the target index when the month-on-month change rate is greater than a preset threshold; for child nodes with linear aggregation relationship, the contribution is calculated based on the ratio of the numerical change to the total change; for child nodes with nonlinear composite relationship, the contribution is calculated based on the product of the partial derivative of the child node with respect to the target index and the numerical change of the child node. The text generation unit is used to extract the semantic description of the child node with the highest contribution and combine it with the month-on-month change rate to synthesize the trend interpretation text.

[0010] Preferably, the rendering and compositing module includes: A density evaluation unit is used to calculate the information density of the result data, wherein the information density is defined as the ratio of the number of data records of the result data to the number of physical pixels of the current display device along the rendering axis. The component adaptation unit is configured to execute the following logic: if the information density is less than or equal to a preset density threshold, select a graphical chart component for rendering; if the information density is greater than the preset density threshold, perform data pruning or automatically switch to a statistical summary table component for rendering.

[0011] Preferably, the data virtualization module supports heterogeneous physical data sources in the medical field, including relational databases, time-series databases, and unstructured document repositories.

[0012] Preferably, the dynamic compilation engine is also used for: It handles cross-source query requests and performs real-time alignment and aggregation calculations on data streams from different physical data sources in memory. The rendering and compositing module is used to update the view within a preset response time threshold after receiving the calculated results of the aggregation.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This system maps physical fields to standardized atomic data elements through a data virtualization module, and uses an indicator graph construction module to generate an atomic indicator dependency graph that only stores logical operation relationships, avoiding redundant data storage; combined with a context-aware module and a dynamic compilation engine, the system injects tenant role identifiers and time window constraints as environment variables into the graph, instantiates and parses logical nodes, and generates structured query instructions for the underlying source, realizing the one-time definition of atomic indicator logic and dynamic reuse in multiple scenarios; 2. This system introduces a differential calculation module to optimize response performance. This module is based on the topological sorting of the atomic index-dependent graph and accurately identifies the affected downstream nodes when the screening conditions change. The system only triggers recalculation instructions for the affected nodes and maintains the cached state of the unaffected nodes, thereby avoiding a full data refresh and significantly reducing the consumption of computing resources. 3. Through the intelligent attribution module, the system performs path traversal analysis and reverse backtracking on the atomic indicator dependency graph to locate the leaf or intermediate node that causes numerical fluctuations; the module calculates the contribution of child nodes based on linear aggregation or nonlinear composite relationships and automatically generates trend interpretation text in natural language form, realizing automated insight from numerical fluctuations to business semantics. Attached Figure Description

[0014] Figure 1 This is a structural diagram of the system of the present invention; Figure 2 This is a flowchart of the indicator graph construction module. Figure 3 This is a flowchart of the intelligent attribution module. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0016] Example 1: Please see Figure 1 The personalized report configuration platform system is applied to the real-time linkage monitoring and clinical auxiliary decision-making scenarios of hospital HIS and EMR data. The system includes: a data virtualization module, which is used to establish the mapping relationship between heterogeneous physical data sources and logical layers in medical care, map physical fields into standardized atomic data elements, and build an atomic data pool. The atomic data elements include at least admission time, discharge time, diagnosis code, and cost details. The indicator graph construction module interacts with the data virtualization module to obtain atomic data elements, defines business indicators through declarative logic, such as departmental resource consumption and outpatient volume, and generates atomic indicator dependency graphs. The atomic indicator dependency graph is a directed acyclic graph structure that stores the logical operation relationships between business indicators rather than the calculation results. The context-aware module is used to capture the access environment parameters of the current session in real time at runtime. The access environment parameters include at least the tenant role identifier corresponding to the specific department of the hospital and the time window constraint. The dynamic compilation engine integrates with the indicator graph construction module and the context-aware module through the instruction stream. It is used to inject access environment parameters as environment variables into the atomic indicator dependency graph and to instantiate and parse the logical nodes in the atomic indicator dependency graph. This allows the structured query instructions to automatically carry permission filtering conditions based on tenant role identifiers and range retrieval conditions based on time window constraints when the underlying data source is executed, thereby generating structured query instructions for the underlying data source. The rendering and compositing module is configured at the output of the dynamic compilation engine. It is used to execute structured query instructions to obtain result data and map the result data to front-end view components based on preset visualization rules.

[0017] This embodiment details the overall architecture and data flow logic of the system, aiming to resolve the contradiction between medical data silos and the personalized needs of multiple departments. The system starts the data virtualization module, which acts as middleware between the logical layer and the physical layer, connecting to the underlying heterogeneous database through an adapter. During this process, the module does not migrate data, but establishes metadata mapping, defining the columns in the physical table as atomic data elements, thereby shielding the differences in the underlying database schema. Based on these atomic data elements, the indicator graph construction module constructs an atomic indicator dependency graph in response to the user's declarative input. This graph exists in the form of a directed acyclic graph, which only records the logical path of how to calculate, without storing the calculated value. This design significantly reduces storage overhead and ensures the atomicity of indicator definition. When a user initiates a report access request, the context-aware module immediately captures the current runtime environment and extracts the tenant role identifier and time window constraints. These parameters are passed to the dynamic compilation engine, which treats the logical nodes in the graph as uninstantiated classes, injects environment variables into them, and generates structured query instructions for specific database dialects. The rendering and compositing module executes these instructions, obtains the raw result data, and automatically matches the front-end view components based on the data characteristics to complete the final page rendering. This embodiment achieves one-time logic definition and multiple dynamic instantiations in a multi-tenant medical scenario through the collaborative work of a dynamic compilation engine and indicator graph. This architecture avoids the heavy work of developing hard-coded reports for each department, and since the graph only stores logical relationships, the system can handle the management of massive indicators with extremely low resource consumption, significantly improving the maintainability and response speed of the system.

[0018] Example 2: The indicator graph construction module includes: an atomic definition unit, used to define the attribute type and source path of atomic data elements; and a logical arrangement unit, used to combine multiple atomic data elements into composite indicator nodes through preset operators. The dependency resolution unit is used to resolve the dependency paths of composite index nodes to atomic data elements and establish the parent-child node topology in the atomic index dependency graph.

[0019] This embodiment further refines the internal processing flow of the indicator map construction module, such as... Figure 2 As shown, the core is to realize the automated transformation from discrete data elements to structured logic graphs; the atomic definition unit provides a metadata access interface, where operators can define the physical attributes of atomic data elements, such as integer, floating-point, string, date and time, etc. This step ensures that every leaf node in the graph can be accurately traced back to a specific field in the physical data source, establishing an anchor point between the logical world and the physical world; Based on this, the logic orchestration unit provides an operator library, allowing users to combine atomic data elements into composite index nodes by dragging and dropping or using scripts, using addition, subtraction, multiplication, division or aggregation operators; the preset operators are encapsulated at the system level as computational primitives with deterministic mathematical definitions to eliminate ambiguity at the implementation level; Specifically, for binary arithmetic operators, the system has built-in null value circuit breaking and exception protection logic. For example, the addition operator is defined as: Among them, parameters and These represent the instantiation values ​​of the left and right operand nodes in the current context, respectively. The function is used to return the first non-empty value in its argument list, and the formula is obtained by... The function ensures that any operand that is NULL is automatically converted to 0 for calculation; the division operator is defined as: Among them, parameters The value of the dividend node, parameter The value of the divisor node. The function is used to return a null value when two specified expressions are equal; the formula utilizes... The function forces a return value of NULL when the division by zero, thereby preventing the database from throwing a division by zero exception, thus ensuring that no manual handling of basic boundary conditions is required when writing the logic code; For aggregation operators, the system defines a target field. With filtering predicates The operational structure of conditional summation, taking it as an example, has the following specific mathematical operation definition: in, For the data record set of the current computing context, Representing records target field The value, Representative predicate After filtering, for the target field The final numerical result obtained by performing conditional summation; based on the above-mentioned explicit mathematical operation definitions, those skilled in the art can directly construct the corresponding operator class library; the dependency parsing unit intervenes and performs recursive parsing on each newly created composite index node; this unit traverses the calculation formula of the index, automatically identifies its directly dependent child nodes, and traces down the dependency chain to the leaf node, thereby constructing a complete parent-child node topology; this process automatically detects circular dependencies and reports errors, ensuring that the generated graph strictly conforms to the requirements of a directed acyclic graph structure; This embodiment eliminates the error-proneness of manually maintaining indicator relationships through an automated dependency resolution mechanism. When faced with complex business systems with hundreds or thousands of indicators, this module can automatically sort out clear calculation lineage relationships, providing a solid topological data foundation for subsequent automated attribution analysis and change impact assessment.

[0020] Example 3: The dynamic compilation engine includes: a subgraph extraction unit, used to extract relevant logical subgraphs from the atomic metric dependency graph based on the report view requested by the user; a context injection unit, used to load corresponding differential filtering conditions or calculation formulas into the logical subgraphs based on the tenant role identifier, so as to realize polymorphic logical calculation of the same atomic metric under different tenants; and a query optimization unit, used to combine time window constraints to convert the context-injected logical subgraphs into SQL query statements that can be executed by the database.

[0021] This embodiment reveals in detail the on-demand computing mechanism of the dynamic compilation engine, which aims to solve the performance bottleneck problem caused by full graph calculation. In response to the report request initiated by the front end, the subgraph extraction unit does not load the entire indicator graph. Instead, it takes the target indicator involved in the request as the starting point, performs a reverse traversal in the graph, and extracts only the upstream nodes necessary for calculating these target indicators to construct a lightweight logical subgraph. The context injection unit reads the current tenant role identifier and injects differentiated filtering conditions into specific nodes in the logical subgraph according to the preset rule table. For example, for the indicator of department resource consumption, if the role is identified as cardiology, the system operates on the abstract syntax tree (AST) of the logical subgraph, uses the visitor pattern to locate the table scan operator of all leaf nodes, and inserts a filtering operator node containing the condition "department ID = cardiology" at its direct parent position, instead of simple text concatenation. This ensures that the logical structure completes the permission constraints at the underlying data reading stage, preventing data overreach and redundant loading. Furthermore, for the differentiated calculation formulas mentioned in the embodiments, the context injection unit adopts an AST subtree replacement strategy; when the traverser identifies an indicator node with polymorphic attributes, the system retrieves the corresponding calculation logic based on the tenant configuration table; The engine dynamically constructs the corresponding expression subtree and replaces the indicator nodes in the original abstract syntax tree. This process occurs before SQL generation, ensuring that different tenants execute completely different arithmetic operation logic when accessing the same indicator name, thereby achieving deep polymorphism of indicator logic. Specifically, the context injection unit maintains a tenant-variable mapping table; when the AST traverser accesses a node with an environment variable tag, it looks up the mapping table and replaces it with a specific constant literal node. In addition, to prevent runtime errors caused by mismatch between injected variables and physical field types, the context injection unit has built-in type enforcement and adaptation logic; before performing AST node replacement, the system reads the type definition of the target field in the metadata pool; If the environment variable is a string, the system automatically triggers a dictionary lookup, retrieving the corresponding numeric code from the cached dictionary table and constructing the replacement node as a numeric literal. If there is no dictionary lookup, it automatically wraps the variable with the CAST or TRY_CAST function operator to ensure that the generated SQL statement conforms to strong type constraints. The CAST operator is used to forcibly convert one built-in data type to another to ensure type compatibility of the generated SQL statement. The TRY_CAST operator attempts to convert an expression to a specified data type. If the conversion is successful, it returns the converted value; if the conversion fails, it returns NULL, thus avoiding query execution interruption due to underlying data anomalies. The query optimization unit, combined with time window constraints, compiles the logical subgraph after context injection. This unit traverses the AST using the visitor pattern and applies predicate pushdown and column pruning techniques to convert the logical graph into SQL query statements optimized for a specific database. When generating SQL, the adapter not only performs field name replacement but also serializes the data according to the target database's syntax template, especially performing precise dialect conversion for date functions and pagination syntax to ensure that the generated SQL statements can be directly executed in the target physical database without syntax errors. This embodiment achieves fine-grained allocation of computing resources through subgraph extraction and context injection techniques. In the high-concurrency query scenario during the morning rush hour in hospitals, this mechanism ensures that the system only calculates the data required for the current view and can automatically present the corresponding data slices based on the doctor's department, greatly improving query response speed and data security.

[0022] Example 4: The system also includes: a differential calculation module, which communicates and associates with the rendering and compositing module to respond to changes in the filtering conditions of the front-end view components; The differential calculation module is specifically used for topological sorting based on atomic index-dependent graphs, identifying downstream nodes affected by changes in screening conditions, and triggering recalculation instructions only for affected nodes, while maintaining the cached state of unaffected nodes.

[0023] This embodiment introduces a differential calculation module, which focuses on optimizing system performance in interactive analysis scenarios. The system is in a listening state. When the user modifies the filtering conditions on the front-end view component, the differential calculation module immediately captures the change operation. Based on the topological sorting of the atomic index dependency graph, the module performs downstream traversal starting from the change point, accurately identifies all downstream nodes that depend on the change condition, and marks these nodes as dirty. The system generates recalculation instructions only for nodes marked as dirty, calling the dynamic compilation engine to regenerate the query; for nodes in the graph that are unrelated to the changed path, the system directly reuses the cached data in memory, skipping the database query step; the rendering and compositing module merges the newly calculated data with the cached data and refreshes the front-end view; This embodiment utilizes the topological characteristics of the graph to achieve intelligent cache updates. In complex dashboard scenarios, when users frequently switch filtering dimensions, this mechanism avoids indiscriminate re-refreshing of full-screen data, reducing the system's interaction latency from seconds to milliseconds, and significantly improving the user's exploratory analysis experience.

[0024] Example 5: The system also includes: an intelligent attribution module, which is integrated with the dynamic compilation engine to perform path traversal analysis on the atomic indicator dependency graph to generate trend interpretation text in natural language form; and an intelligent attribution module configured to backtrack along the atomic indicator dependency graph to locate the leaf node or intermediate calculation node that caused the numerical fluctuation when a numerical fluctuation of the target indicator is detected. The intelligent attribution module includes: a fluctuation detection unit, used to calculate the month-on-month change rate of the target indicator in the current time window and the historical time window; The path traversal unit is used to identify the child nodes constituting the target indicator when the month-on-month change rate is greater than a preset threshold; for child nodes with linear aggregation relationships, the contribution is calculated based on the ratio of the numerical change to the total change; for child nodes with nonlinear composite relationships, the contribution is calculated based on the product of the partial derivative of the child node with respect to the target indicator and the numerical change of the child node; the text generation unit is used to extract the semantic description of the child node with the highest contribution and combine it with the month-on-month change rate to synthesize the trend interpretation text.

[0025] This embodiment describes in detail the algorithm principle of the intelligent attribution module. Please refer to [link / reference]. Figure 3 The aim is to transform black-box data fluctuations into interpretable business insights; the fluctuation detection unit continuously monitors key indicators and defines the month-on-month change rate. ;like and Then a preset non-zero smoothing factor is introduced. ,in For the smallest positive number that approaches zero, for example, a value of The year-on-year change rate at this point is defined as Therefore, based on The actual size is scaled up proportionally to accurately preserve the magnitude difference characteristics of data when it surges or slightly increases from zero, avoiding the loss of magnitude information; if and Then define ;like and If the signs are opposite, then take the denominator as . To uniformly measure the magnitude of change and reflect the true growth or decline trend; if The formula to be executed is: in, The indicator value for the current time window; The indicator value is a historical time window value; in response to When the absolute value exceeds a preset threshold, the path traversal unit is activated. Specifically, this preset threshold is not obtained by fixed assignment, but by dynamic calculation based on the principle of statistical process control: the system extracts historical data of the target indicator over the past N periods and calculates its mean. with standard deviation ; The preset threshold is set to ,in, The sensitivity coefficient, with a value ranging from 2 to 3; based on the distribution of typical medical outpatient volume data, when The empirical coefficient of variation calculated at that time is usually in the range of 0.15 to 0.20. This is used as a criterion to ensure that only statistically significant abnormal fluctuations are captured. The path traversal unit executes differentiated attribution logic based on the operation type of the target indicator in the graph: For linear aggregation nodes, i.e., addition or subtraction nodes: this unit backtracks along the graph to obtain all direct child nodes of the target indicator and calculates the contribution of each child node. ;like Then define the contribution of the child node. ;like The formula is as follows: in, child node The current value; child node The historical value. This formula directly uses the ratio of the change in the value of the child node to the change in the total value of the target indicator, thus eliminating the sign correction term that may lead to weight bias; For nonlinear composite nodes, i.e., multiplication or division nodes: given the failure of the direct difference method under nonlinear relationships, the system performs partial derivative approximation decomposition logic; that is, it utilizes historical values. The partial derivative at a given point is used as a sensitivity coefficient, multiplied by the variation of the child nodes; let the node relationship be... ,in Representing the variables of each child node, the system first calculates the absolute contribution of each child node. The calculation formula is as follows: in, This represents the defined functional relationship between the target indicator and each child node, i.e. ; This represents a vector consisting of all child node variables; This indicates that all child nodes take the historical value. In the state of the function For specific child nodes The partial derivatives; Define the contribution of each child node in a nonlinear scenario. For: If Then define ;like The formula is as follows: To address the problem of calculating partial derivatives for arbitrary user-defined formulas, the path traversal unit, as a lower-level component of the intelligent attribution module, integrates a symbolic differential engine within its internal architecture. This engine is based on the AST stored in the atomic index dependency graph and employs a post-order traversal algorithm. When an operator node is accessed, the engine follows the chain rule and a pre-defined operator derivative rule library, which fully incorporates division. power functions The engine defines the derivatives of fundamental operators such as exponential, logarithmic, and trigonometric functions, and dynamically constructs a derivative expression tree. For operators without predefined rules or user-defined functions via scripts, the symbolic differentiation engine is equipped with a numerical differentiation downgrade mechanism: when the AST traverses to an unknown operator node, the engine pauses symbolic parsing and automatically switches to the central difference algorithm, using the formula: in, Values Alternatively, the engine can dynamically adjust based on machine accuracy, calculate the local derivative value of the node, and use the result as a constant node to fill the derivative expression tree, thereby ensuring the completeness of partial derivative calculations for any formula; the engine will store the historical values ​​of each child node. By injecting this derivative expression tree, the accurate partial derivative values ​​are calculated through constant folding, thus avoiding the truncation error introduced by the numerical difference method. Furthermore, to address the Taylor expansion residuals that may exist in the nonlinear approximation calculation, the system performs residual attribution correction: calculating the model predicts the residuals. The formula is defined as: in, This represents the actual total change in the target indicator; This indicates that for all child nodes absolute contribution The summation represents the total change predicted by the model.

[0026] like The 5% correction threshold is a system-preset default parameter, obtained by statistically analyzing the actual residual distribution generated after approximating the target index with first-order partial derivatives in historical data test sets. This threshold is used to distinguish between nonlinear residuals and systematic computational noise. In actual configuration, the system supports dynamically increasing this threshold based on the specific operator type contained in the composite node to accommodate its larger nonlinear truncation error. The system automatically creates a virtual child node named "multi-factor coupling effect" to handle this difference. ; The text generation unit performs semantic template matching logic; this unit maintains a semantic mapping table based on symbolic logic. ,in The sign of the total rate of change. Symbol for the contribution of child nodes: like and The matching template is positively driven by the increase in node names; like and The matching template attempted to offset the increase despite the decline in [node name]; like and The main reason for matching the template is the shortening of the node name; like and Although the number of matching template node names has rebounded somewhat, it is not enough to reverse the downward trend; Text generation unit selection The highest-ranking Top-1 child node, combined with the above logic, generates the final trend interpretation text, thereby ensuring strict consistency between the generated natural language conclusions and mathematical logic and business intuition.

[0027] Example 6: The rendering and compositing module includes: a density evaluation unit, used to calculate the information density of the result data, which is defined as the ratio of the number of data records in the result data to the number of physical pixels on the current display device along the rendering axis; The component adaptation unit is configured to execute the following logic: if the information density is less than or equal to the preset density threshold, select the graphical chart component for rendering; if the information density is greater than the preset density threshold, perform data pruning or automatically switch to the statistical summary table component for rendering.

[0028] This embodiment illustrates the adaptive display logic in the rendering and compositing module, aiming to solve the visual congestion problem on mobile devices when displaying massive amounts of data; the density evaluation unit immediately calculates the information density after receiving the query results. The formula is as follows: in, This represents the number of rows in the result set. This is the physical pixel width of the current display area along the rendering axis. The component adaptation unit will The density is compared with a preset density threshold, typically set between 0.5 and 0.8; in response to If the value is less than or equal to the threshold, the system calls the graphical chart component for rendering; Conversely, in response to If the value exceeds a threshold, the system identifies a potential risk of visual confusion and automatically triggers a degradation strategy. To eliminate the uncertainty between pruning operations and table switching, the component adaptation unit further calculates the coefficient of variation of the data sequence. : like This indicates that the data is stabilizing, and the system will prioritize switching to the statistical summary table component, displaying only the minimum / maximum / average statistical values ​​and paginated details, in order to avoid rendering thousands of rows of meaningless tiled tables on mobile devices; like This indicates a significant fluctuation trend. The system maintains the graphical rendering mode and performs data pruning operations, specifically using the maximum triangle three-bucket algorithm. The specific execution steps of the LTTB algorithm are as follows: sort the data sequence Divided into For each bucket, the first and last points are fixed and retained; for the middle bucket... Calculate the next bucket The geometric center of all points Let the previous bucket be... The points already selected are ; Coordinate normalization mapping: To prevent timestamp values ​​from being displayed on the X-axis, such as... The significant difference in dimensions between the index values ​​on the Y-axis and the actual values ​​leads to an imbalance in the area calculation weights. Therefore, the system pre-maps all data points within the current processing window to a unit square using a linear normalization method. In the coordinate system; let the mapped coordinates be... ; Traverse the current bucket Each data point in Calculate from point , , The area of ​​the normalized triangle formed is calculated using the following formula: Select the current bucket The original point that makes the area the largest As a reserved point; This embodiment achieves deterministic adaptive degradation of visualization by introducing physical resolution constraints, coefficient of variation judgment logic, and LTTB pruning algorithm based on normalized coordinates, effectively preventing chart overlap and feature loss caused by excessively dense data points or disparate dimensions.

[0029] Example 7: The data virtualization module supports heterogeneous physical data sources in the medical field, including relational databases, time-series databases, and unstructured document repositories. The dynamic compilation engine is also used to: process cross-source query requests and perform real-time alignment and aggregation calculations on data streams from different physical data sources in memory. The rendering and compositing module is used to update the view within a preset response time threshold after receiving the aggregated calculation results.

[0030] This embodiment demonstrates the system's cross-source integration capabilities in a complex medical data environment; the data virtualization module concurrently accesses relational databases, time-series databases, and document repositories; in this scenario, the dynamic compilation engine is configured to support federated queries; the engine breaks down and distributes complex queries, and aggregates data streams in application-layer memory; The engine executes a time-bucket alignment join algorithm in memory: defining a global standard time grid sequence. Step length Aggregate high-frequency time series data to grid points. For unstructured document data, semantic feature discretization is performed: the system extracts key text based on pre-configured regularization rules; to address the problem of zero weight or overfitting caused by sample sparsity in the feature weight calculation algorithm within a short time window, i.e., the total number of documents... This equals the number of documents containing that word, resulting in The system introduces a hybrid corpus weighting mechanism: the general formula for calculating inverse document frequency (IDF) is defined as follows: in, This corresponds to the total number of documents in the corpus. For included terms The number of documents. Based on this, the weighted formula is as follows: in, The comprehensive semantic feature weights of medical terms, These are the mixed weighting coefficients, with a value range of [value range missing]. The default possible value is This is used to balance the weights of local window features and global features; Calculated based on the document set within the current query window. Based on the system's daily pre-calculation of the full historical medical record database; the system will calculate the... Compare with the preset feature extraction threshold; if If it exceeds this threshold, then the term will be... Mark the valid event features of the current time grid k and store them in the sequence to be processed; This mechanism ensures that even in the extreme window of having only one medical record, rare and critical medical terms, such as ventricular fibrillation, still receive high weight and are thus accurately extracted as event markers for that time bucket; a hash join is performed based on the patient ID and the time grid index k. To mitigate the risk of memory overflow, the dynamic compilation engine employs a semaphore-based streaming backpressure mechanism: the system maintains a bounded blocking queue with a capacity threshold set to [value missing]. For example, 50MB or 10,000 batch objects; when the number of micro-batches to be processed in memory reaches At this time, the data reading thread is forcibly blocked, pausing the retrieval of new data from the underlying database until the rendering and compositing module consumes and releases memory space; in addition, for the selection of association keys, the engine executes an automatic dimension degradation strategy, prioritizing the downgrading of aggregation operators with high cardinality dimensions to reduce the amount of data added to memory; This embodiment solves the problems of feature loss and resource out-of-control in cross-modal data alignment by using hybrid corpus IDF weighting and backpressure flow control technology, ensuring the robust operation of the system in medical big data scenarios.

[0031] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A personalized report configuration platform system, the system is applied to the real-time linkage monitoring of hospital HIS and EMR data and the clinical auxiliary decision-making scene, characterized in that, The system includes: The data virtualization module is used to establish a mapping relationship between heterogeneous physical data sources in the medical field and the logical layer, mapping physical fields to standardized atomic data elements and constructing an atomic data pool; wherein, the atomic data elements include admission time, discharge time, diagnosis code and cost details; The indicator graph construction module interacts with the data virtualization module to obtain the atomic data elements, which are used to define business indicators through declarative logic based on the atomic data elements and generate an atomic indicator dependency graph; wherein, the atomic indicator dependency graph is a directed acyclic graph structure that stores the logical operation relationships between business indicators rather than the calculation results. The context-aware module is used to capture the access environment parameters of the current session in real time at runtime. The access environment parameters include at least the tenant role identifier corresponding to the specific department of the hospital and time window constraints. The dynamic compilation engine is integrated with the indicator graph construction module and the context-aware module through the instruction stream. It is used to inject the access environment parameters as environment variables into the atomic indicator dependency graph, and to instantiate and parse the logical nodes in the atomic indicator dependency graph to generate structured query instructions for the underlying data source. This allows the structured query instructions to automatically carry permission filtering conditions based on the tenant role identifier and range retrieval conditions based on the time window constraint when executed by the underlying data source, thus generating structured query instructions for the underlying data source. The rendering and compositing module is configured at the output of the dynamic compilation engine to execute the structured query instructions to obtain result data and map the result data to front-end view components based on preset visualization rules.

2. The personalized report configuration platform system according to claim 1, characterized in that, The indicator graph construction module includes: An atomic definition unit is used to define the attribute type and source path of the atomic data element. A logical arrangement unit is used to combine multiple atomic data elements into a composite index node through preset operators; The dependency resolution unit is used to resolve the dependency path of the composite index node to the atomic data element and establish the parent-child node topology relationship in the atomic index dependency graph.

3. The personalized report configuration platform system according to claim 1, characterized in that, The dynamic compilation engine includes: The subgraph extraction unit is used to extract relevant logical subgraphs from the atomic indicator dependency graph based on the report view requested by the user. The context injection unit is used to load the corresponding differentiated filtering conditions or calculation formulas in the logical subgraph according to the tenant role identifier, so as to realize the polymorphic logical calculation of the same atomic index under different tenants. The query optimization unit is used to combine the time window constraints to convert the logical subgraph after injection context into a database-executable SQL query statement.

4. The personalized report configuration platform system according to claim 1, characterized in that, The system also includes: The differential calculation module is communicatively associated with the rendering and compositing module and is used to respond to changes in the filtering conditions of the front-end view component. The differential calculation module is specifically used to identify downstream nodes affected by the change in the screening conditions based on the topological sorting of the atomic index dependency graph, and to trigger recalculation instructions only for the affected downstream nodes, while maintaining the cached state of the unaffected nodes.

5. The personalized report configuration platform system according to claim 1, characterized in that, The system also includes: The intelligent attribution module, integrated with the dynamic compilation engine, is used to perform path traversal analysis on the atomic indicator dependency graph to generate trend interpretation text in natural language form. The intelligent attribution module is configured to backtrack along the atomic index dependency graph to locate the leaf node or intermediate calculation node that caused the numerical fluctuation when a numerical fluctuation is detected.

6. The personalized report configuration platform system according to claim 5, characterized in that, The intelligent attribution module includes: The fluctuation detection unit is used to calculate the month-on-month change rate of the target indicator in the current time window and the historical time window; The path traversal unit is used to identify the child nodes constituting the target index when the month-on-month change rate is greater than a preset threshold; for child nodes with linear aggregation relationship, the contribution is calculated based on the ratio of the numerical change to the total change; for child nodes with nonlinear composite relationship, the contribution is calculated based on the product of the partial derivative of the child node with respect to the target index and the numerical change of the child node. The text generation unit is used to extract the semantic description of the child node with the highest contribution and combine it with the month-on-month change rate to synthesize the trend interpretation text.

7. The personalized report configuration platform system according to claim 1, characterized in that, The rendering and compositing module includes: A density evaluation unit is used to calculate the information density of the result data, wherein the information density is defined as the ratio of the number of data records of the result data to the number of physical pixels of the current display device along the rendering axis. The component adaptation unit is configured to execute the following logic: if the information density is less than or equal to a preset density threshold, select a graphical chart component for rendering; if the information density is greater than the preset density threshold, perform data pruning or automatically switch to a statistical summary table component for rendering.

8. The personalized report configuration platform system according to claim 1, characterized in that, The data virtualization module supports heterogeneous physical data sources in the medical field, including relational databases, time-series databases, and unstructured document repositories.

9. The personalized report configuration platform system according to claim 8, characterized in that, The dynamic compilation engine is also used for: It handles cross-source query requests and performs real-time alignment and aggregation calculations on data streams from different physical data sources in memory. The rendering and compositing module is used to update the view within a preset response time threshold after receiving the calculated results of the aggregation.

Images