A report document generation method and system based on dynamic data driving

Abstract

The application discloses a kind of report document generation method and system based on dynamic data driving, its method includes: establishing communication connection with dynamic data source, real-time acquisition structured and unstructured data stream according to mixed monitoring strategy, after denoising and normalization processing, store into buffer queue generation dynamic data stream and carry out multidimensional feature extraction, through template decision model matching the report structure template suitable for;According to the content slot of template, call data processing component, convert data stream into candidate content segment;According to chapter level assembly, rendering generates target report document, and establishes field level dynamic binding relationship with data source for updatable element;Continuously monitor data update event, locate affected element based on binding relationship, regenerate partial content and incrementally replace to document corresponding position.It realizes the full-process automation of report document from data acquisition, template matching, content generation to dynamic updating, significantly reduces manual maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of intelligent report generation technology, and in particular to a method and system for generating report documents based on dynamic data. Background Technology

[0002] Currently, in data-intensive enterprises, government agencies, and research institutions, regular or ad-hoc data reports (such as daily operation and maintenance reports, financial analysis reports, weekly market monitoring reports, and equipment status assessment reports) are important carriers supporting management decisions and business monitoring. Traditional methods of generating these reports mainly rely on manual data collection, manual writing using office software (such as Word and Excel), or batch generation using semi-automated tools with fixed templates.

[0003] However, existing technologies have the following obvious shortcomings in practical applications:

[0004] First, report templates are fixed and cannot adapt to changes in data characteristics. Traditional report generation methods typically use pre-designed static templates. Once the chapter structure, content granularity, and analytical perspective of the template are fixed, they are difficult to adjust. When the field composition, data distribution, or business focus of the data source changes, the original static template cannot adapt automatically and requires manual redesign of the template, resulting in high maintenance costs and poor flexibility of the report generation system.

[0005] Second, data and document updates are out of sync, resulting in poor real-time performance. In dynamic data environments, data sources are constantly changing, while traditionally generated report documents are one-time snapshots that cannot reflect the latest state of the data. To update the report, the entire data extraction, processing, and formatting process must be re-executed to generate a completely new document. This not only incurs high computational costs but also results in the loss of user annotations, comments, or view positions on the original document, making it impossible to achieve partial, incremental, real-time updates.

[0006] Third, the generation process lacks automation and intelligence, relying on manual intervention. While existing semi-automated tools can populate data based on templates, they require manual specification of data sources, template selection, and configuration of the mapping relationship between data fields and document locations. When the data volume is large, from multiple sources, and updated frequently, the workload of manual configuration is enormous and prone to errors. Furthermore, there is a lack of automatic identification capabilities for data flow characteristics, making it difficult to automatically recommend or switch to the most suitable report structure based on data content.

[0007] Fourth, the granularity of content updates is too coarse, resulting in significant resource waste. When only a few fields in the source data change, traditional methods often require regenerating the entire report or chapter, failing to achieve precise incremental replacement at the field or element level. This not only consumes a large amount of computing and I / O resources but also causes noticeable flickering or content jumps during document display, impacting the user experience. Summary of the Invention

[0008] To address the problems mentioned above, this invention provides a report document generation method and system based on dynamic data to solve the problems mentioned in the background art, such as report lag, template rigidity, maintenance difficulties, and low update efficiency.

[0009] A method for generating report documents based on dynamic data-driven approaches includes the following steps:

[0010] Establish a communication connection with the dynamic data source, acquire the structured and unstructured data streams of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured and unstructured data streams into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data streams;

[0011] Multidimensional feature extraction is performed on the dynamic data stream in the data buffer queue to generate feature vectors that reflect data attributes, statistical distribution and time series change trends. The feature vectors are then input into a trained template decision model to dynamically match a report structure template that is compatible with the dynamic data stream.

[0012] Based on the content slots and content types in the report structure template, the corresponding data processing components in the rule base are called to transform the dynamic data stream into candidate content fragments corresponding to each slot.

[0013] The candidate content fragments are assembled according to the chapter hierarchy of the report structure template, formatted and rendered to generate the target report document. At the same time, during the rendering process, field-level dynamic binding relationships are established between the updatable elements in the target report document and the dynamic data source.

[0014] While the target report document is in a display or storage state, continuously listen for data update events from dynamic data sources, locate the affected document elements based on field-level dynamic binding relationships, drive the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document.

[0015] Preferably, the step of establishing a communication connection with the dynamic data source, acquiring the structured and unstructured data streams of the dynamic data source in real time according to a hybrid monitoring strategy, storing the structured and unstructured data streams into a data buffer queue according to time series, and performing noise reduction and format normalization processing to generate a dynamic data stream includes:

[0016] Determine the database type of the dynamic data source and configure the corresponding data source connector based on the database type;

[0017] Determine the real-time metrics and load status of the dynamic data source, and based on the real-time metrics and load status, support the listening types. Configure a hybrid listening strategy according to the supported listening types, and use the hybrid listening strategy through the data source connector to obtain the structured and unstructured data streams of the dynamic data source in real time.

[0018] The structured and unstructured data streams are marked with timestamps respectively, and then stored in a preset data buffer queue in chronological order.

[0019] Identify and remove outliers and redundant records in structured data, and eliminate noisy text fragments and duplicate content in unstructured data;

[0020] The denoised structured data is converted into a preset first unified data format, and the denoised unstructured data is converted into a preset second unified data format, generating a dynamic data stream with a unified interface description and time sequence attributes.

[0021] Preferably, the step of performing multi-dimensional feature extraction on the dynamic data stream in the data buffer queue to generate feature vectors reflecting data attributes, statistical distribution, and temporal change trends, and inputting the feature vectors into a trained template decision model to dynamically match a report structure template adapted to the dynamic data stream, includes:

[0022] The dynamic data stream that has been denoised and normalized is obtained from the data buffer queue, and the dynamic data stream is divided into multiple sequentially arranged data windows according to the preset time window.

[0023] Multidimensional feature extraction is performed on each data window to generate feature vectors that reflect data attributes, statistical distribution, and temporal trends.

[0024] The feature vector is input into the trained template decision model, which outputs the matching probability distribution between the current data window and each template in the preset report template library.

[0025] Based on the matching probability distribution, determine the window-level matching report structure template with the highest matching degree to the current data window from the preset report template library;

[0026] The window-level matching report structure templates are arranged into a template decision sequence according to the time order of each data window. The global consistency analysis of the template decision sequence is performed by a preset template aggregation strategy to calculate the global adaptation score of each window-level matching report structure template.

[0027] Based on the global adaptation score, the final report structure template is determined from multiple window-level matching report structure templates.

[0028] Preferably, the step of calling the corresponding data processing component in the rule base to convert the dynamic data stream into candidate content fragments corresponding to each slot based on the content slot and content type of the report structure template includes:

[0029] Obtain the chapter hierarchy structure within the report structure template, determine multiple content slots based on the chapter hierarchy structure, and define the required content type for each content slot;

[0030] Based on the content type required for each content slot, the corresponding data processing component is called from the pre-set rule base according to the mapping relationship between the content type and the data processing component;

[0031] Using a pre-built semantic mapping engine, target data is aggregated from the dynamic data stream based on the semantic roles and data requirements bound to the content slot and input into the called data processing component to generate candidate content fragments corresponding to the content slot.

[0032] Each generated candidate content fragment is associated and bound to its corresponding content slot to form a slot-content mapping set.

[0033] Preferably, the process of assembling candidate content fragments according to the chapter hierarchy of the report structure template, formatting and rendering them to generate a target report document, and simultaneously establishing field-level dynamic binding relationships between updatable elements within the target report document and the dynamic data source during the rendering process, includes:

[0034] Following the chapter hierarchy of the report structure template, the process traverses downwards from the root chapter, locates the content slots contained in each chapter, and extracts candidate content fragments associated with each content slot from the slot-content mapping set.

[0035] According to the content order and layout relationship defined in the template, the extracted candidate content fragments are inserted into the document flow of the corresponding chapter to complete the content assembly of the report document;

[0036] The candidate content fragments in each content slot are applied with a preset rendering style template according to their content type. The rendering style template defines the font, paragraph spacing, color theme, chart size and alignment for natural language description, statistical charts and pivot tables respectively, and automatically numbers and generates a table of contents for each chapter to form an initial report document with a uniform format.

[0037] Identify updatable elements within the initial report document that have data dependencies on dynamic data sources, and establish field-level dynamic binding relationships for each updatable element;

[0038] A target report document containing binding metadata is generated based on field-level dynamic binding relationships. The binding metadata is embedded in the structural description of the target report document and is used to support precise incremental refresh of updatable elements during document display or storage.

[0039] Preferably, when the target report document is in a display or storage state, continuously monitoring data update events from the dynamic data source, locating affected document elements based on field-level dynamic binding relationships, and driving the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document, includes:

[0040] When the target report document is in a display or storage state, start the real-time consistency maintenance process, continuously listen for data update events of dynamic data sources, and obtain data update information including data source identifier, changed field path and changed data value;

[0041] Parse the data update information and, based on the field-level dynamic binding relationship embedded in the target report document, retrieve all document element nodes that match the data source identifier and the path of the changed field to determine the local area that needs to be updated.

[0042] For each affected document element node, based on its corresponding content generation logic configuration and original content type, the corresponding data processing component in the rule base is called, and the local content fragment is regenerated using the changed data value as input.

[0043] An incremental replacement algorithm is used to replace the newly generated local content fragments with the corresponding local areas of the target report document, while preserving the rendering state, view position, and user interaction state of the unchanged parts of the document.

[0044] Preferred options also include:

[0045] When generating the target report document, determine the data correlation index between each candidate content segment, and construct a causal dependency graph between content segments based on the data correlation index;

[0046] When data updates trigger field changes, forward propagation is performed based on the causal dependency graph to determine all related content fragments that have direct and indirect dependencies on the local content fragment;

[0047] The update cost of a local content segment is determined based on the associated content segments. If the update cost exceeds the threshold, the entire section or the entire article is regenerated.

[0048] Record user interaction behavior on each content segment of the target report document, calculate the popularity value based on the interaction behavior, and use a long short-term memory network to predict the hot reporting area in the next time window based on the popularity value. Pre-generate target candidate content segments for the hot reporting area in advance to reduce update delay.

[0049] Preferably, the process of acquiring structured and unstructured data streams from dynamic data sources in real time according to a hybrid monitoring strategy also includes:

[0050] Configure a load metric collection strategy for the dynamic data source connector. The load metric collection strategy defines the set of runtime metrics to be collected, the collection period, and the collection method.

[0051] According to the collection cycle, the raw indicator data of the runtime indicator set is periodically obtained through the indicator exposure interface built into the dynamic data source connector.

[0052] The raw indicator data is preprocessed to obtain a standardized runtime indicator sequence. Based on the standardized runtime indicator sequence, local features that characterize the current load status of the dynamic data source connector are extracted to generate a local load feature vector.

[0053] Configure a global resource indicator collection strategy for the report document generation system. The global resource indicator collection strategy defines the set of system-level resource indicators to be collected, the collection period, and the data aggregation granularity.

[0054] According to the collection cycle, the system resource probe, which interacts with the operating environment of the report document generation system, periodically obtains the raw indicator data of the system-level resource indicator set;

[0055] The acquired raw indicator data is preprocessed to obtain a standardized system resource indicator sequence. Based on the standardized system resource indicator sequence, features representing the overall system load status and resource usage patterns are extracted from the report document to generate a system resource feature vector.

[0056] The local load feature vector and the system resource feature vector are normalized and then concatenated to generate a joint state vector.

[0057] According to the preset time alignment rules, the local load feature vector and the system resource feature vector are matched to the same time window to generate paired samples;

[0058] A statistical scaling transformation method is used to map the feature values ​​of different dimensions and numerical ranges in paired samples to a preset unified numerical range, thereby obtaining normalized local load feature vectors and normalized system resource feature vectors.

[0059] The normalized local load feature vector and the normalized system resource feature vector are concatenated according to a predetermined dimensional concatenation order to generate a multidimensional joint state vector.

[0060] The multidimensional joint state vector is input as an input sequence into a preset load prediction model to extract the temporal dependency features in the input sequence and output the expected load sequence for a preset number of time windows in the future.

[0061] Based on the expected load sequence, the adjustment parameters of the hybrid listening strategy are dynamically calculated using a preset strategy to adjust the mapping relationship. The adjustment parameters include the weight ratio between polling listening and event-driven listening, as well as the polling time interval of polling listening.

[0062] The calculated weighting ratio and polling interval are sent to the acquisition module so that when the future time window arrives, polling listening and event-driven listening are allocated or switched in the listening tasks of each dynamic data source according to the weighting ratio, and polling operation is performed according to the polling interval to realize dynamic adjustment of the hybrid listening strategy.

[0063] Preferably, when generating the target report document, determining the data correlation index between each candidate content segment and constructing a causal dependency graph between content segments based on the data correlation index includes:

[0064] Record the set of dynamic data source field paths and data transformation operation chain that each candidate content fragment depends on;

[0065] For any two candidate content fragments, determine the data field overlap, data flow propagation strength, semantic logical correlation, and time sequence synchronization dependency based on the dynamic data source field path set and data transformation operation chain.

[0066] The overlap of data fields, the strength of data stream propagation, the degree of semantic and logical relevance, and the time sequence synchronization dependency are weighted and fused according to preset weights to calculate the data correlation index between any two candidate content fragments:

[0067]

[0068] in, Let represent the data correlation index between the i-th candidate content fragment and the j-th candidate content fragment. This is represented as the overlap weight of the data fields. This represents the overlap of data fields between the i-th and j-th candidate content fragments. Represented as the data stream propagation strength weight, This represents the data stream propagation strength between the i-th candidate content fragment and the j-th candidate content fragment. Represented as semantic logical relevance weight, Let represent the semantic logical correlation between the i-th candidate content fragment and the j-th candidate content fragment. Represented as the temporal synchronization dependency weight, It is represented as the temporal synchronization dependency between the i-th candidate content segment and the j-th candidate content segment;

[0069] Based on the comparison between the data correlation index and the preset threshold, it is determined whether to establish a directed edge between two candidate content segments, and the direction of the directed edge is determined according to the data flow direction to generate a causal dependency graph.

[0070] A report document generation system based on dynamic data-driven methods, the system comprising:

[0071] The acquisition module is used to establish a communication connection with the dynamic data source, acquire the structured and unstructured data streams of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured and unstructured data streams into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data streams.

[0072] The matching module is used to extract multi-dimensional features from the dynamic data stream in the data buffer queue, generate feature vectors that reflect data attributes, statistical distribution and time series change trends, and input the feature vectors into the trained template decision model to dynamically match the report structure template that is adapted to the dynamic data stream.

[0073] The conversion module is used to call the corresponding data processing components in the rule base to convert the dynamic data stream into candidate content fragments corresponding to each slot, based on the content slots and content types of the report structure template.

[0074] The generation module is used to assemble candidate content fragments according to the chapter hierarchy of the report structure template, format and render them to generate the target report document. At the same time, during the rendering process, it establishes field-level dynamic binding relationships between the updatable elements in the target report document and the dynamic data source.

[0075] The incremental replacement module continuously listens for data update events from dynamic data sources when the target report document is in a display or storage state. Based on field-level dynamic binding relationships, it locates the affected document elements, drives the data processing component to regenerate local content fragments for the changed data, and incrementally replaces them with the corresponding positions in the target report document.

[0076] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0077] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0078] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0079] Figure 1 A flowchart illustrating the workflow of a report document generation method based on dynamic data-driven approach provided by this invention;

[0080] Figure 2 Another workflow diagram for a report document generation method based on dynamic data-driven approach provided by the present invention;

[0081] Figure 3 This is another flowchart of a report document generation method based on dynamic data-driven approach provided by the present invention.

[0082] Figure 4 This is a schematic diagram of the structure of a report document generation system based on dynamic data driven by the present invention. Detailed Implementation

[0083] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0084] A report document generation method based on dynamic data-driven methods, such as Figure 1 As shown, it includes the following steps:

[0085] Step S101: Establish a communication connection with the dynamic data source, acquire the structured and unstructured data streams of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured and unstructured data streams into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data streams.

[0086] Step S102: Perform multi-dimensional feature extraction on the dynamic data stream in the data buffer queue to generate feature vectors that reflect data attributes, statistical distribution and time series change trends. Input the feature vectors into the trained template decision model to dynamically match the report structure template that is compatible with the dynamic data stream.

[0087] Step S103: Based on the content slots and content types of the report structure template, call the corresponding data processing components in the rule base to convert the dynamic data stream into candidate content fragments corresponding to each slot;

[0088] Step S104: Assemble the candidate content fragments according to the chapter hierarchy of the report structure template, format and render them to generate the target report document. At the same time, during the rendering process, establish field-level dynamic binding relationships between the updatable elements in the target report document and the dynamic data source.

[0089] Step S105: When the target report document is in display or storage state, continuously listen for data update events from the dynamic data source, locate the affected document elements based on the field-level dynamic binding relationship, drive the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document.

[0090] In this embodiment, the hybrid monitoring strategy refers to a data acquisition strategy that dynamically switches between polling monitoring based on a fixed time window and event-driven monitoring based on a message queue, based on the real-time metrics and load status of the data source, in order to efficiently acquire data streams.

[0091] In this embodiment, denoising refers to removing outliers and redundant records from structured data, and filtering out noisy text fragments and duplicate content from unstructured data; format normalization refers to converting structured data into a preset unified data format and converting unstructured data into a preset text or document format.

[0092] In this embodiment, dynamic data stream refers to a data sequence that has been processed by denoising and format normalization, and has a unified interface description and temporal attributes, which is used for subsequent feature extraction and template matching.

[0093] In this embodiment, the template decision model refers to a trained classification or regression model that takes a feature vector as input and outputs the matching probability distribution between the current data window and each template in the preset report template library.

[0094] In this embodiment, a content slot refers to a predefined placeholder area in the report structure template used to fill specific types of content. Each slot is associated with a content type and a semantic role.

[0095] In this embodiment, the data processing component refers to a software module used to convert data into content fragments.

[0096] In this embodiment, the field-level dynamic binding relationship refers to the mapping relationship between updatable elements in the report document and specific data fields in the dynamic data source. It records the data source identifier, field path, filtering conditions, and update triggering strategy for precise positioning and partial refresh.

[0097] In this embodiment, incremental replacement refers to replacing only the part of the document affected by data changes without regenerating the entire document; DOM node replacement or streaming editing algorithms are used to preserve the rendering state and user interaction state of the unchanged parts.

[0098] The beneficial effects of the above technical solution are as follows: It acquires structured and unstructured data streams from dynamic data sources in real time through a hybrid monitoring strategy, generates dynamic data streams after denoising and normalization; dynamically matches report structure templates based on multi-dimensional feature extraction and template decision models; transforms data into content fragments using data processing components in the rule base, assembles and renders them according to templates, and establishes field-level dynamic binding relationships; finally, it continuously monitors data updates to achieve incremental replacement of local content. This achieves full automation of the report document process from data collection, template matching, content generation to dynamic updates, significantly reducing manual writing and maintenance costs and ensuring real-time consistency between the document and the data source.

[0099] In one embodiment, establishing a communication connection with the dynamic data source, acquiring structured and unstructured data streams from the dynamic data source in real time according to a hybrid monitoring strategy, storing the structured and unstructured data streams into a data buffer queue according to time series, and performing noise reduction and format normalization processing to generate a dynamic data stream includes:

[0100] Determine the database type of the dynamic data source and configure the corresponding data source connector based on the database type;

[0101] Determine the real-time metrics and load status of the dynamic data source, and based on the real-time metrics and load status, support the listening types. Configure a hybrid listening strategy according to the supported listening types, and use the hybrid listening strategy through the data source connector to obtain the structured and unstructured data streams of the dynamic data source in real time.

[0102] The structured and unstructured data streams are marked with timestamps respectively, and then stored in a preset data buffer queue in chronological order.

[0103] Identify and remove outliers and redundant records in structured data, and eliminate noisy text fragments and duplicate content in unstructured data;

[0104] The denoised structured data is converted into a preset first unified data format, and the denoised unstructured data is converted into a preset second unified data format, generating a dynamic data stream with a unified interface description and time sequence attributes.

[0105] In this embodiment, the database types include relational databases, time-series databases, streaming databases, message middleware libraries, etc.

[0106] In this embodiment, the supported listening types include polling listening based on a fixed time window and event-driven listening based on a message queue. The hybrid listening strategy dynamically switches between polling listening and event-driven listening based on the real-time metrics and load status of the data source. Polling listening refers to actively querying the data source at fixed time intervals to obtain the latest data; event-driven listening refers to passively receiving data update notifications by subscribing to the message queue or change log of the data source.

[0107] In this embodiment, the data buffer queue is configured to support mixed caching of structured and unstructured data units.

[0108] In this embodiment, the data source connector refers to a dedicated communication component configured according to the database type, used to establish a connection with the data source and perform data acquisition operations.

[0109] In this embodiment, the real-time performance indicator refers to the requirements of the dynamic data source for data update latency, such as seconds or milliseconds, and is used to determine the focus of the monitoring strategy.

[0110] In this embodiment, the load status refers to the current resource usage of the dynamic data source, such as query pressure, number of connections, and response time, in order to avoid excessive polling that could lead to system overload.

[0111] In this embodiment, the first unified data format refers to the standard format after the structured data is converted, such as a key-value pair set or tabular data, which includes field names, data types, and values.

[0112] In this embodiment, the second unified data format refers to the standard format after unstructured data conversion, such as cleaned plain text, tagged document fragments, or vectorized representation.

[0113] The beneficial effects of the above technical solution are as follows: By configuring connectors according to database type, dynamically configuring hybrid monitoring strategies based on real-time indicators and load status, storing data in a buffer queue after being timestamped, and performing noise reduction and format normalization on both structured and unstructured data, the adaptability and real-time performance of data acquisition are improved, system load is reduced, and the temporal integrity and unified interface of data are ensured, providing a high-quality, standardized dynamic data stream for subsequent feature extraction and template matching.

[0114] In one embodiment, the step of performing multi-dimensional feature extraction on the dynamic data stream in the data buffer queue to generate feature vectors reflecting data attributes, statistical distribution, and temporal change trends, and inputting the feature vectors into a trained template decision model to dynamically match a report structure template adapted to the dynamic data stream, includes:

[0115] The dynamic data stream that has been denoised and normalized is obtained from the data buffer queue, and the dynamic data stream is divided into multiple sequentially arranged data windows according to the preset time window.

[0116] Multidimensional feature extraction is performed on each data window to generate feature vectors that reflect data attributes, statistical distribution, and temporal trends.

[0117] The feature vector is input into the trained template decision model, which outputs the matching probability distribution between the current data window and each template in the preset report template library.

[0118] Based on the matching probability distribution, determine the window-level matching report structure template with the highest matching degree to the current data window from the preset report template library;

[0119] The window-level matching report structure templates are arranged into a template decision sequence according to the time order of each data window. The global consistency analysis of the template decision sequence is performed by a preset template aggregation strategy to calculate the global adaptation score of each window-level matching report structure template.

[0120] Based on the global adaptation score, the final report structure template is determined from multiple window-level matching report structure templates.

[0121] In this embodiment, the data attribute features include the field composition, data type and source identifier of the data in the data window; the statistical distribution features include the mean, variance, skewness and kurtosis of the numerical fields and the frequency distribution of the categorical fields; and the time series change trend features include the difference, rate of change and trend slope calculated based on the changes of key indicators between adjacent windows.

[0122] In this embodiment, the preset aggregation strategy calculates the global adaptation score of each window-level matching report structure template based on the frequency of occurrence, confidence score, temporal distribution characteristics, and structural compatibility constraints between templates.

[0123] In this embodiment, a data window refers to a data block obtained by dividing a dynamic data stream into segments with a fixed time length or a fixed number of records. Adjacent windows may or may not overlap.

[0124] In this embodiment, the window-level matching report structure template refers to the report structure template that best matches the data characteristics of a single data window through a template decision model.

[0125] In this embodiment, the template decision sequence refers to the sequence formed by multiple window-level matching report structure templates arranged in chronological order.

[0126] In this embodiment, the global adaptation score refers to the score obtained by comprehensively considering the template's support, stability and compatibility with previous and subsequent templates in the sequence, and is used to finally determine the unique report structure template.

[0127] The beneficial effects of the above technical solution are as follows: By dividing the dynamic data stream into time windows, multi-dimensional features such as data attributes, statistical distribution, and temporal trends are extracted and input into the template decision model to obtain the matching probability distribution, thereby obtaining window-level templates; then, a global consistency analysis is performed through a template aggregation strategy to calculate the global adaptation score and determine the final template. This solves the problem of template instability or inconsistency that may be caused by single-window matching, improves the accuracy and robustness of template selection, and enables the report structure to adapt to the dynamic changes of the data stream.

[0128] In one embodiment, such as Figure 2 As shown, the step of converting the dynamic data stream into candidate content fragments corresponding to each slot by calling the corresponding data processing component in the rule base according to the content slots and content types of the report structure template includes:

[0129] Step S201: Obtain the chapter hierarchy structure inside the report structure template, and determine multiple content slots and the required content type for each content slot based on the chapter hierarchy structure;

[0130] Step S202: Based on the content type required for each content slot, call the data processing component corresponding to the content type from the preset rule base according to the mapping relationship between the content type and the data processing component;

[0131] Step S203: Using a pre-built semantic mapping engine, based on the semantic roles and data requirements bound to the content slot, aggregate the target data from the dynamic data stream and input it into the called data processing component to generate candidate content fragments corresponding to the content slot.

[0132] Step S204: Associate and bind each generated candidate content fragment with its corresponding content slot to form a slot-content mapping set.

[0133] In this embodiment, the data processing component includes a natural language generation unit, a chart rendering unit, and a cross-analysis unit.

[0134] In this embodiment, the semantic mapping engine automatically aggregates the corresponding data fields from the dynamic data stream based on the semantic roles and data requirements bound to the content slots.

[0135] In this embodiment, semantic roles refer to the semantic meaning of content slots in the report, such as "title", "numerical description", "trend analysis", "comparison chart", etc., which are used to guide data aggregation.

[0136] The beneficial effects of the above technical solution are as follows: By parsing the chapter hierarchy and content slots of the template, the corresponding data processing components are called from the rule base according to the content type, and the target data is aggregated from the data stream using a semantic mapping engine to generate candidate content fragments and bind them to the slots. This achieves componentization and automation of content generation, ensures accurate mapping from data to content, improves the matching accuracy between content fragments and template slots, and facilitates subsequent assembly and dynamic updates.

[0137] In one embodiment, assembling candidate content fragments according to the chapter hierarchy of the report structure template, formatting and rendering them to generate a target report document, and simultaneously establishing field-level dynamic binding relationships between updatable elements within the target report document and the dynamic data source during the rendering process, includes:

[0138] Following the chapter hierarchy of the report structure template, the process traverses downwards from the root chapter, locates the content slots contained in each chapter, and extracts candidate content fragments associated with each content slot from the slot-content mapping set.

[0139] According to the content order and layout relationship defined in the template, the extracted candidate content fragments are inserted into the document flow of the corresponding chapter to complete the content assembly of the report document;

[0140] The candidate content fragments in each content slot are applied with a preset rendering style template according to their content type. The rendering style template defines the font, paragraph spacing, color theme, chart size and alignment for natural language description, statistical charts and pivot tables respectively, and automatically numbers and generates a table of contents for each chapter to form an initial report document with a uniform format.

[0141] Identify updatable elements within the initial report document that have data dependencies on dynamic data sources, and establish field-level dynamic binding relationships for each updatable element;

[0142] A target report document containing binding metadata is generated based on field-level dynamic binding relationships. The binding metadata is embedded in the structural description of the target report document and is used to support precise incremental refresh of updatable elements during document display or storage.

[0143] In this embodiment, the field-level dynamic binding relationship record is a mapping data structure between document object model nodes and dynamic data source identifiers, data field paths, filtering conditions, and update triggering strategies.

[0144] In this embodiment, updatable elements refer to elements in the report document whose content depends on a dynamic data source and needs to be refreshed as the data changes, such as dynamic text blocks, statistical charts, and data tables.

[0145] In this embodiment, binding metadata refers to a set of data embedded in the target report document structure description, containing complete information on field-level dynamic binding relationships.

[0146] The beneficial effects of the above technical solution are as follows: Candidate content fragments are extracted and traversed according to the template chapter hierarchy, inserted into the document flow, and rendered style templates for different content types are applied, automatically generating numbers and a table of contents; simultaneously, updatable elements are identified and field-level dynamic binding relationships are established, embedding the binding metadata into the document structure. This ensures the uniformity and professional aesthetics of the report format and provides a precise positioning and refresh mechanism for subsequent incremental updates, avoiding the performance overhead of a full document redraw.

[0147] In one embodiment, such as Figure 3 As shown, the step of continuously monitoring data update events from dynamic data sources when the target report document is in a display or storage state, locating affected document elements based on field-level dynamic binding relationships, and driving the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document includes:

[0148] Step S301: When the target report document is in the display or storage state, start the real-time consistency maintenance process, continuously listen for data update events of dynamic data sources, and obtain data update information including data source identifier, changed field path and changed data value;

[0149] Step S302: Parse the data update information and, based on the field-level dynamic binding relationship embedded in the target report document, retrieve all document element nodes that match the data source identifier and the changed field path to determine the local area that needs to be updated.

[0150] Step S303: For each affected document element node, based on its corresponding content generation logic configuration and original content type, call the corresponding data processing component in the rule base, and regenerate the local content fragment using the changed data value as input.

[0151] Step S304: Using an incremental replacement algorithm, the newly generated local content fragments are replaced with the corresponding local areas of the target report document, while preserving the rendering state, view position, and user interaction state of the unchanged parts of the document.

[0152] In this embodiment, the content generation logic configuration refers to the configuration information associated with document element nodes, which explains how to convert raw data into the final display content.

[0153] In this embodiment, the incremental replacement algorithm refers to a local document update algorithm, such as node replacement based on the document object model, difference merging based on the content editable area, or patch application based on version vector, to ensure that only the necessary parts are changed.

[0154] The beneficial effects of the above technical solution are as follows: By locating the affected document element nodes through field-level binding relationships, calling the corresponding data processing components to regenerate local content fragments with the changed data, and using an incremental replacement algorithm to replace local areas while retaining the rendering state and user interaction state of the unchanged parts. This achieves real-time, accurate, and local updates to report documents, significantly reducing response latency and computing resource consumption, and improving user experience and data timeliness.

[0155] In one embodiment, it also includes:

[0156] When generating the target report document, determine the data correlation index between each candidate content segment, and construct a causal dependency graph between content segments based on the data correlation index;

[0157] When data updates trigger field changes, forward propagation is performed based on the causal dependency graph to determine all related content fragments that have direct and indirect dependencies on the local content fragment;

[0158] The update cost of a local content segment is determined based on the associated content segments. If the update cost exceeds the threshold, the entire section or the entire article is regenerated.

[0159] Record user interaction behavior on each content segment of the target report document, calculate the popularity value based on the interaction behavior, and use a long short-term memory network to predict the hot reporting area in the next time window based on the popularity value. Pre-generate target candidate content segments for the hot reporting area in advance to reduce update delay.

[0160] In this embodiment, the data association index refers to a numerical value that quantifies the strength of data dependence between two content fragments, such as the number of data fields used in common or the number of transformation steps the data flows through.

[0161] In this embodiment, the causal dependency graph is a directed graph where nodes represent content fragments and directed edges represent the data output of one fragment as the input of another fragment, used to describe the data propagation relationship between content.

[0162] In this embodiment, the update cost refers to a quantified value of the computational resources, time, or complexity required to update a content fragment and all its indirectly dependent fragments.

[0163] In this embodiment, the popularity score refers to a score calculated based on the user's interaction with the content segment, reflecting the user's level of attention to the segment.

[0164] In this embodiment, the Long Short-Term Memory network is a variant of a recurrent neural network used for time series forecasting; in this scheme, it is used to predict reporting areas that may become hotspots within a future time window based on historical heat value sequences.

[0165] The beneficial effects of the above technical solution are as follows: By constructing a causal dependency graph through calculating the data association index between content segments, forward propagation is used during data updates to identify all directly and indirectly dependent segments and assess the update cost, triggering a complete section or article regeneration if necessary; simultaneously, popularity values ​​are calculated based on user interaction behavior, and long short-term memory networks are used to predict hotspot areas and pre-generate content. This optimizes the efficiency of cascading updates, avoids frequent regeneration caused by excessively long dependency chains, and reduces update latency in hotspot areas through the pre-generation mechanism, further improving response speed and intelligence.

[0166] In one embodiment, the process of acquiring structured and unstructured data streams from dynamic data sources in real time according to a hybrid monitoring strategy further includes:

[0167] Configure a load metric collection strategy for the dynamic data source connector. The load metric collection strategy defines the set of runtime metrics to be collected, the collection period, and the collection method.

[0168] According to the collection cycle, the raw indicator data of the runtime indicator set is periodically obtained through the indicator exposure interface built into the dynamic data source connector.

[0169] The raw indicator data is preprocessed to obtain a standardized runtime indicator sequence. Based on the standardized runtime indicator sequence, local features that characterize the current load status of the dynamic data source connector are extracted to generate a local load feature vector.

[0170] Configure a global resource indicator collection strategy for the report document generation system. The global resource indicator collection strategy defines the set of system-level resource indicators to be collected, the collection period, and the data aggregation granularity.

[0171] According to the collection cycle, the system resource probe, which interacts with the operating environment of the report document generation system, periodically obtains the raw indicator data of the system-level resource indicator set;

[0172] The acquired raw indicator data is preprocessed to obtain a standardized system resource indicator sequence. Based on the standardized system resource indicator sequence, features representing the overall system load status and resource usage patterns are extracted from the report document to generate a system resource feature vector.

[0173] The local load feature vector and the system resource feature vector are normalized and then concatenated to generate a joint state vector.

[0174] According to the preset time alignment rules, the local load feature vector and the system resource feature vector are matched to the same time window to generate paired samples;

[0175] A statistical scaling transformation method is used to map the feature values ​​of different dimensions and numerical ranges in paired samples to a preset unified numerical range, thereby obtaining normalized local load feature vectors and normalized system resource feature vectors.

[0176] The normalized local load feature vector and the normalized system resource feature vector are concatenated according to a predetermined dimensional concatenation order to generate a multidimensional joint state vector.

[0177] The multidimensional joint state vector is input as an input sequence into a preset load prediction model to extract the temporal dependency features in the input sequence and output the expected load sequence for a preset number of time windows in the future.

[0178] Based on the expected load sequence, the adjustment parameters of the hybrid listening strategy are dynamically calculated using a preset strategy to adjust the mapping relationship. The adjustment parameters include the weight ratio between polling listening and event-driven listening, as well as the polling time interval of polling listening.

[0179] The calculated weighting ratio and polling interval are sent to the acquisition module so that when the future time window arrives, polling listening and event-driven listening are allocated or switched in the listening tasks of each dynamic data source according to the weighting ratio, and polling operation is performed according to the polling interval to realize dynamic adjustment of the hybrid listening strategy.

[0180] In this embodiment, the load metric collection strategy refers to a set of rules configured for the dynamic data source connector, which defines the runtime metrics to be collected (such as CPU utilization, memory usage, network latency, query response time, number of connections, etc.), collection period, and collection method.

[0181] In this embodiment, the runtime metrics set refers to quantitative metrics that reflect the current operating status of the data source connector, including: the connector's own CPU utilization, memory usage, request queue length, average / maximum response time, error rate, throughput (messages / second), etc.

[0182] In this embodiment, the local load feature vector refers to the feature vector extracted from the runtime metrics of the data source connector to characterize its current load state.

[0183] In this embodiment, the global resource metric collection strategy refers to a set of rules configured for the report document generation system, which defines the system-level resource metrics to be collected (such as overall CPU utilization, memory utilization, disk I / O, network bandwidth, etc.), collection period, and data aggregation granularity.

[0184] In this embodiment, the system resource probe refers to a software component that interacts with the runtime environment of the report document generation system to obtain system-level resource usage data.

[0185] In this embodiment, the system resource feature vector refers to the feature vector extracted from system-level resource indicators and used to characterize the overall load status of the report document generation system.

[0186] In this embodiment, the joint state vector refers to a multi-dimensional vector formed by concatenating the local load feature vector and the system resource feature vector after normalization, which is used to comprehensively describe the load status of the data source connector and the system as a whole.

[0187] In this embodiment, the time alignment rule refers to the method of matching feature vectors from different sources to the same time window.

[0188] In this embodiment, the statistical scaling method refers to a standardization method that maps feature values ​​with different dimensions and numerical ranges to a unified numerical interval.

[0189] In this embodiment, the multidimensional joint state vector refers to the final vector obtained after normalization and dimension concatenation, and its dimension is equal to the sum of the local load feature vector dimension and the system resource feature vector dimension.

[0190] In this embodiment, the load prediction model refers to a pre-trained time series prediction model that takes a multi-dimensional joint state vector sequence as input and outputs a sequence of expected load values ​​for multiple future time windows.

[0191] In this embodiment, the expected load sequence refers to the load prediction values ​​for multiple consecutive time windows output by the load prediction model, where each value represents the average or peak load within that window.

[0192] In this embodiment, the policy adjustment mapping relationship refers to the function or rule table that converts the load values ​​in the expected load sequence into hybrid monitoring policy adjustment parameters. For example, under high load, the weight ratio of event-driven monitoring is increased and the polling frequency is reduced; under low load, the polling frequency can be appropriately increased or kept balanced.

[0193] In this embodiment, the weight ratio between polling listening and event-driven listening refers to the weight ratio of the two listening methods in the hybrid listening strategy. This ratio determines the allocation of resources or the priority of which method to use to obtain data in the supervisor's listening task.

[0194] In this embodiment, the polling time interval refers to the time interval between two active queries to the data source in the polling listening method. The shorter the interval, the higher the real-time performance but the greater the system load, and vice versa.

[0195] In this embodiment, the acquisition module is responsible for executing the data source connection and hybrid monitoring strategy.

[0196] The beneficial effects of the above technical solution are as follows: By collecting local load metrics from the data source connector and global resource metrics from the report document generation system in real time, a joint state vector is constructed, and a load prediction model is used to predict future load. This allows for dynamic adjustment of the weight ratio and polling interval between polling and event-driven monitoring in the hybrid monitoring strategy, achieving adaptive optimization of the data acquisition strategy. It can intelligently allocate monitoring resources based on the current system load and resource usage, achieving a dynamic balance between data real-time performance and system overhead. This avoids performance degradation under high load or resource waste under low load caused by fixed strategies, thereby improving the overall system stability, data acquisition efficiency, and resource utilization. It is particularly suitable for high-concurrency, highly dynamic data monitoring and reporting scenarios.

[0197] In one embodiment, determining the data correlation index between each candidate content fragment when generating the target report document, and constructing a causal dependency graph between content fragments based on the data correlation index, includes:

[0198] Record the set of dynamic data source field paths and data transformation operation chain that each candidate content fragment depends on;

[0199] For any two candidate content fragments, determine the data field overlap, data flow propagation strength, semantic logical correlation, and time sequence synchronization dependency based on the dynamic data source field path set and data transformation operation chain.

[0200] The overlap of data fields, the strength of data stream propagation, the degree of semantic and logical relevance, and the time sequence synchronization dependency are weighted and fused according to preset weights to calculate the data correlation index between any two candidate content fragments:

[0201]

[0202] in, Let represent the data correlation index between the i-th candidate content fragment and the j-th candidate content fragment. This is represented as the overlap weight of the data fields. This represents the overlap of data fields between the i-th and j-th candidate content fragments. Represented as the data stream propagation strength weight, This represents the data stream propagation strength between the i-th candidate content fragment and the j-th candidate content fragment. Represented as semantic logical relevance weight, Let represent the semantic logical correlation between the i-th candidate content fragment and the j-th candidate content fragment. Represented as the temporal synchronization dependency weight, It is represented as the temporal synchronization dependency between the i-th candidate content segment and the j-th candidate content segment;

[0203] Based on the comparison between the data correlation index and the preset threshold, it is determined whether to establish a directed edge between two candidate content segments, and the direction of the directed edge is determined according to the data flow direction to generate a causal dependency graph.

[0204] In this embodiment, the dynamic data source field path set refers to the location path of the original data field on which the candidate content fragment depends in the dynamic data source.

[0205] In this embodiment, the data transformation operation chain refers to a series of data processing operations that take place from the original data field to the final content fragment.

[0206] In this embodiment, the data field overlap weight is used to measure the contribution of the degree of overlap of the data source fields on which two candidate content fragments depend to the overall relevance. The physical meaning is the importance of data homology. A larger value indicates that the two fragments are strongly related based solely on shared fields, making it suitable for scenarios where fields are closely tied to business meanings and shared fields inevitably lead to content interconnection.

[0207] In this embodiment, the data stream propagation strength weight is used to measure the contribution of the existence of a direct data processing chain between candidate content segments to the overall relevance. Physically, it represents the importance of the predecessor-successor processing relationship. A larger value indicates a strong dependency, meaning that as long as the output of one segment is the input of another segment.

[0208] In this embodiment, the semantic logical relevance weight is used to measure the contribution of the semantic similarity of fragment text descriptions to the overall relevance. The physical meaning is the importance of logical topic consistency. A higher value indicates that even if they do not share the same fields, as long as the content semantically belongs to the same topic, they should be considered relevant. This is suitable for natural language generation reports, avoiding missing inherent connections due to different field names.

[0209] In this embodiment, the temporal synchronization dependency weight is used to measure the contribution of the sequential tightness of segments in the report to the overall relevance. Physically, it represents the importance of document pre- and post-departure dependencies. A larger value indicates that adjacent segments are more likely to depend on each other, suitable for reports that strictly follow a data-first, conclusion-later format.

[0210] In this embodiment, data field overlap refers to the ratio of the intersection size to the union size of the data field sets on which two candidate content fragments depend, used to quantify the degree of overlap at the data source level.

[0211] In this embodiment, the data stream propagation intensity refers to the data transmission intensity from one candidate content segment to another, which can be calculated comprehensively based on factors such as whether there is an output-input relationship in the data transformation operation chain and the proportion of the amount of data transmitted.

[0212] In this embodiment, semantic logical relevance refers to the degree of relevance between two candidate content fragments in terms of business semantics, such as whether they describe the same entity, whether they have a causal relationship, or whether they are adjacent or related in the reporting context. It can be calculated using a semantic similarity model or a domain knowledge graph.

[0213] In this embodiment, temporal synchronization dependency refers to whether the data on which two candidate content segments depend have a synchronous update requirement in time.

[0214] In this embodiment, the data association index is a value between 0 and 1, representing the degree of data dependency between two candidate content segments. The higher the index, the stronger the correlation between the two segments in terms of data source, transformation process, semantics, or temporal sequence, and the more likely directed edges should be established in the causal dependency graph.

[0215] In this embodiment, the preset threshold is a critical value used to determine whether a directed edge is established between two content segments, and is set to 0.6.

[0216] In this embodiment, the direction of the directed edges is determined based on the data flow direction: if the data of the i-th candidate content fragment is transformed and used to generate the j-th candidate content fragment, then the direction of the edge is i→j. If they are mutually dependent, bidirectional edges can be established.

[0217] The beneficial effects of the above technical solution are as follows: By weighting the data association index between content fragments from four dimensions—data field overlap, data flow propagation strength, semantic logical correlation, and time-series synchronization dependency—and constructing a causal dependency graph representing the dependency relationship based on the threshold, the solution achieves accurate modeling and quantification of complex dependency relationships between content fragments in the report document. When the data source changes, it can quickly locate all directly and indirectly dependent content fragments, intelligently assess the update cost, and select the optimal update strategy, avoiding the performance overhead caused by blind full refresh. It significantly reduces cascading update latency and computing resource consumption, and improves the real-time response efficiency and user experience of the report document under dynamic data-driven conditions.

[0218] In one embodiment, this embodiment also discloses a report document generation system based on dynamic data-driven methods, such as... Figure 4 As shown, the system includes:

[0219] The acquisition module 401 is used to establish a communication connection with the dynamic data source, acquire the structured data stream and unstructured data stream of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured data stream and unstructured data stream into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data stream.

[0220] The matching module 402 is used to perform multi-dimensional feature extraction on the dynamic data stream in the data buffer queue, generate feature vectors that reflect data attributes, statistical distribution and time series change trends, input the feature vectors into the trained template decision model, and dynamically match the report structure template that is adapted to the dynamic data stream.

[0221] The conversion module 403 is used to call the corresponding data processing component in the rule base to convert the dynamic data stream into candidate content fragments corresponding to each slot, based on the content slot and content type of the report structure template.

[0222] The generation module 404 is used to assemble candidate content fragments according to the chapter level of the report structure template, format and render them to generate the target report document. At the same time, during the rendering process, it establishes field-level dynamic binding relationships between the updatable elements in the target report document and the dynamic data source.

[0223] The incremental replacement module 405 is used to continuously listen for data update events from dynamic data sources when the target report document is in a display or storage state. Based on the field-level dynamic binding relationship, it locates the affected document elements, drives the data processing component to regenerate local content fragments for the changed data, and incrementally replaces them with the corresponding positions in the target report document.

[0224] The working principle and beneficial effects of the above technical solution have been explained in the method embodiments, and will not be repeated here.

[0225] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0226] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for generating report documents based on dynamic data-driven approaches, characterized in that, Includes the following steps: Establish a communication connection with the dynamic data source, acquire the structured and unstructured data streams of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured and unstructured data streams into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data streams; Multidimensional feature extraction is performed on the dynamic data stream in the data buffer queue to generate feature vectors that reflect data attributes, statistical distribution and time series change trends. The feature vectors are then input into a trained template decision model to dynamically match a report structure template that is compatible with the dynamic data stream. Based on the content slots and content types in the report structure template, the corresponding data processing components in the rule base are called to transform the dynamic data stream into candidate content fragments corresponding to each slot. The candidate content fragments are assembled according to the chapter hierarchy of the report structure template, formatted and rendered to generate the target report document. At the same time, during the rendering process, field-level dynamic binding relationships are established between the updatable elements in the target report document and the dynamic data source. While the target report document is in a display or storage state, continuously listen for data update events from dynamic data sources, locate the affected document elements based on field-level dynamic binding relationships, drive the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document.

2. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The process of establishing a communication connection with the dynamic data source, acquiring structured and unstructured data streams from the dynamic data source in real time according to a hybrid monitoring strategy, storing the structured and unstructured data streams into a data buffer queue according to time series, and performing noise reduction and format normalization processing to generate a dynamic data stream includes: Determine the database type of the dynamic data source and configure the corresponding data source connector based on the database type; Determine the real-time metrics and load status of the dynamic data source, and based on the real-time metrics and load status, support the listening types. Configure a hybrid listening strategy according to the supported listening types, and use the hybrid listening strategy through the data source connector to obtain the structured and unstructured data streams of the dynamic data source in real time. The structured and unstructured data streams are marked with timestamps respectively, and then stored in a preset data buffer queue in chronological order. Identify and remove outliers and redundant records in structured data, and eliminate noisy text fragments and duplicate content in unstructured data; The denoised structured data is converted into a preset first unified data format, and the denoised unstructured data is converted into a preset second unified data format, generating a dynamic data stream with a unified interface description and time sequence attributes.

3. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The process involves multi-dimensional feature extraction of the dynamic data stream in the data buffer queue to generate feature vectors reflecting data attributes, statistical distribution, and temporal trends. These feature vectors are then input into a trained template decision model to dynamically match a report structure template adapted to the dynamic data stream. This includes: The dynamic data stream that has been denoised and normalized is obtained from the data buffer queue, and the dynamic data stream is divided into multiple sequentially arranged data windows according to the preset time window. Multidimensional feature extraction is performed on each data window to generate feature vectors that reflect data attributes, statistical distribution, and temporal trends. The feature vector is input into the trained template decision model, which outputs the matching probability distribution between the current data window and each template in the preset report template library. Based on the matching probability distribution, determine the window-level matching report structure template with the highest matching degree to the current data window from the preset report template library; The window-level matching report structure templates are arranged into a template decision sequence according to the time order of each data window. The global consistency analysis of the template decision sequence is performed by a preset template aggregation strategy to calculate the global adaptation score of each window-level matching report structure template. Based on the global adaptation score, the final report structure template is determined from multiple window-level matching report structure templates.

4. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The step of converting the dynamic data stream into candidate content fragments corresponding to each slot by calling the corresponding data processing component in the rule base based on the content slots and content types of the report structure template includes: Obtain the chapter hierarchy structure within the report structure template, determine multiple content slots based on the chapter hierarchy structure, and define the required content type for each content slot; Based on the content type required for each content slot, the corresponding data processing component is called from the pre-set rule base according to the mapping relationship between the content type and the data processing component; Using a pre-built semantic mapping engine, target data is aggregated from the dynamic data stream based on the semantic roles and data requirements bound to the content slot and input into the called data processing component to generate candidate content fragments corresponding to the content slot. Each generated candidate content fragment is associated and bound to its corresponding content slot to form a slot-content mapping set.

5. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The process of assembling candidate content fragments according to the chapter hierarchy of the report structure template, formatting and rendering them to generate a target report document, and simultaneously establishing field-level dynamic binding relationships between updatable elements within the target report document and the dynamic data source during the rendering process includes: Following the chapter hierarchy of the report structure template, the process traverses downwards from the root chapter, locates the content slots contained in each chapter, and extracts candidate content fragments associated with each content slot from the slot-content mapping set. According to the content order and layout relationship defined in the template, the extracted candidate content fragments are inserted into the document flow of the corresponding chapter to complete the content assembly of the report document; The candidate content fragments in each content slot are applied with a preset rendering style template according to their content type. The rendering style template defines the font, paragraph spacing, color theme, chart size and alignment for natural language description, statistical charts and pivot tables respectively, and automatically numbers and generates a table of contents for each chapter to form an initial report document with a uniform format. Identify updatable elements within the initial report document that have data dependencies on dynamic data sources, and establish field-level dynamic binding relationships for each updatable element; A target report document containing binding metadata is generated based on field-level dynamic binding relationships. The binding metadata is embedded in the structural description of the target report document and is used to support precise incremental refresh of updatable elements during document display or storage.

6. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The process of continuously monitoring data update events from dynamic data sources while the target report document is in a display or storage state, locating affected document elements based on field-level dynamic binding relationships, and driving the data processing component to regenerate local content fragments for the changed data and incrementally replace them with the corresponding positions in the target report document includes: When the target report document is in a display or storage state, start the real-time consistency maintenance process, continuously listen for data update events of dynamic data sources, and obtain data update information including data source identifier, changed field path and changed data value; Parse the data update information and, based on the field-level dynamic binding relationship embedded in the target report document, retrieve all document element nodes that match the data source identifier and the path of the changed field to determine the local area that needs to be updated. For each affected document element node, based on its corresponding content generation logic configuration and original content type, the corresponding data processing component in the rule base is called, and the local content fragment is regenerated using the changed data value as input. An incremental replacement algorithm is used to replace the newly generated local content fragments with the corresponding local areas of the target report document, while preserving the rendering state, view position, and user interaction state of the unchanged parts of the document.

7. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, Also includes: When generating the target report document, determine the data correlation index between each candidate content fragment, and construct a causal dependency graph between content fragments based on the data correlation index; When data updates trigger field changes, forward propagation is performed based on the causal dependency graph to determine all related content fragments that have direct and indirect dependencies on the local content fragment; The update cost of a local content segment is determined based on the associated content segments. If the update cost exceeds the threshold, the entire section or the entire article is regenerated. Record user interaction behavior on each content segment of the target report document, calculate the popularity value based on the interaction behavior, and use a long short-term memory network to predict the hot reporting area in the next time window based on the popularity value. Pre-generate target candidate content segments for the hot reporting area in advance to reduce update delay.

8. The report document generation method based on dynamic data-driven approach according to claim 1, characterized in that, The process of acquiring structured and unstructured data streams from dynamic data sources in real time according to a hybrid monitoring strategy also includes: Configure a load metric collection strategy for the dynamic data source connector. The load metric collection strategy defines the set of runtime metrics to be collected, the collection period, and the collection method. According to the collection cycle, the raw indicator data of the runtime indicator set is periodically obtained through the indicator exposure interface built into the dynamic data source connector. The raw indicator data is preprocessed to obtain a standardized runtime indicator sequence. Based on the standardized runtime indicator sequence, local features that characterize the current load status of the dynamic data source connector are extracted to generate a local load feature vector. Configure a global resource indicator collection strategy for the report document generation system. The global resource indicator collection strategy defines the set of system-level resource indicators to be collected, the collection period, and the data aggregation granularity. According to the collection cycle, the system resource probe, which interacts with the operating environment of the report document generation system, periodically obtains the raw indicator data of the system-level resource indicator set; The acquired raw indicator data is preprocessed to obtain a standardized system resource indicator sequence. Based on the standardized system resource indicator sequence, features representing the overall system load status and resource usage patterns are extracted from the report document to generate a system resource feature vector. The local load feature vector and the system resource feature vector are normalized and then concatenated to generate a joint state vector. According to the preset time alignment rules, the local load feature vector and the system resource feature vector are matched to the same time window to generate paired samples; A statistical scaling transformation method is used to map the feature values ​​of different dimensions and numerical ranges in paired samples to a preset unified numerical range, thereby obtaining normalized local load feature vectors and normalized system resource feature vectors. The normalized local load feature vector and the normalized system resource feature vector are concatenated according to a predetermined dimensional concatenation order to generate a multidimensional joint state vector. The multidimensional joint state vector is input as an input sequence into a preset load prediction model to extract the temporal dependency features in the input sequence and output the expected load sequence for a preset number of time windows in the future. Based on the expected load sequence, the adjustment parameters of the hybrid listening strategy are dynamically calculated using a preset strategy to adjust the mapping relationship. The adjustment parameters include the weight ratio between polling listening and event-driven listening, as well as the polling time interval of polling listening. The calculated weighting ratio and polling interval are sent to the acquisition module so that when the future time window arrives, polling listening and event-driven listening are allocated or switched in the listening tasks of each dynamic data source according to the weighting ratio, and polling operation is performed according to the polling interval to realize dynamic adjustment of the hybrid listening strategy.

9. The report document generation method based on dynamic data-driven approach according to claim 7, characterized in that, The step of determining the data correlation index between each candidate content fragment when generating the target report document, and constructing a causal dependency graph between content fragments based on the data correlation index, includes: Record the set of dynamic data source field paths and data transformation operation chain that each candidate content fragment depends on; For any two candidate content fragments, determine the data field overlap, data flow propagation strength, semantic logical correlation, and time sequence synchronization dependency based on the dynamic data source field path set and data transformation operation chain. The overlap of data fields, the strength of data stream propagation, the degree of semantic and logical relevance, and the time sequence synchronization dependency are weighted and fused according to preset weights to calculate the data correlation index between any two candidate content fragments: ； in, Let represent the data correlation index between the i-th candidate content fragment and the j-th candidate content fragment. This is represented as the overlap weight of the data fields. This represents the overlap of data fields between the i-th and j-th candidate content fragments. Represented as the data stream propagation strength weight, This represents the data stream propagation intensity between the i-th candidate content fragment and the j-th candidate content fragment. Represented as semantic logical relevance weight, Let represent the semantic logical correlation between the i-th candidate content fragment and the j-th candidate content fragment. Represented as the weight of time-series synchronization dependency. It is represented as the temporal synchronization dependency between the i-th candidate content segment and the j-th candidate content segment; Based on the comparison between the data correlation index and the preset threshold, it is determined whether to establish a directed edge between two candidate content segments, and the direction of the directed edge is determined according to the data flow direction to generate a causal dependency graph.

10. A report document generation system based on dynamic data-driven methods, characterized in that, The system includes: The acquisition module is used to establish a communication connection with the dynamic data source, acquire the structured and unstructured data streams of the dynamic data source in real time according to the hybrid monitoring strategy, store the structured and unstructured data streams into the data buffer queue according to the time series, and perform noise reduction and format normalization processing to generate dynamic data streams. The matching module is used to extract multi-dimensional features from the dynamic data stream in the data buffer queue, generate feature vectors that reflect data attributes, statistical distribution and time series change trends, and input the feature vectors into the trained template decision model to dynamically match the report structure template that is adapted to the dynamic data stream. The conversion module is used to call the corresponding data processing components in the rule base to convert the dynamic data stream into candidate content fragments corresponding to each slot, based on the content slots and content types of the report structure template. The generation module is used to assemble candidate content fragments according to the chapter hierarchy of the report structure template, format and render them to generate the target report document. At the same time, during the rendering process, it establishes field-level dynamic binding relationships between the updatable elements in the target report document and the dynamic data source. The incremental replacement module continuously listens for data update events from dynamic data sources when the target report document is in a display or storage state. Based on field-level dynamic binding relationships, it locates the affected document elements, drives the data processing component to regenerate local content fragments for the changed data, and incrementally replaces them with the corresponding positions in the target report document.

Images