A deep learning-based configuration report intelligent jump method and device

Abstract

The application relates to the technical field of data processing, and discloses a configuration report intelligent jump method and device based on deep learning. The method comprises the following steps: collecting user data in a report system to construct a training data set, performing multidimensional feature extraction and standardization processing, constructing a jump intention prediction model and a query condition importance sorting model and training, making the model lightweight to form an inference engine embedded in a report system public service layer, defining a page jump rule and designing a standardized parameter passing format, binding a jump correlation in a configuration end and supporting visual configuration, calling the model to carry out jump with parameters when a user triggers jump, and automatically returning and retrying when the jump fails, so that the technical problems of lack of standardization in page jump of an existing report system, incapability of automatically delivering query conditions, insufficient support for multi-tab linkage and cross-menu jump are solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and apparatus for intelligent navigation of configurable reports based on deep learning. Background Technology

[0002] In current enterprise-level dynamically configurable reporting systems, data in different rows of reports often have different business type attributes (such as waybill type, settlement type, expense type, etc.). These differences directly determine the different form fields required when editing the data in that row—the editing fields, operation buttons, and even pop-up layouts corresponding to different types of data are fundamentally different.

[0003] Current technologies generally adopt a "one form per data type" architecture strategy: developing and deploying a complete set of pop-up forms and corresponding button systems for each data type. While this approach is logically clear, it exposes four core problems in actual engineering implementation: Page redirection lacks standardization: Existing methods mostly involve independent access between pages, lacking unified redirection logic specifications, and there is no unified standard for redirection methods.

[0004] Search criteria cannot be carried over: After switching pages, you need to manually re-enter the search criteria, which is cumbersome and inefficient.

[0005] Multi-tab linkage is not supported: The current method does not support linkage and navigation between multiple tabs under the same menu.

[0006] Cross-menu navigation is complicated: Navigating across menu pages requires manual input of content, and the target page cannot be automatically loaded.

[0007] The aforementioned issues have impacted the efficiency of users searching for detailed data and severely affected the user experience of business data analysis and comparison. There is an urgent need for a new technical solution that can automatically carry over page navigation conditions and automatically backfill queries on the target page. Summary of the Invention

[0008] The main objective of this invention is to provide a method and apparatus for intelligent navigation of configurable reports based on deep learning, in order to solve the technical problems of lack of standardization in page navigation, inability to automatically pass query conditions, insufficient support for multi-tab linkage and cross-menu navigation in existing configurable report systems.

[0009] The first aspect of this invention provides a deep learning-based intelligent navigation method for configurable reports, comprising: Collect user data from the reporting system and build a training dataset; Multi-dimensional feature extraction and standardization are performed on the training dataset; Construct a redirection intent prediction model and a query condition importance ranking model, and train the models; The two trained models are lightweighted and processed to form an inference engine, which is then embedded into the public service layer of the reporting system. Define page redirection rules, unify the logic for collecting query conditions, and design a standardized parameter passing format; The configuration terminal binds the basic association between menus, tabs and target pages, recommends candidate jump targets in real time based on the jump intent prediction model, and supports visually adding or modifying jump relationships through the configuration terminal. When a user triggers a redirect, two sets of models are invoked, carrying standardized routes and core parameters to redirect to the target page. If the redirect fails, the fault tolerance module is activated to automatically fall back and retry.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the step of collecting user data within the reporting system and constructing a training dataset includes: Collect user redirection logs, collecting all user redirection behavior data within the reporting system. Core fields include user unique identifier, redirection source page, redirection target page, redirection trigger time, redirection time, redirection trigger method, and redirection result. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names, field data types, filter values, filter operators, condition selection order, condition validity status, and number of query executions. Collect menu / tab relationships and pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row location, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering.

[0011] A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time, and data preprocessing is completed through ETL tools. The cleaned data was divided into training, validation, and test sets in a 7:2:1 ratio.

[0012] Optionally, in a second implementation of the first aspect of the present invention, the step of performing multi-dimensional feature extraction and standardization on the training dataset includes: Extract user-dimensional features, including user ID, user business role, frequency of user report usage, and user historical navigation preference tags; Extract page scene features, including current menu ID, current tab index, page type, target page popularity, menu level depth, and tab page switching frequency; Extract query condition features, including query field name, field data type, filter value, number of condition combinations, frequency of condition usage, and data volume of the filtered results. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence, and use sequence embedding to convert discrete behaviors into continuous vectors; Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations; Discrete features are encoded using one-hot encoding or ordinal encoding, continuous features are processed using min-max normalization, and effective features are selected through variance analysis and mutual information coefficients.

[0013] Optionally, in a third implementation of the first aspect of the present invention, the step of constructing a jump intent prediction model and a query condition importance ranking model, and training the models, includes: The specific process of building a redirection intent prediction model includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

[0014] Optionally, in a fourth implementation of the first aspect of the present invention, the step of lightweighting the trained model to form an inference engine and embedding it into the public service layer of the reporting system includes: Model quantization and channel pruning techniques are used to compress the trained model, convert floating-point parameters into integer parameters, and remove redundant neurons. The compressed dual model is converted into the ONNX universal format, integrated with the TensorRT inference acceleration engine, and encapsulated into a standardized API interface to provide jump intent prediction and condition importance ranking services. Deploy the inference engine to the public service layer of the reporting system, and interface with the query condition collection module and the jump relationship configuration module to receive front-end page behavior data and output prediction results in real time; Establish a model caching mechanism to cache prediction results for high-frequency users and pages; The design incorporates incremental training and hot update mechanisms, periodically collecting new user behavior data to incrementally fine-tune the model.

[0015] Optionally, in the fifth implementation of the first aspect of the present invention, the definition of page redirection rules, the unified query condition collection logic, and the design of a standardized parameter passing format include: Define rules for linked navigation between multiple tabs within the same menu, and establish standardized navigation paths between different tabs under the same menu. Define rules for seamless navigation across menu pages and establish standardized navigation paths between pages with different menus; Unify the logic for collecting query conditions, and automatically extract time, region, business type, and key field values; Design a standardized JSON parameter format that supports cross-page and cross-service compatibility parsing.

[0016] Optionally, in the sixth implementation of the first aspect of the present invention, when a user triggers a redirection operation, two sets of models are invoked, carrying standardized routes and core parameters to redirect to the target page. If the redirection fails, a fault-tolerant module is activated to automatically fall back and retry, including: When a user triggers a redirect, the system calls the redirect intent prediction model to output the target page; A query condition importance ranking model is used to extract highly important query conditions and remove redundant parameters. Redirecting to the target page with standardized routes and core parameters; The target page uses the adaptive parsing module to backfill conditions, silently query and display data; If the jump fails, the fault tolerance module is activated to automatically roll back and retry.

[0017] A second aspect of the present invention provides a deep learning-based configurable report intelligent redirection device, comprising: The data acquisition module is used to collect user data within the reporting system and build a training dataset. The feature engineering module is used to perform multi-dimensional feature extraction and standardization on the training dataset; The model training module is used to build a jump intent prediction model and a query condition importance ranking model, and to train the models. The model deployment module is used to lightweight process the two trained models to form an inference engine and embed it into the public service layer of the reporting system. The rule definition module is used to define page redirection rules, unify the logic for collecting query conditions, and design standardized parameter passing formats; The configuration binding module is used to bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support adding or modifying jump relationships visually through the configuration terminal. The redirection processing module is used to call two sets of models and redirect to the target page with standardized routes and core parameters when a user triggers a redirection operation. If the redirection fails, the fault tolerance module is activated to automatically fall back and retry.

[0018] A third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the electronic device to execute the above-described deep learning-based configurable report intelligent jump method.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described deep learning-based configurable report intelligent navigation method.

[0020] In the technical solution of this invention, a training dataset is constructed by collecting user jump logs, query condition sequences, menu and tab associations, and data click behaviors. After feature engineering, a jump intent prediction model and a query condition importance ranking model are trained, and the models are deployed as an inference engine embedded in the public service layer of a reporting system. Jump associations are bound at the configuration end, and standardized jump rules are defined. When the system responds to a page jump trigger command, the inference engine outputs the target page and extracts the core query conditions, generates standardized routing parameters to complete the jump, and the target page adaptively parses and backfills the query conditions before triggering data retrieval. Through the above technical solution, multi-tab linked jumps within the same menu and cross-menu jumps with conditions are realized, enabling the target page to automatically perform data retrieval based on the query conditions of the source page. Attached Figure Description

[0021] Figure 1This is a first flowchart of a deep learning-based intelligent navigation method for configurable reports provided in an embodiment of the present invention. Figure 2 The second flowchart of the configuration-based intelligent jump method for reports based on deep learning provided in the embodiments of the present invention; Figure 3 The third flowchart of the configuration-based intelligent jump method for reports based on deep learning provided in the embodiments of the present invention; Figure 4 The fourth flowchart of the configuration-based intelligent jump method for reports based on deep learning provided in the embodiments of the present invention; Figure 5 A schematic diagram of the structure of the configurable report intelligent jump device based on deep learning provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] This invention provides a configurable intelligent report redirection method and apparatus based on deep learning. A training dataset is constructed by collecting user redirection logs, query condition sequences, menu and tab associations, and data click behavior. After feature engineering, a redirection intent prediction model and a query condition importance ranking model are trained and deployed as an inference engine embedded in the public service layer of the reporting system. Redirection relationships are bound at the configuration end, and standardized redirection rules are defined. When the system responds to a page redirection trigger command, the inference engine outputs the target page and extracts the core query conditions, generating standardized routing parameters to complete the redirection. The target page adaptively parses and refills the query conditions before triggering data retrieval.

[0023] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the intelligent navigation method for configurable reports based on deep learning in this invention includes: 101. Collect user data from the reporting system and construct a training dataset; In this embodiment, user jump logs are collected, which collect all user jump behavior data in the reporting system. The core fields include user unique identifier, jump source page (menu ID / Tab index / page type), jump target page, jump trigger time, jump time, jump trigger method (manual click / configured jump / intelligent recommendation), and jump result (success / failure / abnormal). It covers two core scenarios: jumps to multiple tabs within the same menu and jumps across menu pages, and fully records the user's real jump preferences. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names (time / region / business type / waybill number, etc.), field data types, filter values, filter operators (equal to / greater than / contains), condition selection order, condition effective status, and number of query executions, to reconstruct the combination logic and usage habits of user query conditions; Collect menu / tab relationships, pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes (summary table / detail table), and construct the page topology of the report system to provide a foundation for the model to understand page business relationships; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row position, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering, to accurately capture the related behaviors triggered by user data viewing and navigation.

[0025] A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time. Data preprocessing is performed through ETL tools to remove invalid logs (jump timeout / abnormal exit), remove duplicate data, fill missing values ​​(fill with business average / default values), and filter noisy data (test account / bot operation data) to ensure data quality. The cleaned data is divided into training, validation, and test sets in a 7:2:1 ratio. The training set is used for model parameter learning, the validation set is used for model tuning, and the test set is used for model performance evaluation to ensure that the model has generalization ability and can adapt to different users' report navigation behavior.

[0026] 102. Perform multi-dimensional feature extraction and standardization on the training dataset; In this embodiment, user-dimensional features are extracted, including user ID, user business role (operations / finance / administrator), user report usage frequency, and user historical jump preference tags. By associating historical behavior with user unique identifiers, the operating habits of different users are distinguished, and personalized jump prediction is achieved. Extract page scene features, including current menu ID, current tab index, page type (summary / detail), target page popularity, menu level depth, and tab page switching frequency, to characterize the business attributes and structural relationships of the report page, enabling the model to understand the relationship logic between pages; Extract query condition features, including query field name, field data type (numeric / text / time), filter value, number of condition combinations, frequency of condition usage, and data volume of condition filtering results. Quantify the business importance and usage frequency of query conditions to provide a basis for condition filtering. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence. Use sequence embedding to convert discrete behaviors into continuous vectors, preserve the temporal nature of user operations, and adapt to the sequence processing characteristics of the Transformer model. Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations, and strengthen the model's ability to identify high-frequency behaviors and core conditions through statistical regularities; Discrete features are encoded using one-hot or ordinal encoding, continuous features are processed by min-max normalization, and effective features are screened through variance analysis and mutual information coefficients, while redundant features are eliminated, thereby reducing the computational complexity of the model and improving prediction efficiency.

[0027] 103. Construct a jump intent prediction model and a query condition importance ranking model, and train the models; In this embodiment, a jump intent prediction model is constructed, and the specific process includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

[0028] 104. Lightweight the two trained models to form an inference engine and embed it into the public service layer of the reporting system; 105. Define page redirection rules, unify the logic for collecting query conditions, and design a standardized parameter passing format; 106. Bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support visually adding or modifying jump relationships through the configuration terminal; 107. When a user triggers a redirect, two sets of models are invoked, and the user is redirected to the target page with standardized routes and core parameters. If the redirect fails, the fault tolerance module is activated to automatically fall back and retry.

[0029] Please see Figure 2 The second embodiment of the configuration-based intelligent jump method for reports based on deep learning in this invention includes: 201. Collect user data from the reporting system and construct a training dataset; In this embodiment, user jump logs are collected, which collect all user jump behavior data in the reporting system. The core fields include user unique identifier, jump source page (menu ID / Tab index / page type), jump target page, jump trigger time, jump time, jump trigger method (manual click / configured jump / intelligent recommendation), and jump result (success / failure / abnormal). It covers two core scenarios: jumps to multiple tabs within the same menu and jumps across menu pages, and fully records the user's real jump preferences. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names (time / region / business type / waybill number, etc.), field data types, filter values, filter operators (equal to / greater than / contains), condition selection order, condition effective status, and number of query executions, to reconstruct the combination logic and usage habits of user query conditions; Collect menu / tab relationships, pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes (summary table / detail table), and construct the page topology of the report system to provide a foundation for the model to understand page business relationships; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row position, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering, to accurately capture the related behaviors triggered by user data viewing and navigation.

[0030] A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time. Data preprocessing is performed through ETL tools to remove invalid logs (jump timeout / abnormal exit), remove duplicate data, fill missing values ​​(fill with business average / default values), and filter noisy data (test account / bot operation data) to ensure data quality. The cleaned data is divided into training, validation, and test sets in a 7:2:1 ratio. The training set is used for model parameter learning, the validation set is used for model tuning, and the test set is used for model performance evaluation to ensure that the model has generalization ability and can adapt to different users' report navigation behavior.

[0031] 202. Perform multi-dimensional feature extraction and standardization on the training dataset; In this embodiment, user-dimensional features are extracted, including user ID, user business role (operations / finance / administrator), user report usage frequency, and user historical jump preference tags. By associating historical behavior with user unique identifiers, the operating habits of different users are distinguished, and personalized jump prediction is achieved. Extract page scene features, including current menu ID, current tab index, page type (summary / detail), target page popularity, menu level depth, and tab page switching frequency, to characterize the business attributes and structural relationships of the report page, enabling the model to understand the relationship logic between pages; Extract query condition features, including query field name, field data type (numeric / text / time), filter value, number of condition combinations, frequency of condition usage, and data volume of condition filtering results. Quantify the business importance and usage frequency of query conditions to provide a basis for condition filtering. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence. Use sequence embedding to convert discrete behaviors into continuous vectors, preserve the temporal nature of user operations, and adapt to the sequence processing characteristics of the Transformer model. Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations, and strengthen the model's ability to identify high-frequency behaviors and core conditions through statistical regularities; Discrete features are encoded using one-hot or ordinal encoding, continuous features are processed by min-max normalization, and effective features are screened through variance analysis and mutual information coefficients, while redundant features are eliminated, thereby reducing the computational complexity of the model and improving prediction efficiency.

[0032] 203. Construct a jump intent prediction model and a query condition importance ranking model, and train the models; In this embodiment, a jump intent prediction model is constructed, and the specific process includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

[0033] 204. The trained model is compressed using model quantization and channel pruning techniques, converting floating-point parameters into integer parameters and removing redundant neurons. 205. Convert the compressed dual model into the ONNX universal format, integrate the TensorRT inference acceleration engine, encapsulate it into a standardized API interface, and provide jump intent prediction and condition importance ranking services. 206. Deploy the inference engine to the public service layer of the reporting system, and interface with the query condition collection module and the jump relationship configuration module to receive front-end page behavior data and output prediction results in real time; 207. Establish a model caching mechanism to cache prediction results for high-frequency users and pages; 208. Design an incremental training and hot update mechanism to periodically collect new user behavior data to incrementally fine-tune the model; 209. Define page redirection rules, unify the logic for collecting query conditions, and design a standardized parameter passing format; 210. Bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support visually adding or modifying jump relationships through the configuration terminal; 211. When a user triggers a redirect, two sets of models are invoked, and the user is redirected to the target page with standardized routes and core parameters. If the redirect fails, the fault tolerance module is activated to automatically fall back and retry.

[0034] Please see Figure 3 The third embodiment of the configuration-based intelligent jump method for reports based on deep learning in this invention includes: 301. Collect user data from the reporting system and construct a training dataset; In this embodiment, user jump logs are collected, which collect all user jump behavior data in the reporting system. The core fields include user unique identifier, jump source page (menu ID / Tab index / page type), jump target page, jump trigger time, jump time, jump trigger method (manual click / configured jump / intelligent recommendation), and jump result (success / failure / abnormal). It covers two core scenarios: jumps to multiple tabs within the same menu and jumps across menu pages, and fully records the user's real jump preferences. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names (time / region / business type / waybill number, etc.), field data types, filter values, filter operators (equal to / greater than / contains), condition selection order, condition effective status, and number of query executions, to reconstruct the combination logic and usage habits of user query conditions; Collect menu / tab relationships, pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes (summary table / detail table), and construct the page topology of the report system to provide a foundation for the model to understand page business relationships; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row position, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering, to accurately capture the related behaviors triggered by user data viewing and navigation.

[0035] A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time. Data preprocessing is performed through ETL tools to remove invalid logs (jump timeout / abnormal exit), remove duplicate data, fill missing values ​​(fill with business average / default values), and filter noisy data (test account / bot operation data) to ensure data quality. The cleaned data is divided into training, validation, and test sets in a 7:2:1 ratio. The training set is used for model parameter learning, the validation set is used for model tuning, and the test set is used for model performance evaluation to ensure that the model has generalization ability and can adapt to different users' report navigation behavior.

[0036] 302. Perform multi-dimensional feature extraction and standardization on the training dataset; In this embodiment, user-dimensional features are extracted, including user ID, user business role (operations / finance / administrator), user report usage frequency, and user historical jump preference tags. By associating historical behavior with user unique identifiers, the operating habits of different users are distinguished, and personalized jump prediction is achieved. Extract page scene features, including current menu ID, current tab index, page type (summary / detail), target page popularity, menu level depth, and tab page switching frequency, to characterize the business attributes and structural relationships of the report page, enabling the model to understand the relationship logic between pages; Extract query condition features, including query field name, field data type (numeric / text / time), filter value, number of condition combinations, frequency of condition usage, and data volume of condition filtering results. Quantify the business importance and usage frequency of query conditions to provide a basis for condition filtering. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence. Use sequence embedding to convert discrete behaviors into continuous vectors, preserve the temporal nature of user operations, and adapt to the sequence processing characteristics of the Transformer model. Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations, and strengthen the model's ability to identify high-frequency behaviors and core conditions through statistical regularities; Discrete features are encoded using one-hot or ordinal encoding, continuous features are processed by min-max normalization, and effective features are screened through variance analysis and mutual information coefficients, while redundant features are eliminated, thereby reducing the computational complexity of the model and improving prediction efficiency.

[0037] 303. Construct a jump intent prediction model and a query condition importance ranking model, and train the models; In this embodiment, a jump intent prediction model is constructed, and the specific process includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

[0038] 304. Lightweight the two trained models to form an inference engine and embed it into the public service layer of the reporting system; 305. Define rules for linking multiple tabs within the same menu, and establish standardized jump paths between different tabs under the same menu; 306. Define seamless navigation rules across menu pages and establish standardized navigation paths between pages with different menus; 307. Unify the logic for collecting query conditions, and automatically extract time, region, business type, and key field values; 308. Design a standardized JSON parameter format that supports cross-page and cross-service compatible parsing.

[0039] 309. Bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support visually adding or modifying jump relationships through the configuration terminal; 310. When a user triggers a redirect, two sets of models are invoked, and the user is redirected to the target page with standardized routes and core parameters. If the redirect fails, the fault tolerance module is activated to automatically fall back and retry.

[0040] Please see Figure 4 The fourth embodiment of the intelligent navigation method for configurable reports based on deep learning in this invention includes: 401. Collect user data from the reporting system and construct a training dataset; In this embodiment, user jump logs are collected, which collect all user jump behavior data in the reporting system. The core fields include user unique identifier, jump source page (menu ID / Tab index / page type), jump target page, jump trigger time, jump time, jump trigger method (manual click / configured jump / intelligent recommendation), and jump result (success / failure / abnormal). It covers two core scenarios: jumps to multiple tabs within the same menu and jumps across menu pages, and fully records the user's real jump preferences. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names (time / region / business type / waybill number, etc.), field data types, filter values, filter operators (equal to / greater than / contains), condition selection order, condition effective status, and number of query executions, to reconstruct the combination logic and usage habits of user query conditions; Collect menu / tab relationships, pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes (summary table / detail table), and construct the page topology of the report system to provide a foundation for the model to understand page business relationships; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row position, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering, to accurately capture the related behaviors triggered by user data viewing and navigation.

[0041] A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time. Data preprocessing is performed through ETL tools to remove invalid logs (jump timeout / abnormal exit), remove duplicate data, fill missing values ​​(fill with business average / default values), and filter noisy data (test account / bot operation data) to ensure data quality. The cleaned data is divided into training, validation, and test sets in a 7:2:1 ratio. The training set is used for model parameter learning, the validation set is used for model tuning, and the test set is used for model performance evaluation to ensure that the model has generalization ability and can adapt to different users' report navigation behavior.

[0042] 402. Perform multi-dimensional feature extraction and standardization on the training dataset; In this embodiment, user-dimensional features are extracted, including user ID, user business role (operations / finance / administrator), user report usage frequency, and user historical jump preference tags. By associating historical behavior with user unique identifiers, the operating habits of different users are distinguished, and personalized jump prediction is achieved. Extract page scene features, including current menu ID, current tab index, page type (summary / detail), target page popularity, menu level depth, and tab page switching frequency, to characterize the business attributes and structural relationships of the report page, enabling the model to understand the relationship logic between pages; Extract query condition features, including query field name, field data type (numeric / text / time), filter value, number of condition combinations, frequency of condition usage, and data volume of condition filtering results. Quantify the business importance and usage frequency of query conditions to provide a basis for condition filtering. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence. Use sequence embedding to convert discrete behaviors into continuous vectors, preserve the temporal nature of user operations, and adapt to the sequence processing characteristics of the Transformer model. Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations, and strengthen the model's ability to identify high-frequency behaviors and core conditions through statistical regularities; Discrete features are encoded using one-hot or ordinal encoding, continuous features are processed by min-max normalization, and effective features are screened through variance analysis and mutual information coefficients, while redundant features are eliminated, thereby reducing the computational complexity of the model and improving prediction efficiency.

[0043] 403. Construct a jump intent prediction model and a query condition importance ranking model, and train the models; In this embodiment, a jump intent prediction model is constructed, and the specific process includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

[0044] 404. Lightweight the two trained models to form an inference engine and embed it into the public service layer of the reporting system; 405. Define page redirection rules, unify the logic for collecting query conditions, and design a standardized parameter passing format; 406. Bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support visually adding or modifying jump relationships through the configuration terminal; 407. When a user triggers a redirect, the system calls the redirect intent prediction model to output the target page; 408. Query condition importance ranking model: extracts highly important query conditions and removes redundant parameters; 409. Redirect to the target page with standardized routing and core parameters; 410. The target page uses the adaptive parsing module to backfill conditions, silently query and display data; 411. When the jump fails, the fault tolerance module is activated to automatically roll back and retry.

[0045] Please see Figure 5 One embodiment of the configurable report intelligent redirection device based on deep learning in this invention includes: Data acquisition module 501 is used to collect user data within the reporting system and build a training dataset; Feature engineering module 502 is used to perform multi-dimensional feature extraction and standardization processing on the training dataset; The model training module 503 is used to build a jump intent prediction model and a query condition importance ranking model, and to train the models. The model deployment module 504 is used to lightweight process the two trained models to form an inference engine and embed it into the public service layer of the reporting system. Rule definition module 505 is used to define page redirection rules, unify the logic for collecting query conditions, and design standardized parameter passing formats; The configuration binding module 506 is used to bind the basic association between menus, tabs and target pages on the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support adding or modifying jump relationships visually through the configuration terminal. The redirection processing module 507 is used to call two sets of models and carry standardized routes and core parameters to redirect to the target page when the user triggers a redirection operation. If the redirection fails, the fault tolerance module is started to automatically fall back and retry.

[0046] Figure 6 This is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present invention. The electronic device 600 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the electronic device 600.

[0047] Electronic device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 6 The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0048] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a deep learning-based configurable report intelligent jump method.

[0049] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0050] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0051] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A configuration-based intelligent report navigation method based on deep learning, characterized in that, The method includes: Collect user data from the reporting system and build a training dataset; Multi-dimensional feature extraction and standardization are performed on the training dataset; Construct a redirection intent prediction model and a query condition importance ranking model, and train the models; The two trained models are lightweighted and processed to form an inference engine, which is then embedded into the public service layer of the reporting system. Define page redirection rules, unify the logic for collecting query conditions, and design a standardized parameter passing format; The configuration terminal binds the basic association between menus, tabs and target pages, recommends candidate jump targets in real time based on the jump intent prediction model, and supports visually adding or modifying jump relationships through the configuration terminal. When a user triggers a redirect, two sets of models are invoked, carrying standardized routes and core parameters to redirect to the target page. If the redirect fails, the fault tolerance module is activated to automatically fall back and retry.

2. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, The user data collected from the reporting system is used to construct a training dataset, including: Collect user redirection logs, collecting all user redirection behavior data within the reporting system. Core fields include user unique identifier, redirection source page, redirection target page, redirection trigger time, redirection time, redirection trigger method, and redirection result. Collect query condition sequences and collect data on the entire query process of users on the report page, including query field names, field data types, filter values, filter operators, condition selection order, condition validity status, and number of query executions. Collect menu / tab relationships and pull system configuration data from the report configuration end, including menu hierarchy, tab page ownership, page routing address, preset jump relationships, and page business attributes; Collect data click behavior, including user interaction behavior in the report data area, such as clicked data row location, clicked field name, data row business attribute, click frequency, page dwell time, and click probability after data filtering. A distributed log collection service is used to collect front-end event tracking data and back-end operation logs in real time, and data preprocessing is completed through ETL tools. The cleaned data was divided into training, validation, and test sets in a 7:2:1 ratio.

3. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, The multi-dimensional feature extraction and standardization process for the training dataset includes: Extract user-dimensional features, including user ID, user business role, frequency of user report usage, and user historical navigation preference tags; Extract page scene features, including current menu ID, current tab index, page type, target page popularity, menu level depth, and tab page switching frequency; Extract query condition features, including query field name, field data type, filter value, number of condition combinations, frequency of condition usage, and data volume of the filtered results. Extract behavioral sequence features, including click location sequence, jump order sequence, query condition input sequence, and page dwell time sequence, and use sequence embedding to convert discrete behaviors into continuous vectors; Extract statistical features, including user page jump frequency, historical usage rate of query conditions, page jump relevance, and hit rate of condition combinations; Discrete features are encoded using one-hot encoding or ordinal encoding, continuous features are processed using min-max normalization, and effective features are selected through variance analysis and mutual information coefficients.

4. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, The construction of the redirection intent prediction model and the query condition importance ranking model, and the training of the models, include: The specific process of building a redirection intent prediction model includes: This paper employs a Transformer architecture to process user behavior time-series data, converting user behavior sequence features into word embedding vectors and adding positional encoding to preserve temporal information. It utilizes a 6-layer multi-head self-attention mechanism and a feedforward neural network to extract global correlation features of user behavior in parallel, capturing behavioral logic. This behavioral logic, in sequence, includes: query condition setting, data click and jump triggering. The feedforward network performs non-linear feature transformation, and a fully connected layer and Softmax activation function output the predicted probabilities of all candidate target pages. The page with the highest probability is selected as the model's recommended jump target. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer updates parameters. The learning rate and batch size are adjusted using a validation set. The specific process of constructing a query condition importance ranking model includes: The self-attention mechanism is used to extract the correlation features between query conditions, identify the weight of high-frequency combination conditions, input the self-attention extracted features into the LightGBM model, use the query hit rate after condition propagation as the label, train the condition importance ranking model, output the weight score of each query condition, sort them from high to low weight scores, select the Top-K core conditions as jump propagation parameters, and eliminate redundant conditions. Accuracy, recall, F1 score, and ranking error were used as evaluation metrics. The model training was completed when the accuracy of the jump intent prediction model reached the target and the ranking error of the conditional importance model was ≤5%.

5. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, The process of lightweighting the trained model to form an inference engine and embedding it into the public service layer of the reporting system includes: Model quantization and channel pruning techniques are used to compress the trained model, convert floating-point parameters into integer parameters, and remove redundant neurons. The compressed dual model is converted into the ONNX universal format, integrated with the TensorRT inference acceleration engine, and encapsulated into a standardized API interface to provide jump intent prediction and condition importance ranking services. Deploy the inference engine to the public service layer of the reporting system, and interface with the query condition collection module and the jump relationship configuration module to receive front-end page behavior data and output prediction results in real time; Establish a model caching mechanism to cache prediction results for high-frequency users and pages; The design incorporates incremental training and hot update mechanisms, periodically collecting new user behavior data to incrementally fine-tune the model.

6. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, The defined page redirection rules unify the query condition collection logic and design a standardized parameter passing format, including: Define rules for linked navigation between multiple tabs within the same menu, and establish standardized navigation paths between different tabs under the same menu. Define rules for seamless navigation across menu pages and establish standardized navigation paths between pages with different menus; Unify the logic for collecting query conditions, and automatically extract time, region, business type, and key field values; Design a standardized JSON parameter format that supports cross-page and cross-service compatibility parsing.

7. The intelligent navigation method for configurable reports based on deep learning according to claim 1, characterized in that, When a user triggers a redirect, two sets of models are invoked, carrying standardized routes and core parameters to redirect to the target page. If the redirect fails, a fault tolerance module is activated to automatically fall back and retry, including: When a user triggers a redirect, the system calls the redirect intent prediction model to output the target page; A query condition importance ranking model is used to extract highly important query conditions and remove redundant parameters. Redirecting to the target page with standardized routes and core parameters; The target page uses the adaptive parsing module to backfill conditions, silently query and display data; If the jump fails, the fault tolerance module is activated to automatically roll back and retry.

8. A configurable report intelligent redirection device based on deep learning, characterized in that, include: The data acquisition module is used to collect user data within the reporting system and build a training dataset. The feature engineering module is used to perform multi-dimensional feature extraction and standardization on the training dataset; The model training module is used to build a jump intent prediction model and a query condition importance ranking model, and to train the models. The model deployment module is used to lightweight process the two trained models to form an inference engine and embed it into the public service layer of the reporting system. The rule definition module is used to define page redirection rules, unify the logic for collecting query conditions, and design standardized parameter passing formats; The configuration binding module is used to bind the basic association between menus, tabs and target pages in the configuration terminal, recommend candidate jump targets in real time based on the jump intent prediction model, and support adding or modifying jump relationships visually through the configuration terminal. The redirection processing module is used to call two sets of models and redirect to the target page with standardized routes and core parameters when a user triggers a redirection operation. If the redirection fails, the fault tolerance module is activated to automatically fall back and retry.

9. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the deep learning-based configurable report intelligent navigation method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor, implement the steps of the deep learning-based intelligent navigation method for configurable reports as described in any one of claims 1-7.

Images