Business rule constraint-oriented structured data entry method and system

Abstract

The invention discloses a structured data entry method and system oriented to business rule constraints. The method comprises the following steps: firstly, receiving multi-modal semantic input and retrieving a business object identifier, and loading a historical state matrix containing a field-level state locking identifier according to the business object identifier; calling a semantic extraction module to analyze and obtain a business intention and a target field candidate value; through a multi-source data conflict resolution module, under the condition that a target field is locked, value coverage is allowed to be executed and a timestamp is reset only when an error correction semantic signal containing an error correction feature is detected, otherwise, an original value is maintained. Meanwhile, the system judges the integrity based on a preset business constraint knowledge base, and generates structured business data or a questioning instruction. And recording a decision attribution evidence chain containing a rule track and a fault feature identifier in the whole process. Through a field-level state locking mechanism and an automatic attribution evidence storage mechanism, controlled management of field updating behaviors in the multi-round structured data entry process is achieved, tracing analysis of the data entry process and the abnormal state is supported, and therefore the method is suitable for business scenes with high requirements for consistency and stability.

Description

Technical Field

[0001] This invention relates to the fields of computer data processing and natural language processing, and in particular to a multi-round interactive structured data entry method and system oriented towards business rule constraints. Background Technology

[0002] With the rapid development of artificial intelligence (AI) and natural language processing (NLP) technologies, automated data entry solutions that transform unstructured multimodal semantic inputs (such as speech and text) into structured business data have been widely used in finance, industry, healthcare and other fields.

[0003] However, in real-world applications (especially complex multi-turn interaction scenarios), existing structured data entry technologies still suffer from the following technical shortcomings:

[0004] 1. Information Overwriting Issues in Multi-Turn Interactions: During multi-turn semantic extraction, the AI ​​model's interpretation of contextual intent is uncertain. When users input new semantic information in subsequent interactions, existing systems often struggle to accurately identify whether this information is "supplementation of a new field" or "misinterpretation of an already entered field." Due to the lack of a state management mechanism based on field-level locking identifiers, newly generated extraction results are highly susceptible to incorrectly overwriting previously confirmed correct data, making it difficult to guarantee the consistency of structured data.

[0005] 2. Lack of a clear technical judgment mechanism for error correction logic: Existing systems typically employ simple "time priority" or "probability score priority" strategies when processing data updates. This approach fails to effectively distinguish between "normal business updates" and "error correction actions." Without the linkage between error correction semantic recognition and field-level locking indicators, the system cannot guarantee that core fields will only be updated upon receiving a clear error correction signal, thus reducing the rigor of structured data entry. Traditional database concurrency control technologies (such as optimistic locking or Multi-Version Concurrency Control (MVCC)) can resolve data write conflicts, but their core logic is based on "time sequence" or "transaction isolation," failing to understand business semantics. In human-computer interaction scenarios, user "correction" actions do not always occur at the "last moment." Therefore, directly applying existing database locking mechanisms cannot solve the intent conflict problem in AI semantic understanding; a dedicated technical solution integrating semantic understanding (error correction recognition) and state control (locking mechanism) is urgently needed.

[0006] 3. Lack of transparency in the decision-making process and difficulty in attributing failures: Existing data entry systems are mostly "black box" processes, lacking a record of the AI ​​extraction process, rule triggering logic, and field state transitions. When structured results deviate or external semantic interfaces (such as LLM interfaces) respond abnormally, developers and operations personnel find it difficult to trace the basis of decisions and cannot accurately attribute different types of failures, such as "communication timeout" and "logic matching anomalies," failing to meet the auditing and stability requirements of high compliance scenarios.

[0007] Therefore, there is an urgent need for a technical solution that can ensure the accuracy and stability of data entry through field-level status control, accurate identification of error correction intentions, and end-to-end decision-making evidence preservation. Especially against the backdrop of my country's vigorous promotion of the deep integration of artificial intelligence and the real economy, and the construction of an independent and controllable digital economy foundation, developing a structured data entry system that can effectively solve the uncertainty of AI semantic extraction and possess complete decision-making attribution capabilities is of significant practical importance for achieving digital transformation in key business areas, particularly in fields such as finance, government affairs, and high-end manufacturing, which have strict requirements for end-to-end data security and independent intellectual property rights. Summary of the Invention

[0008] (a) Technical issues

[0009] The technical problem this invention aims to solve is that, in existing unstructured data entry processes, semantic extraction results based on artificial intelligence (AI) often exhibit uncertainty. Especially in multi-round interaction scenarios, due to the lack of effective state management and conflict resolution mechanisms, newly generated extraction results can easily overwrite previously confirmed correct data. Furthermore, existing data entry systems lack transparent recording of the decision-making process and effective attribution of external interface anomalies, resulting in an untraceable data entry process and poor system stability in industrial application environments.

[0010] (II) Technical Solution

[0011] To address the aforementioned technical problems, this invention provides a structured data entry method oriented towards business rule constraints, specifically including the following steps:

[0012] 1. State machine-driven context management: Retrieve session drafts based on business object identifiers, load session state data (e.g., state matrix) containing field-level locking identifiers, and provide a state baseline for multi-round data entry;

[0013] 2. Conflict resolution based on error correction semantics: When the target field is locked, the error correction semantic recognition unit determines the error correction signal. Only when a clear error correction signal is detected is it allowed to overwrite and reset the timestamp; otherwise, the original value is maintained, which solves the conflict problem caused by the uncertainty of AI extraction. The error correction semantic recognition unit can be implemented as a sub-functional unit of the semantic extraction module or deployed as an independent semantic analysis module.

[0014] 3. Business rule-driven integrity determination: The system uses a pre-set business constraint knowledge base to perform derivative calculations and determine integrity. For missing fields, it automatically generates structured follow-up questions, realizing a closed-loop guidance for the data entry process. The business constraint knowledge base is coupled with the multi-round business logic processing engine and is called by the multi-round business logic processing engine when it detects a field status change event, so as to realize the closed-loop processing of business logic.

[0015] 4. End-to-end decision audit and attribution: Real-time generation of decision attribution evidence chains including rule trajectories, AI-extracted original text, and fault feature identifiers, which are stored in the audit and evidence storage unit, realizing end-to-end evidence storage and anomaly tracing from semantic extraction to business data entry.

[0016] The present invention also provides a structured data entry system oriented towards business rule constraints for implementing the above method, characterized by including: an input layer, a session draft management unit, a multi-round business logic processing engine, a multi-source data conflict resolution module, an integrity judgment unit, a decision trajectory audit and evidence storage unit, and an anomaly monitoring and automatic attribution module.

[0017] It should be noted that the rule identifiers, status identifiers, and error correction semantic signals described in this invention are only used to describe the decision-making path and processing basis of the system under the constraints of business rules, and their specific naming, quantity, and representation do not constitute a limitation on the technical solution of this invention.

[0018] (III) Beneficial Effects

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. Enhance the certainty and compliance of business decisions: By introducing field-level locking identifiers and error correction semantic recognition mechanisms, the problem of information overwriting in complex multi-round interactions is solved, ensuring the logical consistency of structured data based on preset business rules in the production environment.

[0021] 2. Achieve transparent auditing and explainability of the decision-making chain: Relying on the decision attribution evidence chain mechanism, the semantic extraction process based on artificial intelligence is transformed into a traceable rule-based decision trajectory, enabling the system to achieve transparent auditing of the decision-making process without relying on the internal state of the model, thus meeting the technical auditing requirements of high compliance scenarios.

[0022] 3. Improve data governance efficiency and engineering processing capabilities: Based on business constraints, the automatic triggering of derivative calculations and structured inquiry mechanisms reduces redundant inputs in multi-round interaction processes, improves the overall processing efficiency of converting unstructured semantic data into structured business data, and provides an efficient underlying data access solution for enterprise digital transformation.

[0023] 4. Possesses stable anomaly attribution and business continuity assurance capabilities: The anomaly monitoring mechanism can distinguish and identify communication anomalies and business logic anomalies, and update the anomaly occurrence stage through a state machine, providing technical support for the continuous operation of the system in a large-scale, high-concurrency environment.

[0024] 5. Supports full-link traceability and playback: Through the structured storage of full snapshot data, the system can achieve traceable and verifiable playback of the business input decision path based on decision attribution records without relying on external semantic reasoning, thereby improving the auditability and certainty of system decisions.

[0025] Those skilled in the art should understand that the functional modules described in this invention can be implemented in a centralized or distributed manner, and their functional logic can also be accomplished through different software structures, hardware architectures, or combinations of software and hardware. Any equivalent substitutions or improvements that those skilled in the art can conceive of under the technical guidance of this invention are within the scope of protection of this invention. Attached Figure Description

[0026] Figure 1 is a schematic diagram of the logical architecture of the structured data entry system oriented towards business rule constraints involved in the embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the overall process of a structured data entry method oriented towards business rule constraints according to an embodiment of the present invention;

[0028] Figure 3 is a schematic diagram of the method for resolving field-level locking identifiers and multi-source data conflicts in an embodiment of the present invention;

[0029] Figure 4 is a schematic diagram of the external interface anomaly monitoring and automatic attribution and evidence storage process involved in the embodiments of the present invention. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0031] It should be noted in advance that, in this embodiment of the invention, the 'session state management unit' may also be referred to as the 'session draft management unit' or the 'state machine management module', and its core function is to maintain session state data containing field-level locking identifiers; the 'audit and evidence storage unit' can be implemented based on a database, blockchain or other persistent storage media, and is used to store decision attribution records.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0033] [Example 1: Main Process for Structured Entry of Overall Business Data]

[0034] Referring to Figure 2, this embodiment provides a structured data entry method oriented towards business rule constraints, the specific process of which includes:

[0035] Step S101: Receive multimodal semantic input data. The system acquires input sources including but not limited to text obtained from speech conversion and text directly input by the user. In embodiments of the present invention, the multimodal semantic input can also be referred to as multi-source semantic input data, used to represent semantic information from different input channels or different data sources. In this specification, the two terms can be used interchangeably without causing ambiguity.

[0036] Step S102: Retrieve the status information of the target business object identifier. Confirm the context of the current interaction through the business object identifier (such as an asset identifier).

[0037] Step S103: Determine the loading status of the session draft. If an incomplete session draft for the business object is detected, historical session status data (e.g., a status matrix) is loaded from the session draft management unit; otherwise, null session status data (e.g., a status matrix) is initialized.

[0038] Step S104: Semantic parsing. The semantic extraction module is invoked to parse the user's input business intent and entity data.

[0039] Step S105: Invoke the multi-round business logic processing engine to perform processing. This step includes: performing multi-dimensional data dimension normalization processing to unify non-standard units, date semantics, etc. in the input data into a standard data format; triggering derivative calculation logic based on a preset business constraint knowledge base to complete the associated field data related to the current business intent. The derivative calculation logic is a deterministic and repeatable rule derivation process based on the constraint relationship between confirmed fields. It is automatically executed when a field state change event is triggered, and at least one associated field result obtained from the derivation is written into the current session draft state data; invoking the multi-source data conflict resolution module to perform field-level locking and coverage determination; and generating the decision attribution evidence chain for this business processing process.

[0040] Step S106: Integrity Condition Determination. Determine whether the current structured data meets the preset business constraint integrity conditions. The integrity conditions include at least whether the required fields corresponding to the business intent are complete, and whether the field values ​​conform to the preset business constraint rules.

[0041] Step S107: Output Processing. If the integrity condition is met, generate structured business data to be confirmed; if not, generate structured follow-up instructions based on the missing slots.

[0042] Step S108: Closed-loop evidence storage. Write full snapshot data to the decision trajectory audit evidence storage unit and update the session draft status management unit to achieve technical traceability of the entire data entry process.

[0043] [Example 2: Field-level locking flags and multi-source data conflict resolution logic]

[0044] In this embodiment, "field-level lock flag" refers to a logical write protection mechanism. Specific implementations of this mechanism include, but are not limited to: setting a Boolean lock flag, setting a write protection time window, setting a confidence threshold, or comparing data version numbers. Any control logic that dynamically adjusts field write permissions based on semantic judgment results should be considered as the lock state management described in this invention.

[0045] Specifically, the locking mechanism described in this invention differs fundamentally from traditional database locks in that it triggers unlocking not based on the physical timestamp or transaction isolation identifier of the data, but on the logical feature of "error correction intent" parsed by the error correction semantic recognition unit based on multimodal input. In this invention, timestamp resetting is merely a means of state synchronization after conflict resolution, rather than a precondition for determining whether write access is allowed. This dynamic locking mechanism based on intent recognition solves the problem of information mis-overwriting caused by semantic parsing uncertainty in multi-turn dialogues in artificial intelligence.

[0046] Referring to Figure 3, the core mechanisms for handling multi-round interactive data conflicts include:

[0047] Step S201: Obtain the target field to be entered and its candidate values.

[0048] Step S202: Retrieve historical status. Determine if the target field exists in the historical session draft status data. If it does not exist, perform new field creation and writing, and synchronously activate the lock flag, recording the rule flag R-LOCK-NEW.

[0049] Step S203: Lock Status Determination. If the target field already exists, further determine whether the field is in a locked protection state. If not locked, execute the field value writing and activate the lock flag, recording the rule flag R-LOCK-INIT.

[0050] Step S204: Error Correction Semantic Recognition. If the field is locked, the system calls the error correction semantic recognition unit to identify the error correction features in the input semantics.

[0051] Step S205: Conflict resolution decision. If a clear error correction semantic signal is detected, execute branch A (error correction update): allow field value overwrite operation, reset field-level lock timestamp, and record rule identifier R-LOCK-UPDATE; if no clear error correction semantic signal is detected, execute branch B (lock maintenance): activate lock protection and reject overwrite, retain the original value, and record rule identifier R-LOCK-KEEP.

[0052] Step S206: Synchronize and update. Update the session draft state management unit in real time to ensure that the persistent state and logical decisions are synchronized.

[0053] It should be noted that the rule identifiers such as R-LOCK-NEW, R-LOCK-INIT, R-LOCK-UPDATE, and R-LOCK-KEEP mentioned in this embodiment are only used to distinguish logical identifiers of different business processing paths in the field status management process, and their specific naming forms do not constitute a limitation on the technical solution of this invention.

[0054] The core technical solution of this invention lies in: control logic for dynamically locking, unlocking, or maintaining field-level write permissions based on error correction semantic recognition results, and the resulting multi-source data conflict resolution mechanism. The triggering conditions and processing results of the relevant rules are fully illustrated in the field status locking and conflict resolution decision process shown in Figure 3. Those skilled in the art can make equivalent substitutions for the names, quantities, or representations of the above rule identifiers without departing from the technical concept of this invention.

[0055] [Example 3: External Interface Anomaly Monitoring and Automatic Cause Attribution and Evidence Preservation]

[0056] Referring to Figure 4, the method for ensuring system stability includes: Step S301: Trigger an external semantic extraction interface call request; Step S302: Monitor whether the interface response is within the normal communication threshold; Step S303: If the response is successful, parse the intent and entity set, match the semantic verification rules, execute the logical judgment, and construct an evidence chain snapshot.

[0057] Step S304: If the response is abnormal, the system automatically executes the attribution logic: determine whether the abnormal feature matches the communication timeout threshold. If yes, mark the fault feature flag as communication timeout (AI_TIMEOUT); if no, mark the fault feature flag as internal processing abnormality (PROCESS_ERROR). The above abnormality type flags are only exemplary internal flags, and their specific names do not constitute a limitation.

[0058] Step S305: Attribution and Evidence Storage. Associate the fault characteristic identifier with the business object and store it in the audit evidence storage unit. Step S306: Update the system state machine to support subsequent review and retrospection.

[0059] The system architecture and methodological logic described in this invention possess a high degree of decoupling. Specifically, the decision trajectory audit and evidence storage mechanism and the external interface anomaly automatic attribution mechanism can serve as a general "business decision traceability and operational stability assurance solution for artificial intelligence systems."

[0060] [Example 4: Application of Data Replay and Logic Reconstruction]

[0061] In an extended embodiment of the present invention, the system provides a data replay function based on the decision trajectory audit and evidence storage unit. The technical logic of this function is as follows: the system extracts historical decision attribution records according to specific business object identifiers, loads the corresponding multi-round multimodal semantic input original text and its parsing results sequentially according to the time order marked in the decision attribution records, and uses them as driving signals. Combined with the rule identifier triggered at that time (such as R-LOCK-UPDATE) and the corresponding full snapshot data, the field state change process is replayed step by step without changing the original rule execution order, thereby completely reproducing the entire process of data entry decision-making in the logic simulation environment.

[0062] This data playback process can be used for offline regression testing of business logic, as well as for evidence auditing and analysis of complex error correction behaviors, ensuring that the generation process of each piece of structured data has traceable and verifiable decision-making basis.

[0063] [Example 5: Industrial Application Example and Verification of Beneficial Effects]

[0064] The system described in this invention has been applied in the intelligent electronic ledger entry scenario of livestock farms. By receiving multimodal data from sources such as voice, text, and image OCR, it achieves structured entry of livestock farm operating data and generates a comprehensive financial overview. In a series of actual multi-round voice interaction tests of a set of embodiments, the field miscovery rate caused by the AI ​​model's intent understanding deviation was reduced by more than 50% compared to the traditional "latest priority" strategy; at the same time, the system's recognition accuracy for user-initiated correction behavior reached over 95%. The above data are only exemplary test results, used to illustrate the technical effects that this invention may achieve, and do not constitute a limitation on the scope of protection of this invention.

[0065] Furthermore, because the decision attribution evidence chain fully records the original text of semantic extraction and rule judgment identifiers for each round, the time cost of business audit backtracking is reduced by approximately 70%. Based on the architecture described in this invention, the system is also applicable to business scenarios with high requirements for data accuracy and stability, such as financial credit interview data entry and industrial equipment inspection, demonstrating good industry universality. The system described in this embodiment not only performs excellently in the smart ledger scenario of livestock farms, but its core algorithm architecture also has the technical foundation for rapid migration to related industries in various high-tech industrial clusters (such as smart government affairs, financial compliance review, and electronic medical records). The large-scale deployment of this system will effectively support the data security needs of enterprises in the region during the process of 'digital transformation' and 'full business cloudification'. In addition, the decision playback and attribution evidence storage mechanism described in this invention can serve as a core technical component for enterprises to build a 'trustworthy artificial intelligence' evaluation system, with significant industrial synergy effects and socio-economic benefits.

Claims

1. A structured data entry method, characterized in that, include: During multi-round interaction, session state data corresponding to business objects is maintained. The session state data includes multiple business fields and their corresponding field-level locking identifiers. The field-level locking identifiers are used to distinguish between the first confirmation write behavior of a field and the update behavior based on error correction intent during multi-round interaction. When the target business field is detected to be written to for the first time, the corresponding field value is stored in the session state data, and the field-level locking flag corresponding to the target field is activated to restrict subsequent updates to the field. When a new semantic input is received and the candidate value of the target field is parsed, if the field-level locking flag corresponding to the target field is in an active state, the semantic input is used to determine the error correction intent. Only when the determination result indicates that the semantic input contains an error correction intent for the target field, the field-level locking flag corresponding to the target field is released, and field value overwrite operation is allowed for that field.

2. The method as described in claim 1, characterized in that, The field-level status management includes: Based on the business object identifier, the corresponding historical session draft status data is loaded, wherein the session status data includes at least the session draft status data; When the target field obtained from parsing does not exist in the current session draft state data, a field creation and writing operation is performed, and the field-level locking flag of the field is activated at the same time as the field writing is completed.

3. The method as described in claim 1, characterized in that, The determination of the error correction intention includes: The error-correcting semantic recognition unit is invoked to perform semantic feature parsing on the semantic input; Based on a preset error correction semantic feature template or semantic feature matching rule, determine whether error correction semantic features exist in the semantic input; When the judgment result meets the error correction judgment condition, an error correction intent identifier for the corresponding target field is generated; The error correction semantic features include at least one of the following: error correction keyword features, semantic negation structure features, or correction instruction features.

4. The method as described in claim 1, characterized in that, Also includes: When field value overwrite operations are allowed, the field-level lock timestamp corresponding to the target field is reset synchronously, so that subsequent conflict determination for the field is based on the most recent valid modification time of the field, thereby avoiding interference from historical semantic input to newly confirmed fields.

5. The method as described in claim 1, characterized in that, Also includes: Based on a pre-defined business constraint knowledge base, when a field status change event is detected, derivative calculation logic is triggered to complete the associated fields related to the current business intent. It also performs an integrity check on the current structured data to determine whether the preset field completeness conditions are met; When the integrity condition is not met, a structured follow-up instruction corresponding to the missing field is generated.

6. The method as described in claim 1 or 5, characterized in that, It also includes decision auditing and anomaly attribution steps: During the structured data entry process, the triggered rule identifiers, field status changes, and field processing results are written into the audit and evidence storage unit to form a decision attribution record that includes rule trajectories, multi-round multimodal semantic input text, field status changes, and field processing results. It also monitors the response status of the external semantic extraction interface. When an anomaly is detected, it assigns a cause to the anomaly type based on preset anomaly characteristics. The anomaly type includes at least communication timeout anomalies and internal processing anomalies.

7. A structured data entry system, characterized in that, include: The session state management unit is used to maintain session state data during multi-round interactions. The session state data includes multiple business fields and their corresponding field-level locking identifiers. A multi-round business logic processing engine is used to write the corresponding field value to the session state management unit and activate the field-level locking flag corresponding to the target field when the target business field is detected to be written for the first time. The multi-source data conflict resolution module is used to determine the error correction intent of the received semantic input when the field-level locking flag corresponding to the target field is in an active state. The multi-source data conflict resolution module is further configured to: release the locking flag of the corresponding field and allow the field value overwrite operation only when the determination result indicates that the semantic input contains an error correction intent for the target field.

8. The system as described in claim 7, characterized in that: When the target field does not exist in the current session draft state, the multi-source data conflict resolution module performs a field creation and writing operation and simultaneously activates the field-level state lock flag of the corresponding field.

9. The system as described in claim 7, characterized in that: The multi-source data conflict resolution module is configured to synchronously reset the field-level lock timestamp corresponding to the target field when field value overwrite is allowed.

10. The system as described in claim 7, characterized in that: The multi-source data conflict resolution module includes an error correction semantic recognition unit, which is used to determine whether error correction features exist in the user input semantics based on a preset error correction semantic feature template or semantic feature matching rule.

11. The system as described in claim 7, characterized in that: The multi-round business logic processing engine is configured to trigger derivative calculation logic based on a preset business constraint knowledge base when a field state change event is detected, so as to complete the associated fields in the structured data.

12. The system as described in claim 7 or 10, characterized in that: It also includes an anomaly monitoring and automatic attribution module, which monitors the response status of external semantic extraction interfaces and attributes the anomaly type based on preset anomaly features when an anomaly is detected.

13. The method as described in claim 6, characterized in that, It also includes a data playback step: based on the full snapshot data and corresponding rule trajectory in the decision attribution record, the reconstruction of state data is driven in time series order to reconstruct the input decision logic evolution path of a specific business object in a multi-round interaction process.

14. The system as described in claim 12, characterized in that, Also includes: The data playback module is configured to drive the reconstruction of state data in chronological order based on the full snapshot data and corresponding rule trajectories in the decision attribution record, so as to reconstruct the input decision logic evolution path of a specific business object in a multi-round interaction process.

Images