An experimental teaching natural language data query interaction method

Abstract

This invention relates to the field of computer application technology, specifically to a method for interactive natural language data query in experimental teaching, comprising the following steps: S100, in response to a user's login request, obtaining the user's role permissions, and dynamically rendering a three-zone interactive interface on the terminal display interface based on the role permissions. This invention constructs a three-zone dynamic interactive architecture comprising intelligent query, a data dashboard, and a centralized workbench, combined with an experimental teaching knowledge graph and a multi-round slot verification mechanism, to accurately map unstructured natural language commands into cross-domain federated queries and indicator calculation tasks limited by data permission boundaries, effectively solving the problems of difficult cross-domain data retrieval and isolated computation in traditional systems; simultaneously, it achieves contextual state fusion between data retrieval status and business flow status.

Description

Technical Field

[0001] This invention relates to the field of computer application technology, specifically to a method for interactive natural language data query in experimental teaching. Background Technology

[0002] With the development of educational informatization, primary and secondary school experimental teaching and laboratory management are increasingly reliant on digital systems. Currently, schools and education management departments have generally deployed comprehensive laboratory management systems for managing personnel files, equipment assets, and scheduling. However, in actual, complex experimental teaching management scenarios, existing data query and business interaction methods face significant technical bottlenecks.

[0003] Existing laboratory management systems typically employ traditional, deeply hierarchical, tree-structured graphical user interfaces (GUIs), and their underlying data architecture is often domain-isolated (e.g., personnel, asset, and academic affairs databases are independent, lacking a unified computational graph). When administrators need comprehensive, cross-dimensional decision-making information (e.g., querying "the opening rate of the eighth-grade physics group experiments this semester and the available inventory gap of optical benches"), the system cannot directly respond to such complex query requests that integrate academic progress, asset status, and macro-level evaluation indicators. Users must separately access the scheduling module, inventory management module, and evaluation module, manually export multi-source, heterogeneous underlying detailed data tables, and then rely on manual table structure assembly and secondary calculation of indicator formulas, resulting in extremely low system query efficiency and a high risk of human statistical omissions.

[0004] The more fundamental problem lies in the inherent disconnect between the "search, view, and process" processes in existing query interaction mechanisms. In traditional system architectures, the data query module only possesses unidirectional data extraction and static list display capabilities. When users discover business anomalies or resource gaps through querying, the existing interaction logic cannot automate the flow of data within the system based on the query result context. This prevents the data query results from driving updates to the business operation interface or macro-monitoring interface. Users must interrupt their current query session, manually switch, and re-navigate to a specific business flow node (such as the new allocation order form interface) to perform repetitive data entry operations. This disconnect between the query state and the business execution state significantly increases the human-computer interaction cost of the system, failing to meet the technical requirements of data-driven real-time decision-making and agile business flow in modern laboratory governance scenarios. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for querying and interacting with natural language data in experimental teaching, comprising the following steps: S100, in response to a user's login request, obtaining the user's role permissions, and dynamically rendering a three-zone interactive interface on the terminal display interface based on the role permissions; the three-zone interactive interface includes at least: a centralized workbench area for the flow of experimental teaching business processes, a data cockpit area for visually monitoring experimental business indicators, and an intelligent query area for receiving unstructured data. S200, receiving natural language query commands about experimental teaching scenarios input by the user through the intelligent query area; S300, invoke a preset natural language processing model to perform semantic parsing on the natural language query instruction, extract the instruction intent and the business entity features corresponding to the instruction intent; the business entity features include at least one of the following: business action category, organizational hierarchy feature, time period feature, and experimental discipline feature; S400, based on the instruction intent and the characteristics of the business entity, performs cross-domain data retrieval in the underlying interconnected personnel database, asset database and academic affairs database, and calls the preset business rule engine to perform compliance comparison and indicator calculation on the retrieved basic data to generate structured query results; S500, in the intelligent inquiry area, outputs natural language feedback text containing the query results in the form of dialogue interaction, and dynamically updates the corresponding data charts in the data cockpit area according to the indicator attributes of the query results, and / or automatically generates a to-do business entry related to the instruction intent in the centralized workbench area.

[0006] Furthermore, based on the aforementioned role permissions, the three-zone interactive interface is dynamically rendered on the terminal display interface, specifically including: The role permissions are parsed to determine the organizational structure data tag to which the user belongs. The organizational structure data tag includes the management decision-making level, the business execution level, and the terminal application level. Retrieve the interface configuration template uniquely bound to the organizational structure data tag from the preset UI configuration library; Based on the interface configuration template, the corresponding business micro-frontend components and data loading interfaces are dynamically assembled to generate the three-zone interactive interface adapted to the user.

[0007] Furthermore, a workflow engine module is embedded in the integrated workbench area; When rendering the centralized workbench area, component filtering is performed based on the user's role permissions, and non-business function nodes that do not belong to the current permission chain are blocked. Furthermore, the experimental teaching business processes activated under this permission are aggregated and presented in the form of form interaction components within the centralized workbench area to provide a one-way data flow operation environment. When rendering the physical layout of the three-zone interactive interface: The centralized workbench area and the intelligent inquiry area are respectively anchored to the left and right edges of the terminal display interface, and the data cockpit area is laid out as the underlying default view in the center area of ​​the terminal display interface. The intelligent inquiry area is configured as a floating dialog box. When the intelligent inquiry area is activated to receive natural language query commands, the core indicator charts in the data cockpit area adaptively adjust their scaling ratio.

[0008] Furthermore, the system receives natural language query commands from the user regarding experimental teaching scenarios through the intelligent query area, including: When the intelligent query area is in query receiving state, listen for the global focus event of the terminal display interface; The current form context data of the centralized workbench area and the current view filtering parameters of the data cockpit area are collected synchronously and packaged to generate a front-end scene snapshot; The front-end scene snapshot is used as an implicit query condition and bound and encapsulated with the natural language query command explicitly input by the user to form a multi-dimensional joint query request.

[0009] Furthermore, the intelligent inquiry area is equipped with a multimodal input interface, and the receipt of natural language query commands input by the user regarding the experimental teaching scenario specifically includes: The multimodal input interface receives a continuous voice stream from the user and calls an endpoint detection algorithm to identify valid voice segments in the continuous voice stream. Convert the effective speech segments into initial text data; Load a pre-set experimental teaching professional thesaurus, perform error correction and synonym mapping on the initial text data for specific domain entities, and output the natural language query command in a standardized text form; Before receiving the user's natural language query command about the experimental teaching scenario through the intelligent query area, the method further includes: Extract the user's role permissions, and combine the user's historical query logs with the current business time node to pre-generate several candidate query intent templates using a prediction algorithm; When the intelligent inquiry area is detected to be activated, the candidate inquiry intent template is rendered dynamically in front of the input box of the intelligent inquiry area in the form of a quick interaction component, in response to the user's click input or auxiliary statement completion.

[0010] Furthermore, the step of invoking a preset natural language processing model to perform semantic parsing on the natural language query instruction, extracting the instruction intent and the business entity features corresponding to the instruction intent, specifically includes: The natural language query command is converted into a word vector sequence, and the word vector sequence is input into a pre-trained sequence labeling network for local feature extraction to obtain initial entity text containing basic material vocabulary and basic course vocabulary. The pre-constructed knowledge graph of the experimental teaching domain is invoked, which contains the national curriculum standard experimental catalog and the Ministry of Education equipment classification code standard library. Calculate the semantic similarity between the initial entity text and the node entities in the experimental teaching domain knowledge graph, and perform entity alignment operations to map the non-standard initial entity text to the standardized business entity features; Based on the standardized business entity features and the word vector sequence, the intent probability distribution set in the multidimensional space is calculated using an intent classification network, and the category corresponding to the highest probability is determined as the instruction intent.

[0011] Furthermore, after extracting the instruction intent and the business entity features corresponding to the instruction intent, the method also includes a dynamic slot verification step: Based on the identified instruction intent, the corresponding intent-slot mapping template is invoked to determine the set of required business entity slots for executing the instruction intent; Compare the extracted business entity features with the set of required business entity slots. If a missing required business entity is detected, the current semantic parsing process is suspended and a front-end scene snapshot is cached. In the intelligent query area, a follow-up prompt is generated for the missing required business entity. When a supplementary instruction based on the follow-up prompt is received, the entity features in the supplementary instruction are extracted and filled into the missing required business entity slot, and the semantic parsing process is resumed. The business action categories in the business entity characteristics include at least resource operation and maintenance actions, performance evaluation and calculation actions, and security investigation actions; When semantic parsing determines that the business action category belongs to the evaluation calculation action, fine-grained evaluation indicator labels are further parsed from the natural language query instruction based on the associated organizational hierarchy features and experimental discipline features. The evaluation indicator label is configured as a soft switch to activate the corresponding indicator calculation engine at the underlying level during subsequent cross-domain data retrieval; the indicator calculation engine is pre-configured with an algorithm for calculating the full opening rate of experiments, an algorithm for calculating the asset value per student, and an algorithm for calculating the equilibrium index.

[0012] Furthermore, the cross-domain data retrieval in the underlying interconnected personnel database, asset database, and academic affairs database specifically includes: Obtain the organizational structure data tags and corresponding row-level data access permission boundaries registered by the user in the data exchange bus; The instruction intent and the business entity characteristics are compiled into distributed federated query statements, and concurrently routed to the personnel database, the asset database and the academic affairs database through the data exchange bus; Under the constraints of the row-level data access permission boundary, each database engine performs asynchronous query operations and performs primary key mapping, cleaning and fusion of the returned multi-source heterogeneous subset datasets to generate an intermediate basic dataset to support the current query session; The process of calling a preset business rule engine to perform compliance comparisons and indicator calculations on the retrieved basic data includes concurrent computation logic for the experiment preparation scenario: When the instruction intent is identified as an experiment reservation or asset availability query, the standard instrument BOM list and the total number of students to be taught for the target experimental course are extracted from the intermediate basic dataset. Simultaneously extract real-time available inventory data and in-transit borrowing data from the asset database that match the standard instrument BOM list; The resource scheduling algorithm engine is invoked to calculate the difference between the target total number of students and the comprehensive available inventory, generating a resource gap status matrix, and encapsulating intelligent allocation suggestions for the resource gap status matrix into the structured query results.

[0013] Furthermore, the step of calling a preset business rule engine to perform compliance comparison and indicator calculation on the retrieved basic data also includes indicator derivation logic for comprehensive regulatory scenarios: When the instruction intent is identified as a regional compliance assessment, the historical experiment execution logs, equipment asset lists, and personnel staffing data contained in the intermediate basic dataset are input into the built-in regulatory comparison model; Based on the national curriculum standard baseline pre-set in the business rules engine, the experiment completion rate parameter and per-student asset balance index are calculated through the regulatory comparison model for the characteristics of the business entity. The experimental completion rate parameter and the per capita asset balance index are quantitatively graded and determined based on a preset multi-level threshold network, and a visual warning status code corresponding to the risk level is attached to the structured query results.

[0014] Furthermore, based on the indicator attributes of the query results, the corresponding data charts in the data dashboard area are dynamically updated, specifically including: Extract the data dimension labels and accompanying visual warning status codes from the structured query results; Within the DOM tree structure of the data cockpit area, traverse and match the target visualization chart components bound to the data dimension labels; Based on the visual warning status code, an asynchronous partial refresh instruction is sent to the target visualization chart component to drive the target visualization chart component to perform independent data frame redrawing; and for data nodes that reach the preset alarm threshold, coordinate system centering and view highlighting operations are performed. The automatic generation of a pending task entry associated with the instruction intent in the centralized workbench area specifically includes: When the instruction intent is determined by preset rules to be an action instruction with closed-loop operation attributes, the set of key parameters required for business flow is extracted from the structured query results. The workflow engine module embedded in the centralized workbench area is invoked to silently instantiate and assign values ​​to the target business form template using the set of key parameters, generating an instance form with pre-filled data. The instance form is encapsulated as a quick to-do interactive card and pushed to the top-level view stack of the centralized workbench area so that the user can confirm or supplement the data flow with one click; The intelligent query area outputs natural language feedback text containing the query results in the form of a dialogue interaction, including: The natural language generation engine is invoked to match the corresponding response strategy template in the pre-set corpus template library based on the data structure characteristics of the query results. The numerical indicators and entity names in the structured query results are used as dynamic variables and injected into the corresponding placeholders in the response strategy template to generate feedback text that conforms to natural language syntax. The natural language feedback text output in the current round, the associated structured query results, and the corresponding instruction intent are packaged into an interaction context package and written into the global state cache to maintain contextual memory constraints for subsequent continuous query instructions.

[0015] Beneficial effects This invention constructs a three-zone dynamic interactive architecture comprising intelligent inquiry, a data dashboard, and a centralized workbench. Combined with an experimental teaching knowledge graph and a multi-round slot verification mechanism, it accurately maps unstructured natural language commands into cross-domain federated queries and indicator calculation tasks constrained by data permission boundaries. This effectively solves the problems of difficult cross-domain data retrieval and isolated computation in traditional systems. Simultaneously, this invention establishes an asynchronous linkage mechanism between query results and front-end components. Utilizing visual warning status codes and business intent identifiers in structured query messages, it automatically triggers partial redrawing of charts in the monitoring view and dynamic generation of pending business entry points on the workbench. This achieves contextual state fusion between data retrieval status and business flow status, significantly reducing the user's interface interaction layers and operational costs, and effectively improving the intelligence level and business collaboration efficiency of the experimental teaching system in complex data scheduling scenarios. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the three-zone dynamic rendering and spatial layout of the terminal display interface of the present invention; Figure 3 This is a flowchart of the natural language instruction semantic parsing and entity extraction process of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., 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 of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "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 includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] The present invention will now be described in further detail with reference to the accompanying drawings: Example: like Figures 1-3 As shown, an experimental teaching natural language data query and interaction method includes the following steps: S100 responds to the user's login request, obtains the user's role permissions, and dynamically renders the three-zone interactive interface on the terminal display interface based on the role permissions; the three-zone interactive interface includes at least: a centralized workbench area for the flow of experimental teaching business processes, a data cockpit area for visual monitoring of experimental business indicators, and an intelligent query area for receiving unstructured data. The S200 receives natural language query commands from users regarding experimental teaching scenarios through an intelligent query area. S300, invokes a preset natural language processing model to perform semantic parsing on the natural language query command, extracts the command intent and the business entity features corresponding to the command intent; the business entity features include at least one of the following: business action category, organizational hierarchy feature, time period feature, and experimental discipline feature; S400, based on instruction intent and business entity characteristics, performs cross-domain data retrieval in the underlying interconnected personnel database, asset database and academic affairs database, and calls the preset business rule engine to perform compliance comparison and indicator calculation on the retrieved basic data to generate structured query results; The S500 outputs natural language feedback text containing query results in the form of dialogue interaction in the intelligent inquiry area, and dynamically updates the corresponding data charts in the data cockpit area based on the indicator attributes of the query results, and / or automatically generates to-do business entry points associated with the command intent in the centralized workbench area.

[0020] Furthermore, the specific implementation process of step S100 is as follows: Step S100 specifically includes the initialization and dynamic rendering process of the terminal display interface. In response to a login request initiated by the user terminal, the server node parses the request message to extract the user identifier and performs matching mapping in the underlying permission database to extract the role permission data bound to the user identifier. The system's internal permission parsing service reads the role permission data and maps it to the corresponding organizational structure data tags. The organizational structure data tags are explicitly divided into management decision-making level, business execution level, and terminal application level in the data dictionary. Specifically, in the experimental teaching management scenario, the system maps the municipal / county-level education bureau management account or principal account to the management decision-making level, the academic affairs director account and laboratory administrator account to the business execution level, and the front-line teaching teacher account to the terminal application level.

[0021] After obtaining the defined organizational structure data tags, the system front-end framework sends a template request command to the preset UI configuration center. The UI configuration center retrieves and returns the interface configuration template uniquely bound to the organizational structure data tags. The interface configuration template specifically uses JSON or XML data format to describe the page's DOM tree structure data, the hierarchical nesting relationships of each block, and the initial coordinate parameters. The front-end rendering engine reads the interface configuration template and performs physical space division and rendering operations for the three-zone interactive interface on the terminal display interface. In the physical space division, the front-end rendering engine obtains the absolute resolution size of the terminal display interface, anchors the centralized workbench area used for the flow of experimental teaching business processes to the left edge of the terminal display interface, anchors the intelligent query area used for receiving unstructured data to the right edge of the terminal display interface, and calculates the pixel width occupied by the left and right edges. The remaining space is configured as the data cockpit area for visually monitoring experimental business indicators, and the data cockpit area is laid out as the underlying default view at the center of the terminal display interface.

[0022] After completing the physical space division, the system executes the dynamic assembly logic of the micro-frontend components. For the centralized workbench area on the left, the system instantiates and embeds the workflow engine module in the container node of the centralized workbench area. The workflow engine module reads the user's role permission set and traverses the system's pre-built full library of business function nodes. During the traversal operation, the workflow engine module executes component filtering logic, comparing the permission requirements of the currently read business function node with the user's role permission set, and blocking non-business function nodes not included in the role permission set, intercepting the loading requests of the corresponding frontend code package. For experimental teaching business process nodes activated under the current permissions, such as purchase order approval nodes, hazardous chemical outbound registration nodes, or experimental scheduling nodes, the workflow engine module calls the corresponding form interaction components and aggregates and presents them in the left view of the centralized workbench area, forming a one-way data flow and form submission process.

[0023] For the central data dashboard area, the system calls the visualization rendering library and data loading interface to retrieve underlying real-time monitoring indicator data and instantiates chart components to fill the central view. For the intelligent inquiry area on the right, the system configures it as a dialog box floating above the central view using CSS properties. The system binds a global focus listener event to the intelligent inquiry area in the front-end DOM tree. When an external input device clicks the input box in the intelligent inquiry area, or the voice capture module outputs an activation signal, the system triggers the focus listener event, determining that the intelligent inquiry area has entered the awakened state. In response to the awakened state, the front-end rendering engine sends an adaptive scaling command to the main container of the central data dashboard area. After receiving the scaling command, the data dashboard area reads the preset avoidance margin parameters, recalculates the width and height attributes of each chart element inside, and calls the redraw interface to perform a view scaling operation to avoid the screen physical coordinates covered by the intelligent inquiry area, preventing the underlying data chart elements in the central area from being obscured by the active intelligent inquiry area.

[0024] Furthermore, the specific implementation process of step S200 is as follows: When the intelligent inquiry area is in the initialization loading phase or in an idle input state, the system background process reads the role and permission data of the currently logged-in user and extracts the associated historical query log records from the local storage layer. Simultaneously, it calls the underlying system clock service to obtain the characteristics of the current business time node, such as a mid-term evaluation node or a year-end asset inventory node. The system's core service components use the extracted role and permission data, historical query records, and business time node characteristics as input vectors, feeding them into a preset probability prediction algorithm model. This model calculates and outputs several candidate inquiry intent template texts ranking high in probability distribution. The front-end rendering engine receives the generated candidate inquiry intent template texts and instantiates them as quick interaction label components on the DOM nodes surrounding the input text box in the intelligent inquiry area, performing pre-rendering. When a trigger event from an external input device on the quick interaction label component is captured, the front-end script extracts the associated template text and directly writes it into the main value field of the input text box to complete auxiliary statements.

[0025] The intelligent inquiry area is configured with a multimodal data receiving interface, including text data stream and audio data stream interfaces. When receiving continuous speech stream input from the user through the audio data stream interface, the underlying audio processing service component of the system resamples and performs frame-by-frame windowing operations on the continuous speech stream. It then calls the endpoint detection algorithm to extract the short-time energy value and short-time zero-crossing rate parameters of each audio frame, and strips background silence segments and environmental noise segments according to a preset energy threshold, extracting the valid speech segments. The audio processing service component converts the valid speech segments into an initial text data string. After obtaining the initial text data string, the text processing engine loads the experimental teaching professional vocabulary database deployed in memory. The experimental teaching professional vocabulary database contains a mapping table of standard instrument names and common colloquial terms. The text processing engine uses string matching algorithms and edit distance algorithms to perform entity comparison and error correction operations on the initial text data string. At the same time, it performs synonym replacement based on the mapping table, mapping non-standard expressions to standard words, thereby outputting natural language query instructions in standardized text form.

[0026] When the input text box in the intelligent query area is in an active focused state, the front-end framework calls the browser's underlying API to register a global focus event listener. When the submission trigger event of a natural language query command is captured, the system front-end synchronously dispatches a view state acquisition command. The state acquisition script traverses the front-end component tree in the left-hand centralized workbench area, extracting the form context data of the currently active component, including the value of the currently selected instrument category node, the form business type identifier, and the business flow node level. Simultaneously, the state acquisition script reads the state parameters of each dimension filter in the intermediate data cockpit area, extracting the currently active time dimension value, subject dimension value, and organization code value. The system serializes the collected form context data and view state parameters, and packages them to generate a structured front-end scene snapshot object.

[0027] Before initiating a network request, the front-end network module uses the serialized front-end scene snapshot object as an implicit query condition parameter and the extracted standardized text form of natural language query instructions as explicit query parameters to perform key-value pair binding at the data level. Following a preset communication protocol, the system encapsulates the bound dataset into a multi-dimensional joint query request message containing implicit scene features and explicit text features, and then transmits it through the network channel to the back-end natural language processing platform for subsequent processing steps.

[0028] Furthermore, the specific implementation process of step S300 is as follows: Step S300 specifically includes instruction parsing based on deep learning networks and knowledge graph architecture, as well as a multi-round state machine verification operation process.

[0029] After receiving a multi-dimensional joint query request message, the backend natural language processing platform extracts the natural language query instructions from the message's internal payload, calls a pre-defined word segmentation algorithm library to perform word segmentation, and maps it to a pre-trained word nesting model, transforming it into a continuous sequence of word vectors. The system's core service inputs the word vector sequence into a pre-trained sequence labeling network to perform local feature extraction operations, calculates the state transition probabilities of each sequence node, and outputs the boundary labels and entity category labels corresponding to each character, thereby obtaining an initial entity text set containing basic material vocabulary and basic course vocabulary.

[0030] Subsequently, the system initiates the knowledge graph comparison service, calling the experimental teaching domain knowledge graph stored in the graph database. The experimental teaching domain knowledge graph embeds the national curriculum standard experimental catalog data table and the Ministry of Education equipment classification code standard library as core benchmark nodes within its underlying node architecture. The graph comparison service extracts each unstructured word from the initial entity text set and uses the cosine similarity algorithm to calculate the semantic similarity value between the vector of each unstructured word and the entity feature vector of each node in the experimental teaching domain knowledge graph. When the semantic similarity value exceeds a preset alignment threshold, the system performs an entity alignment operation, forcibly mapping and overwriting the initial entity text belonging to colloquial expressions or abbreviations with the standardized affixes of the corresponding standard nodes in the knowledge graph, outputting a standardized set of business entity features. After obtaining a standardized set of business entity features, the system concatenates and fuses the standardized set of business entity features with the previously generated word vector sequence in the tensor dimension, inputs it into the intent classification network layer to perform forward propagation operation, calculates the intent probability distribution set of the fused feature data in the preset multi-dimensional intent space, extracts the probability item with the largest value in the intent probability distribution set, and confirms its corresponding spatial classification label as the core instruction intent.

[0031] After establishing the instruction intent and the corresponding set of business entity features, the dialogue flow controller process within the system triggers dynamic slot verification logic. Using the instruction intent as the retrieval key, the dialogue flow controller retrieves the corresponding intent-slot mapping template from a pre-set business rule dictionary, reading the set of mandatory business entity slots required to execute the current instruction intent. The system comparison function compares the extracted set of business entity features with the set of mandatory business entity slots. If the comparison result returns a boolean value of false, indicating the presence of missing mandatory business entity parameters in an empty state, the dialogue flow controller sends an interruption suspension signal to the main process, placing the current semantic parsing thread in a blocked state, and serializing and storing the front-end scene snapshot parameters attached to the multi-dimensional joint query request message in the cache. Simultaneously, the natural language generation engine assembles a follow-up question prompt string based on the missing mandatory business entity category tags, and sends it to the intelligent query area for rendering and display via the network communication link. When the system receives a supplementary query instruction based on the follow-up prompt string returned by the terminal, it calls the sequence labeling network again to extract the secondary entity features in the supplementary instruction, fills the secondary entity features into the empty required business entity slots, and then the system kernel releases the blocked state, wakes up and resumes the suspended semantic parsing main thread.

[0032] At the business entity parsing level, the business action category parameters in the business entity feature set are predefined in the system data dictionary as enumeration values ​​for resource operation and maintenance actions, performance evaluation calculation actions, and security investigation actions. When the semantic parsing engine determines that the extracted business action category parameter matches the performance evaluation calculation action enumeration value, the parsing engine initiates a drill-down analysis process. Based on the front-end scene snapshot parameters and the organizational hierarchy and experimental discipline features associated in the instruction text, fine-grained evaluation indicator labels are separated in the abstract syntax tree. The system registers and configures the extracted evaluation indicator labels as global soft-switch variables in the memory heap space. The soft-switch variables are connected to the underlying business microservice cluster via the RPC remote procedure call protocol, and the indicator calculation engine for the target dimension is activated in the subsequent cross-domain data retrieval stage. Instances of experimental completion rate calculation algorithms, per-student asset value calculation algorithms, and equilibrium index calculation algorithms are pre-deployed in the underlying microservice cluster. When the Boolean state of the soft switch variable flips, the corresponding index calculation engine instance immediately switches from the dormant state to the running ready state and listens to the system message queue, ready to take on subsequent parallel computing tasks of multi-source heterogeneous data streams.

[0033] Furthermore, the specific implementation process of step S400 is as follows: Step S400 specifically includes a multi-source heterogeneous data joint scheduling and rule engine operation process based on a federated query architecture.

[0034] The system's core routing service component reads user identifiers pre-registered in the data exchange bus and extracts matching organizational structure data tags and associated row-level data access permission boundary thresholds from the permission boundary configuration table. The system's underlying SQL compilation engine receives the instruction intent and business entity characteristics output from the preceding steps and compiles them into a set of federated query statements supporting distributed execution, according to the database abstraction layer protocol. The system initiates concurrent communication threads through the data exchange bus to synchronously distribute and route the federated query statement set to the personnel database master node, asset database master node, and academic affairs database master node.

[0035] Under the constraints of row-level data access permission boundary thresholds, the query execution plan generators of each database master node schedule the underlying storage engine to perform asynchronous I / O read operations, extract the raw data records that match the query conditions, and return the multi-source heterogeneous subset dataset to the upper layer through the network buffer. The ETL service process of the system middleware layer captures the multi-source heterogeneous subset dataset, performs primary key mapping association, null value filtering, and field type unification and cleaning operations according to the preset global data dictionary, performs table structure fusion calculations on the multi-source heterogeneous subset dataset in memory, and outputs the intermediate basic dataset to support the current query session.

[0036] When the received instruction intent parameter is identified as an experiment reservation action or an asset availability query action, the system triggers the concurrent operation logic of the experiment preparation scenario. The business rule engine within the system calls the parsing component to extract the target experiment course feature identifier from the intermediate basic dataset and retrieves the associated standard instrument BOM list data structure from the built-in standard component model library. Simultaneously, it extracts the target total number of students variable from the sub-dataset of the teaching affairs dimension. Simultaneously, the business rule engine reads the asset dimension information from the intermediate basic dataset and extracts the real-time available inventory data quantity parameter and the in-transit borrowing data quantity parameter for each instrument category that match the standard instrument BOM list data structure. The system's underlying layer calls the resource scheduling algorithm engine to perform matrix subtraction on the target total number of students variable and the comprehensive available inventory quantity, generating a resource gap status matrix containing the shortage difference of each required instrument. The system recommendation algorithm component reads the negative matrix elements in the resource gap status matrix and retrieves the coordinates of adjacent available storage nodes matching the materials from the pre-set allocation strategy rule library, generating a smart allocation suggestion text package. This smart allocation suggestion text package is then serialized and incorporated into the structured query result message.

[0037] When the received instruction intent parameter is identified as a regional compliance assessment action, the system triggers the indicator derivation calculation logic for the comprehensive supervision scenario. The system's business rule engine uses historical experiment execution logs, current equipment asset list snapshots, and staffing data from the intermediate basic dataset as feature input streams, pushing them into a pre-loaded supervision comparison model in memory. The supervision comparison model extracts the pre-set national curriculum standard baseline parameters from the business rule engine. At the algorithm logic layer, the supervision comparison model performs an intersection operation between the set of already opened experiments in the historical experiment execution logs and the set of required experiments in the national curriculum standard baseline parameters, outputting the experiment completion rate parameter value. Simultaneously, the supervision comparison model divides the total asset value parameter in the current equipment asset list snapshot by the total number of registered students, calculating and outputting the per-student asset equilibrium index variable using a pre-set statistical variance formula.

[0038] The system engine further reads the multi-level compliance threshold network structure pre-deployed on the local hard drive. This network structure contains multiple level judgment nodes discretely distributed across numerical ranges. The system engine uses conditional branching logic statements to perform range-based matching operations on the calculated experimental completion rate parameter and the per-student asset equilibrium index variable, executing a quantitative grading judgment. After the matching operation is complete, the system engine extracts the preset warning attributes of the corresponding range and appends the corresponding visual warning status code variable to the header of the structured query result message for subsequent color gamut rendering and status indication calls by the front-end chart components.

[0039] Furthermore, the specific implementation process of step S500 is as follows: Step S500 specifically includes a natural language text generation mechanism based on the query result status and asynchronous DOM node update and chart redraw flow control across front-end regions.

[0040] The system's backend communication module pushes the structured query result message, which includes visual warning status code variables, generated in the preceding steps into the response message queue. The system's internal natural language generation engine subscribes to the response message queue, extracting numerical fields and status enumeration values ​​from the message payload. The engine traverses the pre-loaded semantic template tree in memory, locates the matching branch node based on the extracted status enumeration values, fills the discrete numerical fields into preset string placeholders according to a specific grammatical order, performs character concatenation, and outputs a coherent natural language feedback text string. The network transport layer, through a bidirectional long connection channel or asynchronous callback interface, completely sends the natural language feedback text string and the structured query result message to the frontend framework environment of the terminal system.

[0041] After receiving the downlink data stream, the front-end rendering engine first instantiates a text bubble component node in the dialogue flow UI container of the right-hand intelligent inquiry area. The front-end rendering engine uses a virtual DOM difference comparison algorithm to calculate the range of page changes, attaches the natural language feedback text string to the text field attribute of the text bubble component node, and triggers local rendering instructions to complete the content output in the form of dialogue interaction.

[0042] Simultaneously, the front-end global state manager parses the intercepted structured query result message, separating the visual warning status code variable attached to the header and the indicator attribute numerical parameters in the load. The front-end global state manager traverses the data listening callback function set of each underlying chart instance in the intermediate data cockpit area in memory. When a chart instance that matches the mapping bound to the indicator attribute numerical parameters is found, the front-end global state manager dispatches a state change signal downwards. After capturing the state change signal, the underlying graphics rendering engine of the data cockpit area blocks the current regular polling redraw timer and extracts the hexadecimal color parameter configuration corresponding to the visual warning status code variable in the front-end theme configuration file. The underlying graphics rendering engine modifies the color attribute pointer of the chart model object in memory and overwrites the new indicator attribute numerical parameters into the data column of the chart model object. Then, it calls the underlying Canvas or SVG drawing API to perform clearing and pixel redrawing calculations, completing the dynamic linkage update of the corresponding data charts in the data cockpit area without refreshing the global page.

[0043] Simultaneously, the front-end routing and event dispatch bus extracts the instruction intent identifier contained in the structured query result message. When it is determined that the extracted instruction intent identifier falls within the preset business whitelist range requiring manual intervention (such as resource gap allocation events or unmet allocation rate events), the front-end routing and event dispatch bus dispatches a component mounting event to the main container tree of the left-hand centralized workbench area. The workflow engine module embedded in the centralized workbench area intercepts the component mounting event and dynamically instantiates a standard form shortcut entry component object in the heap memory. The workflow engine module serializes the intelligent allocation suggestion text package, gap status matrix data, or evaluation indicator abnormal parameters attached to the message and injects them into the initialization property field of the standard form shortcut entry component object. The front-end DOM operation executor appends and inserts the physical node corresponding to the standard form shortcut entry component object to the end of the to-do task list container node in the centralized workbench area, automatically generating a to-do business entry associated with the instruction intent. After the physical node is mounted, the workflow engine module automatically binds the backend business submission process API to the newly created pending business entry, so that the interface elements enter a ready state that can respond to external input device clicks for activation.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to 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 will 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. An experimental teaching natural language data query interaction method, characterized in that, Includes the following steps: S100, in response to the user's login request, obtain the user's role permissions, and dynamically render the three-zone interactive interface on the terminal display interface based on the role permissions; The three-zone interactive interface includes at least: an integrated workbench area for the flow of experimental teaching business processes, a data cockpit area for visual monitoring of experimental business indicators, and an intelligent query area for receiving unstructured data. S200, receiving natural language query commands about experimental teaching scenarios input by the user through the intelligent query area; S300, invoke a preset natural language processing model to perform semantic parsing on the natural language query instruction, extract the instruction intent and the business entity features corresponding to the instruction intent; the business entity features include at least one of the following: business action category, organizational hierarchy feature, time period feature, and experimental discipline feature; S400, based on the instruction intent and the characteristics of the business entity, performs cross-domain data retrieval in the underlying interconnected personnel database, asset database and academic affairs database, and calls the preset business rule engine to perform compliance comparison and indicator calculation on the retrieved basic data to generate structured query results; S500, in the intelligent inquiry area, outputs natural language feedback text containing the query results in the form of dialogue interaction, and dynamically updates the corresponding data charts in the data cockpit area according to the indicator attributes of the query results, and / or automatically generates a to-do business entry related to the instruction intent in the centralized workbench area.

2. The experimental teaching natural language data query and interaction method according to claim 1, characterized in that, Based on the aforementioned role permissions, the three-zone interactive interface is dynamically rendered on the terminal display screen, specifically including: The role permissions are parsed to determine the organizational structure data tag to which the user belongs. The organizational structure data tag includes the management decision-making level, the business execution level, and the terminal application level. Retrieve the interface configuration template uniquely bound to the organizational structure data tag from the preset UI configuration library; Based on the interface configuration template, the corresponding business micro-frontend components and data loading interfaces are dynamically assembled to generate the three-zone interactive interface adapted to the user.

3. The experimental teaching natural language data query and interaction method according to claim 2, characterized in that, The integrated workbench area is embedded with a workflow engine module; When rendering the centralized workbench area, component filtering is performed based on the user's role permissions, and non-business function nodes that do not belong to the current permission chain are blocked. Furthermore, the experimental teaching business processes activated under this permission are aggregated and presented in the form of form interaction components within the centralized workbench area to provide a one-way data flow operation environment. When rendering the physical layout of the three-zone interactive interface: The centralized workbench area and the intelligent inquiry area are respectively anchored to the left and right edges of the terminal display interface, and the data cockpit area is laid out as the underlying default view in the center area of ​​the terminal display interface. The intelligent inquiry area is configured as a floating dialog box. When the intelligent inquiry area is activated to receive natural language query commands, the core indicator charts in the data cockpit area adaptively adjust their scaling ratio.

4. The experimental teaching natural language data query and interaction method according to claim 3, characterized in that, The intelligent query area receives natural language query commands from users regarding experimental teaching scenarios, including: When the intelligent query area is in query receiving state, listen for the global focus event of the terminal display interface; The current form context data of the centralized workbench area and the current view filtering parameters of the data cockpit area are collected synchronously and packaged to generate a front-end scene snapshot; The front-end scene snapshot is used as an implicit query condition and bound and encapsulated with the natural language query command explicitly input by the user to form a multi-dimensional joint query request.

5. The experimental teaching natural language data query and interaction method according to claim 4, characterized in that, The intelligent inquiry area is equipped with a multimodal input interface, and the receipt of natural language query commands from users regarding experimental teaching scenarios specifically includes: The multimodal input interface receives a continuous voice stream from the user and calls an endpoint detection algorithm to identify valid voice segments in the continuous voice stream. Convert the effective speech segments into initial text data; Load a pre-set experimental teaching professional thesaurus, perform error correction and synonym mapping on the initial text data for specific domain entities, and output the natural language query command in a standardized text form; Before receiving the user's natural language query command about the experimental teaching scenario through the intelligent query area, the method further includes: Extract the user's role permissions, and combine the user's historical query logs with the current business time node to pre-generate several candidate query intent templates using a prediction algorithm; When the intelligent inquiry area is detected to be activated, the candidate inquiry intent template is rendered dynamically in front of the input box of the intelligent inquiry area in the form of a quick interaction component, in response to the user's click input or auxiliary statement completion.

6. The experimental teaching natural language data query and interaction method according to claim 5, characterized in that, The step of invoking a preset natural language processing model to perform semantic parsing on the natural language query command, extracting the command intent and the business entity features corresponding to the command intent, specifically includes: The natural language query command is converted into a word vector sequence, and the word vector sequence is input into a pre-trained sequence labeling network for local feature extraction to obtain initial entity text containing basic material vocabulary and basic course vocabulary. The pre-constructed knowledge graph of the experimental teaching domain is invoked, which contains the national curriculum standard experimental catalog and the Ministry of Education equipment classification code standard library. Calculate the semantic similarity between the initial entity text and the node entities in the experimental teaching domain knowledge graph, and perform entity alignment operations to map the non-standard initial entity text to the standardized business entity features; Based on the standardized business entity features and the word vector sequence, the intent probability distribution set in the multidimensional space is calculated using an intent classification network, and the category corresponding to the highest probability is determined as the instruction intent.

7. The experimental teaching natural language data query and interaction method according to claim 6, characterized in that, After extracting the instruction intent and the business entity features corresponding to the instruction intent, the method further includes a dynamic slot verification step: Based on the identified instruction intent, the corresponding intent-slot mapping template is invoked to determine the set of required business entity slots for executing the instruction intent; Compare the extracted business entity features with the set of required business entity slots. If a missing required business entity is detected, the current semantic parsing process is suspended and a front-end scene snapshot is cached. In the intelligent query area, a follow-up prompt is generated for the missing required business entity. When a supplementary instruction based on the follow-up prompt is received, the entity features in the supplementary instruction are extracted and filled into the missing required business entity slot, and the semantic parsing process is resumed. The business action categories in the business entity characteristics include at least resource operation and maintenance actions, performance evaluation and calculation actions, and security investigation actions; When semantic parsing determines that the business action category belongs to the evaluation calculation action, fine-grained evaluation indicator labels are further parsed from the natural language query instruction based on the associated organizational hierarchy features and experimental discipline features. The evaluation indicator label is configured as a soft switch to activate the corresponding indicator calculation engine at the underlying level during subsequent cross-domain data retrieval; the indicator calculation engine is pre-configured with an algorithm for calculating the full opening rate of experiments, an algorithm for calculating the asset value per student, and an algorithm for calculating the equilibrium index.

8. The experimental teaching natural language data query and interaction method according to claim 7, characterized in that, The cross-domain data retrieval in the underlying interconnected personnel database, asset database, and academic affairs database specifically includes: Obtain the organizational structure data tags and corresponding row-level data access permission boundaries registered by the user in the data exchange bus; The instruction intent and the business entity characteristics are compiled into distributed federated query statements, and concurrently routed to the personnel database, the asset database and the academic affairs database through the data exchange bus; Under the constraints of the row-level data access permission boundary, each database engine performs asynchronous query operations and performs primary key mapping, cleaning and fusion of the returned multi-source heterogeneous subset datasets to generate an intermediate basic dataset to support the current query session; The process of calling a preset business rule engine to perform compliance comparisons and indicator calculations on the retrieved basic data includes concurrent computation logic for the experiment preparation scenario: When the instruction intent is identified as an experiment reservation or asset availability query, the standard instrument BOM list and the total number of students to be taught for the target experimental course are extracted from the intermediate basic dataset. Simultaneously extract real-time available inventory data and in-transit borrowing data from the asset database that match the standard instrument BOM list; The resource scheduling algorithm engine is invoked to calculate the difference between the target total number of students and the comprehensive available inventory, generating a resource gap status matrix, and encapsulating intelligent allocation suggestions for the resource gap status matrix into the structured query results.

9. The experimental teaching natural language data query and interaction method according to claim 8, characterized in that, The process of calling a preset business rule engine to perform compliance comparisons and indicator calculations on the retrieved basic data also includes indicator derivation logic for comprehensive regulatory scenarios: When the instruction intent is identified as a regional compliance assessment, the historical experiment execution logs, equipment asset lists, and personnel staffing data contained in the intermediate basic dataset are input into the built-in regulatory comparison model; Based on the national curriculum standard baseline pre-set in the business rules engine, the experiment completion rate parameter and per-student asset balance index are calculated through the regulatory comparison model for the characteristics of the business entity. The experimental completion rate parameter and the per capita asset balance index are quantitatively graded and determined based on a preset multi-level threshold network, and a visual warning status code corresponding to the risk level is attached to the structured query results.

10. The experimental teaching natural language data query and interaction method according to claim 9, characterized in that, Based on the indicator attributes of the query results, the corresponding data charts in the data dashboard area are dynamically updated, specifically including: Extract the data dimension labels and accompanying visual warning status codes from the structured query results; Within the DOM tree structure of the data cockpit area, traverse and match the target visualization chart components bound to the data dimension labels; Based on the visual warning status code, an asynchronous partial refresh instruction is sent to the target visualization chart component to drive the target visualization chart component to perform independent data frame redrawing; and for data nodes that reach the preset alarm threshold, coordinate system centering and view highlighting operations are performed. The automatic generation of a pending task entry associated with the instruction intent in the centralized workbench area specifically includes: When the instruction intent is determined by preset rules to be an action instruction with closed-loop operation attributes, the set of key parameters required for business flow is extracted from the structured query results. The workflow engine module embedded in the centralized workbench area is invoked to silently instantiate and assign values ​​to the target business form template using the set of key parameters, generating an instance form with pre-filled data. The instance form is encapsulated as a quick to-do interactive card and pushed to the top-level view stack of the centralized workbench area so that the user can confirm or supplement the data flow with one click; The intelligent query area outputs natural language feedback text containing the query results in the form of a dialogue interaction, including: The natural language generation engine is invoked to match the corresponding response strategy template in the pre-set corpus template library based on the data structure characteristics of the query results. The numerical indicators and entity names in the structured query results are used as dynamic variables and injected into the corresponding placeholders in the response strategy template to generate feedback text that conforms to natural language syntax. The natural language feedback text output in the current round, the associated structured query results, and the corresponding instruction intent are packaged into an interaction context package and written into the global state cache to maintain contextual memory constraints for subsequent continuous query instructions.

Images