A large model inference data processing method and system for structured data

Abstract

The application discloses a large model inference data processing method and system for structured data, and relates to the technical field of data processing, and comprises the following steps: receiving a natural language query request sent by a user based on a financial service; analyzing the semantic content in the natural language query request; generating a first SQL query statement based on the semantic content; obtaining query feedback information fed back by a database based on the first SQL query statement; identifying a limited data element in the semantic content based on the query feedback information; performing semantic content evaluation on data elements in the first SQL query statement based on the semantic content and the limited data element to obtain alternative query content; and feeding back the query result obtained from the database based on the second SQL query statement to the user. The application enables analysts to not need to have professional database operation skills, greatly improving the efficiency and decision-making speed of data analysis.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method and system for large-scale model inference data processing for structured data. Background Technology

[0002] In the field of modern financial data analytics, institutions have widely deployed intelligent data processing systems to efficiently process massive amounts of transaction records, customer information, and market data, and to extract valuable insights. The core of these systems is typically a large model that understands the queries analysts submit in natural language and automatically translates them into standard SQL queries that the database can understand and execute. With increasingly stringent regulations on data security and personal privacy protection, financial institutions have become more granular in their access control to internal databases. Access control is no longer simply about restricting a user's access to a specific data table; it extends to every specific field within the table. This means that even the same analyst viewing the same customer information table might be allowed to view "customer ID" and "transaction amount," but prohibited from viewing "customer contact information." Even more challenging is that these permissions are not static; they are adjusted in real-time based on various factors such as the analyst's current project involvement, the data's confidentiality level, and even the time the query occurred.

[0003] When a large database model translates an analyst's natural language commands into SQL queries, it primarily relies on the overall database structure. However, it doesn't know which fields the analyst has specific access permissions to. Therefore, when an analyst submits a seemingly normal query, the large model might generate an SQL query containing fields the analyst lacks access rights to. When this SQL query is sent to the database for execution, the database's security access control layer will immediately reject the query request.

[0004] What's even more perplexing is that, for security reasons, databases typically don't explicitly tell the system which specific field lacks sufficient permissions when rejecting SQL query requests. Instead, they return vague messages like "Access denied" or "Query error." When a large model system receives this ambiguous failure feedback, it can't accurately determine the true reason for the SQL query failure and therefore doesn't know how to modify the rejected query. The system's initial approach is to try simplifying the SQL query, such as speculatively removing some fields it deems less important, and then trying again. However, this trial-and-error approach is extremely inefficient. It may still fail after multiple attempts, or, in order to make the SQL query pass, it may have to delete the key fields that the user truly needs, resulting in incomplete data and losing its original analytical value. This makes the intelligent system, originally intended to improve efficiency, unreliable in the new security environment. Analysts are forced to revert to the primitive method of manually writing SQL queries, significantly reducing the system's usability. Summary of the Invention

[0005] This application discloses a method and system for processing large model inference data for structured data. It aims to solve the problem that existing large model systems cannot accurately determine the cause of SQL query failure when faced with fine-grained database access control. The method of this invention improves the reliability of large model inference data processing for structured data, enabling analysts to efficiently obtain the required data in a strictly secure environment without having professional database operation skills, which greatly improves the efficiency of data analysis and decision-making speed.

[0006] In a first aspect, this application discloses a method for processing large-scale model inference data for structured data, comprising the following steps: Receive natural language query requests from users based on financial transactions; Parse the semantic content in natural language query requests; Generate the first SQL query statement based on semantic content; Execute the first SQL query statement and obtain the query feedback information from the database based on the first SQL query statement; Restricted data elements in semantic content are identified based on query feedback information; Based on semantic content and restricted data elements, semantic content evaluation is performed on the data elements in the first SQL query statement to obtain alternative query content, which includes alternative data elements and / or query granularity. A second SQL query is generated based on the alternative query content, and the query results obtained from the database based on the second SQL query are returned to the user.

[0007] Furthermore, parsing the semantic content in a natural language query request includes: Identify query keywords and query entities in natural language query requests; The query should be based on the query keywords and query entities to analyze the actual business needs. Based on preset business strategies, the corresponding business objectives are matched with real business needs. Map each piece of data in the natural language query request to the specific field in the database corresponding to the business objective.

[0008] Furthermore, mapping each piece of data in the natural language query request to the specific field in the database corresponding to the business objective includes: Match each piece of data with the database table structure of the business objectives to obtain the database elements for each piece of data; Analyze the business value priority of each piece of data within the business objectives; Adjust the order of database elements in the database table structure based on business value priority; The database table structure based on the arrangement order maps each piece of data to a specific field corresponding to the business objective.

[0009] Furthermore, based on the query feedback information, the restricted data elements identified in the semantic content include: Determine if there is an abnormal query status in the query feedback information. An abnormal query status includes access being denied or query error information. If an abnormal query status is detected, a field permission detection process is generated based on each field in the first SQL query statement; Each field permission detection process is executed sequentially, and the field query results returned by the database are obtained based on each field permission detection process; Identify the failure status in the field query results, and determine that the field corresponding to the field permission detection process is a restricted data element based on the failure status.

[0010] Furthermore, the field permission detection process based on each field in the first SQL query statement includes: Identify the data columns, filter conditions, and aggregation operations for each field in the first SQL query statement; A field permission detection process is generated based on data columns, filtering conditions, and aggregation operations.

[0011] As a technological improvement, alternative query content is obtained by performing semantic content evaluation on the data elements in the first SQL query statement based on semantic content and restricted data elements, including: Based on semantic content and restricted data elements, the corresponding substitution strategy is matched from the set of financial business data substitution rules; Based on the alternative strategy, a compliance pre-check is performed on the potential alternative data elements to obtain the potential alternative data elements that meet the compliance requirements; Based on the alternative strategy and the current working context, potential alternative data elements that meet compliance requirements are prioritized to obtain the priority ranking results; Based on the priority ranking results, potential alternative data elements with a high degree of semantic matching are selected as alternative query content.

[0012] Furthermore, based on the alternative strategy and the current working context, potential alternative data elements that meet compliance requirements are prioritized, resulting in the following priority ranking: Listen for external dynamic events related to the current work context; Assess the relevance of external dynamic events to potential alternative data elements that meet compliance requirements; The adjusted priority weights are obtained by adjusting the business value and compliance importance of potential alternative data elements that meet compliance requirements based on external dynamic events. Based on priority weights and preset weights for alternative strategies, potential alternative data elements that meet compliance requirements are weighted and scored to obtain priority ranking results.

[0013] More specifically, assessing the relevance of external dynamic events to potential alternative data elements for meeting compliance requirements includes: Identify the types of external dynamic events and the areas of impact of these events; Based on the type and impact area of ​​external dynamic events, search the financial business logic transmission path library for transmission paths related to potential alternative data elements that meet compliance requirements; Analyze the transmission hierarchy and latency defined in the transmission paths related to potential alternative data elements that meet compliance requirements; Adjust the effective window of the impact of external dynamic events according to the delay time defined in the transmission path; Assess the degree of indirectness of the impact of external dynamic events based on the transmission hierarchy defined in the transmission path; Assess the correlation between external dynamic events and potential alternative data elements that meet compliance requirements based on the effective window and degree of indirectness.

[0014] Furthermore, the query results obtained from the database based on the second SQL query statement will be fed back to the user, including: Receive the query results returned by the database based on the second SQL query statement; Configure the interpretation content corresponding to the second SQL query statement based on the query results; The query results and explanations will be provided to the user.

[0015] Secondly, this application also discloses a large-scale model inference data processing system for structured data, the system comprising: The receiving module is used to receive natural language query requests sent by users based on financial transactions; The parsing module is used to parse the semantic content in natural language query requests; The first generation module is used to generate the first SQL query statement based on semantic content; The first execution module is used to execute the first SQL query statement and obtain the query feedback information returned by the database based on the first SQL query statement; The identification module is used to identify restricted data elements in semantic content based on query feedback information; The second generation module is used to perform semantic content evaluation on the data elements in the first SQL query statement based on semantic content and restricted data elements to obtain alternative query content, which includes alternative data elements and / or query granularity. The second execution module is used to generate a second SQL query statement based on the alternative query content, and to return the query results obtained from the database based on the second SQL query statement to the user.

[0016] This application discloses a data processing method for large-scale model inference of structured data. It receives natural language query requests from users based on financial transactions, parses their semantic content, and generates a first SQL query statement. After executing the SQL query statement and obtaining query feedback information from the database, this application can identify restricted data elements in the semantic content based on the query feedback information. Furthermore, this application performs semantic content evaluation on the data elements in the first SQL query statement based on the semantic content and restricted data elements to obtain alternative query content. This alternative query content may include alternative data elements and / or query granularity. Finally, this application generates a second SQL query statement based on the alternative query content and returns the query results obtained from the database based on the second SQL query statement to the user.

[0017] Through the above technical solution, this application effectively solves the problem in existing large-model systems where, when faced with granular database access control, the inability to accurately determine the cause of SQL query failure due to fuzzy error information returned by the database. By introducing a restricted data element identification mechanism, this application can accurately locate the specific field causing the query failure, avoiding the inefficiency of traditional trial-and-error methods. Simultaneously, through semantic content evaluation and the generation of alternative query content, this application can intelligently adjust query strategies, providing effective query results while ensuring data compliance and security, overcoming the limitations of existing technologies where insufficient permissions lead to incomplete data or query failures. Therefore, this application significantly improves the reliability, efficiency, and user experience of large-model inference data processing for structured data, enabling analysts to efficiently obtain the required data in a strictly secure environment without requiring specialized database operation skills, greatly improving the efficiency of data analysis and the speed of decision-making. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the large model inference data processing method for structured data in an embodiment of the present invention. Figure 2 This is a schematic diagram of the method for identifying restricted data elements in semantic content based on query feedback information in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a large model inference data processing system for structured data in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] Specifically, Figure 1 This illustration shows a flowchart of a large-model inference data processing method for structured data according to an embodiment of the present invention. The implementation environment of this application is typically a financial data analysis platform, which integrates a large model, a database system, and a permission management module, aiming to provide financial analysts with efficient and secure data query services. The core of the large-model inference data processing method for structured data proposed in this application lies in intelligently handling query failures caused by permission restrictions and providing effective alternatives. This method includes the following steps: S101. Receive natural language query requests sent by users based on financial transactions; Here, a natural language query request refers to a data query need submitted by a user in everyday language, such as "Query the total transaction amount of all customers in the past three months." Specifically, the method first receives natural language query requests sent by users based on financial transactions. For example, a user might enter, "Please help me query the contact information and transaction records of all VIP customers for the past year." This request can be in text form, entered through the front-end interface, or in voice form, converted into text using speech recognition technology.

[0023] S102. Parse the semantic content in the natural language query request; The semantic content here refers to the core meaning and intent extracted from natural language query requests, including the topic, object, and conditions of the query.

[0024] The parsing process aims to extract the user's true query intent and required data from their natural language. For example, for a request to "query the contact information and transaction records of all VIP customers for the past year," the semantic content might be parsed as: the query object is "VIP customers," and the required data includes "contact information" and "transaction records for the past year." Parsing can employ Natural Language Processing (NLP) techniques, such as lexical analysis, syntactic analysis, and named entity recognition, to transform unstructured natural language into a structured semantic representation.

[0025] S103. Generate a first SQL query statement based on the semantic content; The first SQL query here is a structured query language statement initially generated based on semantic content, used to retrieve data from the database.

[0026] After obtaining the semantic content, the system will convert it into one or more SQL query statements based on the preset database table structure and field mapping rules. For example, if "VIP customer" corresponds to the "customer" table in the database, and "contact information" corresponds to the "contact_info" field, and "transaction records for the most recent year" corresponds to the "amount" and "date" fields in the "transaction" table, then an SQL statement similar to "SELECT customer.contact_info,transaction.amount, transaction.date FROM customer JOIN transaction ON customer.id = transaction.customer_id WHERE customer.type = 'VIP' AND transaction.date >= DATE_SUB(CURDATE(), INTERVAL 1 YEAR)" might be generated.

[0027] S104. Execute the first SQL query statement and obtain the query feedback information returned by the database based on the first SQL query statement; The query feedback information here refers to the status and results returned by the database after executing the first SQL query statement. It may include data from successful queries, or it may include error or access denial prompts.

[0028] The first SQL query will be sent to the database management system for execution. After execution, the database will return the corresponding query results or error information. For example, if the user does not have access rights to the "contact_info" field, the database may return an "Access Denied" error message instead of the actual query data.

[0029] S105. Based on the query feedback information, identify the restricted data elements in the semantic content; Restricted data elements here refer to database fields or data items that the user's current permissions cannot access.

[0030] When the database returns query feedback indicating a query failure (e.g., returning "Access Denied" or "Query Error"), the system needs to further analyze the cause of the failure. One way to identify restricted data elements is for the system to attempt to query permissions for each field listed in the initial SQL query. For example, a simple query can be constructed for each field, and the database's response can be observed. If a query attempt for a particular field fails, that field is determined to be a restricted data element.

[0031] S106. Based on the semantic content and the restricted data elements, perform semantic content evaluation on the data elements in the first SQL query statement to obtain alternative query content; It should be noted that the alternative query content includes alternative data elements and / or query granularity; here, alternative query content refers to the new data elements or adjusted query granularity used to replace the restricted data elements after semantic evaluation and compliance checks have been performed after the restricted data elements have been identified.

[0032] After identifying restricted data elements, the system performs semantic content evaluation on the data elements in the first SQL query statement based on the semantic content and the restricted data elements to obtain alternative query content. Alternative query content includes alternative data elements and / or query granularity. For example, if "contact information" is a restricted data element, the system will attempt to find suitable alternatives based on the user's original semantic content (querying VIP customer information) and the restricted data element (contact information). One alternative is that the system can match an alternative strategy from a predefined set of financial business data alternative rules. For example, if "contact information" is restricted, the alternative strategy might be to provide non-sensitive information such as "customer number" or "customer type" as alternative data elements. Another alternative is that the system can adjust the query granularity, for example, replacing "contact information" with "customer's city," or adjusting "transaction records for the past year" to "total transaction amount for the past month," to circumvent access restrictions.

[0033] S107. Generate a second SQL query statement based on the alternative query content, and feed back the query results obtained from the database based on the second SQL query statement to the user.

[0034] The second SQL query here is a regenerated SQL query based on the alternative query content, designed to avoid permission issues and successfully retrieve the data.

[0035] After determining the alternative query content, the system reconstructs a new SQL query statement, known as the second SQL query statement. This new SQL query statement will no longer contain restricted data elements, but will instead use alternative data elements or an adjusted query granularity. For example, if "Contact Information" is replaced with "Customer ID," the second SQL query statement might become: "SELECT customer.id, transaction.amount, transaction.date FROM customer JOINtransaction ON customer.id = transaction.customer_id WHERE customer.type = 'VIP' AND transaction.date >= DATE_SUB(CURDATE(), INTERVAL 1 YEAR)". This second SQL query statement is sent to the database for execution, and the query results are obtained. Finally, the query results, along with corresponding explanations (e.g., informing the user that "Contact Information" cannot be provided due to permission issues and has been replaced with "Customer ID"), are returned to the user.

[0036] This application introduces a restricted data element identification and semantic content evaluation mechanism, which can intelligently identify and replace data elements that are not authorized to be accessed, thereby generating executable SQL query statements. This effectively solves the problems of query failure and incomplete data caused by permission issues in the prior art, and significantly improves the efficiency and accuracy of financial data analysis.

[0037] First, the method in this embodiment of the invention accurately captures the semantic content of a user's natural language query request by receiving and parsing it, and initially generates a first SQL query statement. Crucially, after executing the first SQL query statement and obtaining query feedback information, this application can intelligently identify restricted data elements within the semantic content based on the feedback information. This step avoids the inefficiency of blind trial and error in traditional methods, directly locating the specific field with restricted permissions.

[0038] Furthermore, this application performs semantic content evaluation on the data elements in the first SQL query statement based on the original semantic content and the identified restricted data elements, thereby obtaining alternative query content. The alternative query content may include alternative data elements or adjustments to the query granularity. This mechanism enables the system to provide compliant and valuable alternatives even when the original query conditions are not fully met, ensuring the continuity and effectiveness of the query. For example, when a user does not have permission to access the sensitive "contact information" field, the system can intelligently replace it with non-sensitive but still analytically valuable data elements such as "customer number" or "customer's city," or adjust the query granularity from "detailed transaction records" to "total transaction amount" to circumvent access restrictions.

[0039] Finally, this application generates a second SQL query statement based on the alternative query content and returns the execution result to the user. This process ensures that the data returned to the user has undergone permission checks and adjustments, satisfying data security and compliance requirements while preserving the user's query intent and value to the greatest extent possible.

[0040] This application, by accurately identifying restricted data elements and combining this with semantic content evaluation, provides more precise and intelligent alternatives, thereby generating executable SQL queries. This approach not only significantly improves query success rate and efficiency but also ensures data integrity and analytical value under restricted access conditions. In this way, this application enables large models to better adapt to the increasingly stringent access control requirements in the financial sector when processing structured data, providing users with more reliable and efficient data analysis services.

[0041] Specifically, parsing the semantic content in the natural language query request in this application includes: identifying the query keywords and query entities in the natural language query request; parsing the actual business needs based on the query keywords and query entities; matching the actual business needs with the corresponding business objectives based on a preset business strategy; and mapping each piece of data in the natural language query request to the specific field corresponding to the business objective in the database.

[0042] Identifying query keywords and query entities in the natural language query request refers to using natural language processing (NLP) techniques, such as lexical analysis, syntactic analysis, and named entity recognition, to extract core query intent words (query keywords) and specific objects related to financial transactions (query entities) from the user's input natural language query request. For example, in a request to "check Zhang San's bank card balance," "check" can be identified as a query keyword, and "Zhang San," "bank card," and "balance" can be identified as query entities.

[0043] Furthermore, parsing the actual business needs based on the query keywords and query entities means, after identifying the query keywords and query entities, combining them with knowledge graphs or preset rules in the financial business domain to understand the deeper business intent of the user's query. For example, when the query keyword is "query" and the query entities include "bank card" and "balance", the actual business need can be parsed as "retrieving the bank card account balance information of a specified user".

[0044] Based on this, matching the actual business needs with corresponding business objectives according to preset business strategies refers to comparing the parsed actual business needs with predefined financial business objectives within the system. Preset business strategies may include a series of business rules, business process definitions, or business models to guide how to map users' abstract needs to specific, executable business operations. For example, the actual business need to obtain bank card balances might be matched with the business objective of "account information query."

[0045] Finally, mapping each piece of data in the natural language query request to the specific field corresponding to the business objective in the database means, after determining the business objective, associating each data element involved in the query request (such as "Zhang San" or "bank card number") with the specific table and field corresponding to that business objective in the database. This typically involves database schema matching, data dictionary lookups, and business logic validation to ensure the accuracy and effectiveness of the data mapping.

[0046] This application's solution progressively refines and structures complex natural language query requests. It first identifies key information at the surface level, then delves into the user's true business intent, precisely matching it with the system's pre-defined business objectives. This ensures the accuracy of subsequent data mapping, laying a solid foundation for generating effective SQL queries. Through this progressive parsing mechanism, unstructured natural language can be effectively transformed into structured query elements that can be processed by the database, thereby enabling large models to make accurate inferences about structured data.

[0047] The method of mapping each piece of data in a natural language query request to the specific field corresponding to the business target in the database in this invention includes: matching each piece of data with the database table structure of the business target to obtain the database element of each piece of data; parsing the business value priority of each piece of data in the business target; adjusting the arrangement order of the database elements of each piece of data in the database table structure based on the business value priority; and mapping each piece of data to the specific field corresponding to the business target based on the database table structure under the arrangement order.

[0048] Each piece of data refers to an independent unit of information or entity identified from a query request after parsing its semantic content. For example, in a query request to "query Zhang San's account balance," both "Zhang San" and "account balance" can be considered data. The database table structure for business objectives refers to the logical organization and physical storage structure of database tables related to the matched business objectives, defining information such as fields, data types, primary keys, foreign keys, and indexes. Database elements, on the other hand, refer to the specific fields or data items corresponding to each piece of data in the database table structure.

[0049] Furthermore, prioritizing the business value of each piece of data within the business objective means evaluating and ranking the importance of each piece of data in achieving the business objective based on pre-defined business rules, user roles, or the current business context. For example, when processing a transaction, the transaction amount may have a higher priority than the transaction time. Adjusting the arrangement order of database elements in the database table structure based on business value priority means optimizing the logical order of database elements during querying or processing according to the importance of the data, ensuring that high-priority data is processed first or utilized more effectively. For example, high-priority data elements can be placed earlier in the query statement or given higher processing weight in the data processing flow. Thus, the database table structure based on the arrangement order maps each piece of data to the specific field corresponding to the business objective, ensuring the accuracy of data mapping and business relevance, thereby improving the efficiency of subsequent SQL query generation and the accuracy of query results.

[0050] This application's solution first matches each piece of data with the database table structure of the business objective, thereby initially determining its corresponding database element. Based on this, by analyzing the business value priority of each piece of data within the business objective, the system can understand the importance of different data within the business context. Furthermore, based on the obtained business value priority, the arrangement order of each piece of data's database elements in the database table structure is adjusted, ensuring that high-value data elements are prioritized or utilized more effectively during subsequent SQL query generation. Finally, based on this optimized arrangement order and database table structure, each piece of data is precisely mapped to the specific field corresponding to the business objective, thus laying the foundation for generating accurate SQL queries.

[0051] Specifically, Figure 2 This invention illustrates a flowchart of a method for identifying restricted data elements in semantic content based on query feedback information, as shown in an embodiment of the invention. The method specifically includes: S201. Determine whether there is an abnormal query status in the query feedback information; It should be noted that the abnormal query status includes access denied or query error messages. Specifically, an abnormal query status means that the database failed to return the expected valid data when executing the first SQL query statement, instead returning information indicating operation failure or insufficient permissions. This may include error codes and error messages returned by the database, such as "Access denied," "Insufficient permissions," or "Table or column does not exist." These messages typically indicate that the user or the current session does not have sufficient access rights to certain data elements, or that the query itself contains syntax or logical errors that prevent normal execution.

[0052] S202. When it is determined that there is an abnormal query status, a field permission detection process is generated based on each field in the first SQL query statement. When an abnormal query is detected, to pinpoint which specific data elements caused the permission issue or query failure, the system generates a field permission probing process for each field in the first SQL query. This field permission probing process can be understood as constructing a separate, minimal query request for each data column, filter condition, or aggregation operation involved in the first SQL query to test the access permissions of that specific field. For example, if the first SQL query involves three fields: "User ID," "Account Balance," and "Transaction Date," separate probing processes will be generated for each of these three fields.

[0053] S203. Execute the permission detection process for each field one by one, and obtain the field query results returned by the database based on the permission detection process for each field; These field permission detection processes will be executed one by one. The database will return the field query results based on each field permission detection process. These results will be used to determine the access permissions or query status of each field.

[0054] S204. Identify the failure status in the field query results, and determine that the field corresponding to the field permission detection process is a restricted data element based on the failure status.

[0055] The system identifies failure states in field query results. A failure state indicates that the database returned an error message similar to an abnormal query state when executing a field permission probe process, suggesting that the field cannot be accessed or queried normally. Based on this failure state, the system can accurately determine that the field corresponding to the field permission probe process is a restricted data element. A restricted data element is a data field that the user currently does not have permission to access or query, requiring subsequent alternative processing.

[0056] This application's solution improves processing efficiency by first assessing the overall query feedback for anomalies, avoiding unnecessary field-level permission probes when queries are normal. When an anomaly is detected, refined field-level permission probes pinpoint the specific restricted data element causing the query failure. This mechanism of probing and identifying failure states field by field ensures high accuracy and reliability in identifying restricted data elements, providing accurate foundational information for generating subsequent alternative query content.

[0057] Specifically, the field permission detection process based on each field in the first SQL query statement in this invention includes: identifying the data column, filtering conditions, and aggregation operations of each field in the first SQL query statement; and generating the field permission detection process based on the data column, filtering conditions, and aggregation operations.

[0058] In the process of identifying restricted data elements in the semantic content based on the query feedback information, when an abnormal query state is determined, permission detection needs to be performed on each field in the first SQL query statement. To effectively generate the field permission detection process, a thorough analysis of each field in the first SQL query statement is first required. Specifically, this involves identifying the data column, filter conditions, and aggregation operations for each field in the first SQL query statement. The data column refers to the column name requested to be returned in the SQL query; the filter condition refers to the conditional expression used to filter data for that field; and the aggregation operation refers to statistical operations such as summation, counting, and averaging performed on that field. By identifying these key components, a comprehensive understanding of the specific role and context of each field in the original query can be achieved.

[0059] Furthermore, a field permission detection process is generated based on the data columns, filtering conditions, and aggregation operations. This means that a dedicated, targeted permission detection process can be built for each field. For example, if a field is used as a data column, the detection process may need to verify whether the user has permission to read the data in that column; if a field involves filtering conditions, the detection process may need to verify whether the user has permission to use those conditions for filtering; if a field involves aggregation operations, the detection process may need to verify whether the user has permission to execute those aggregation operations. Therefore, permission detection logic can be tailored to each field, ensuring the accuracy and comprehensiveness of the detection.

[0060] The solution in this application identifies the data columns, filtering conditions, and aggregation operations corresponding to each field in the first SQL query statement, enabling a refined analysis of the specific role and permission requirements of each field in the SQL query. It is precisely this refined analysis that allows the subsequently generated field permission detection process to more accurately simulate how the database accesses that field when executing the original query. By closely integrating the detection process with the actual usage of the field, overly broad or overly narrow detection logic can be avoided, thereby improving the efficiency and accuracy of permission detection.

[0061] To address this, this application further proposes a method for obtaining alternative query content by semantically evaluating data elements in a first SQL query statement based on semantic content and restricted data elements. The method includes the following steps: matching corresponding substitution strategies from a set of financial business data substitution rules based on the semantic content and the restricted data elements; performing a compliance pre-check on potential alternative data elements based on the substitution strategies to obtain potential alternative data elements that meet compliance requirements; prioritizing the potential alternative data elements that meet compliance requirements based on the substitution strategies and the current working context to obtain a priority ranking result; and selecting potential alternative data elements with a high degree of matching with the semantic content as alternative query content based on the priority ranking result.

[0062] Specifically, upon receiving a user's natural language query request and identifying the restricted data elements it contains, the system first matches the most suitable alternative strategy from a pre-built set of financial business data substitution rules based on the semantic content and the restricted data elements. This set of financial business data substitution rules can be understood as a knowledge base containing various predefined rules and logics, used to guide how to safely, compliantly, and effectively find alternative data elements or adjust query granularity in specific financial business scenarios when the original data is restricted. These rules can be categorized and defined based on dimensions such as data sensitivity, business type, user permissions, and regulatory requirements.

[0063] Furthermore, once the alternative strategy is determined, the system will conduct a compliance pre-check on all potential alternative data elements based on that strategy. The compliance pre-check refers to the initial screening and verification of candidate alternative data elements before actually generating alternative query content, to ensure they comply with relevant laws and regulations, industry standards, and the company's internal data usage policies. For example, it checks whether the alternative data involves personal privacy, whether anonymization is required, and whether it meets the principle of data minimization. This step effectively filters out potential alternative data elements that do not meet compliance requirements, thus obtaining potential alternative data elements that do meet compliance requirements and avoiding compliance risks in subsequent operations.

[0064] Based on this, the system will prioritize potential alternative data elements that meet compliance requirements according to the matched alternative strategies and the current specific working context. The current working context can include, but is not limited to, dynamic factors such as system load, real-time market data, regulatory policy changes, user roles, and query purposes. For example, during periods of high market volatility, alternative data related to risk management may be given higher priority; when the user is a senior analyst, more granular but processed alternative data may be provided. Prioritization ensures that the most suitable alternative data elements for the current business needs and context are selected from among numerous compliant alternatives.

[0065] Ultimately, the system will select the potential alternative data element with the highest matching degree to the original semantic content as the final alternative query content based on the priority ranking results. A high matching degree means that the selected alternative data element is semantically closest to the user's original query intent, which can preserve the business value of the query to the greatest extent while avoiding the problems caused by restricted data elements. Therefore, the alternative query content can include alternative data elements and / or query granularity. For example, replacing the query for sensitive transaction details of a specific customer with the query for anonymized transaction statistics of the customer's group, or adjusting the query for transaction data accurate to the day to the query for transaction data accurate to the month.

[0066] This application's solution constructs a multi-dimensional and intelligent alternative query content generation process by introducing a set of rules for financial business data substitution, compliance pre-checks, and a priority ranking mechanism based on the current working context. This enables the system to not only find compliant alternative data when raw data is limited, but also to select the most optimized and relevant alternative solutions based on actual business needs and dynamic contexts.

[0067] In some preferred embodiments, a specific example is given below. Suppose a user initiates a natural language query request: "Query customer Zhang San's credit card spending details for the past month." After parsing the semantic content, the system generates the first SQL query statement. However, during execution, it discovers that "customer Zhang San's credit card spending details" is a restricted data element because it involves personal privacy, and the user does not have sufficient permissions to directly access it.

[0068] At this point, the system will initiate the alternative query content generation process. First, based on the semantic content ("Customer Zhang San", "Credit Card Transaction Details", "Last Month") and the restricted data element ("Customer Zhang San's Credit Card Transaction Details"), the system will match the corresponding alternative strategy from a pre-set set of financial business data alternative rules. This strategy may stipulate that for sensitive personal transaction details, anonymized, aggregated, or higher-granularity alternative data may be provided.

[0069] Next, the system performs a compliance pre-check on potential alternative data elements based on the substitution strategy. For example, potential alternative data elements might include: "average spending trends of customer Zhang San's customer group," "statistics on customer Zhang San's anonymized spending categories," and "total credit card spending in customer Zhang San's region." The compliance pre-check ensures that these alternatives do not violate personal information protection regulations.

[0070] The system then prioritizes potential alternative data elements that meet compliance requirements based on alternative strategies and the current work context. For example, if the current work context indicates that the user's role is a risk analyst and there has been a recent credit card fraud alert, the system may assign higher priority to "Anonymized spending category statistics for customer Zhang San" or "Average spending trends of customer Zhang San's customer group," because this data is more valuable for risk assessment.

[0071] Finally, based on the priority ranking results, the system selects the potential alternative data element with the highest match to the original semantic content as the alternative query content. For example, if "customer Zhang San's anonymized consumption category statistics" is rated as the highest priority and has a high match with the user's query intent, the system will use it as the alternative query content and generate a second SQL query statement based on it, thereby providing the user with compliant and valuable query results.

[0072] To address this, this application further proposes prioritizing potential alternative data elements that meet compliance requirements based on the substitution strategy and the current work context. The prioritization result includes: monitoring external dynamic events associated with the current work context; assessing the relevance of the external dynamic events to the potential alternative data elements that meet compliance requirements; adjusting the business value and compliance importance of the potential alternative data elements based on the external dynamic events to obtain adjusted priority weights; and weighting the potential alternative data elements based on the priority weights and the preset weights of the substitution strategy to obtain the priority ranking result.

[0073] Specifically, monitoring external dynamic events related to the current work context refers to the system continuously monitoring external information sources relevant to the current financial business context, such as financial market news, regulatory policy updates, industry reports, and social media trends. These external dynamic events can be structured data streams or unstructured text information, with the aim of acquiring external changes that may affect the business value and compliance of data elements in real time.

[0074] Assessing the relevance of the external dynamic event to the potential substitute data elements that meet compliance requirements can be understood as analyzing the degree and direction of the potential impact of the monitored external dynamic event on the potential substitute data elements. For example, an external dynamic event about "stricter regulation in a certain industry" may be highly correlated with potential substitute data elements such as "customer data related to that industry," but less correlated with "financial data from other industries." This assessment aims to quantify the potential impact of external events on specific data elements.

[0075] In practical applications, the business value and compliance importance of potential substitute data elements that meet compliance requirements are adjusted based on the external dynamic events, resulting in adjusted priority weights. Specifically, after identifying the correlation between external dynamic events and potential substitute data elements, the system dynamically adjusts the inherent business value and compliance importance of these potential substitute data elements based on the nature and impact of the event and pre-defined business logic. For example, if an external event indicates an increase in compliance risk for a specific type of data, the compliance importance of that data will be increased, and its business value may be enhanced due to its key role in risk management, thus affecting its final priority weight. The adjusted priority weights reflect the true value and risk level of the data elements in the current dynamic context.

[0076] Furthermore, the priority ranking result is obtained by weighting the potential alternative data elements that meet the compliance requirements based on the priority weights and the preset weights of the alternative strategies. This involves comprehensively weighting the priority weights (which reflect real-time business value and compliance importance) adjusted by external dynamic events with the preset weights in the alternative strategies (which reflect general business rules and preferences). Through this weighted scoring mechanism, a comprehensive score can be generated for each potential alternative data element. This score can fully reflect its applicability and optimality in the current dynamic context. Finally, these scores are used to rank the data elements, resulting in the final priority ranking result.

[0077] This application's solution introduces a mechanism for monitoring and evaluating external dynamic events, making the prioritization of potential alternative data elements no longer static or based solely on preset rules. It is precisely because of the ability to perceive changes in the external environment in real time and dynamically adjust the business value and compliance importance of data elements accordingly that the priority weights can more accurately reflect the true applicability of data elements in the current complex financial business context. By combining this dynamically adjusted priority weight with the preset weights of the alternative strategy for weighted scoring, it ensures that the final priority ranking not only conforms to the established business strategy but also flexibly responds to the challenges and opportunities of the external environment, thereby effectively solving the potential lag and inaccuracy issues in the prioritization of basic solutions in dynamic scenarios.

[0078] In some preferred embodiments, a specific example is given below. Suppose a user queries "a customer's earnings for the previous quarter," but the "earnings" data is not directly accessible due to access restrictions. The system identifies "earnings" as a restricted data element and finds potential alternative data elements based on an alternative strategy, such as "customer asset changes," "customer transaction volume," and "customer risk exposure."

[0079] At this point, the system will monitor external dynamic events related to the current work context. For example, if the system detects an external dynamic event about "financial regulatory agencies issuing new rules requiring strengthened disclosure and management of customer risk exposures".

[0080] Next, the system assesses the correlation between this external dynamic event and potential alternative data elements. Clearly, "Customer Risk Exposure" is highly correlated with the new regulatory rule, while "Customer Asset Changes" and "Customer Transaction Volume" show lower correlations.

[0081] Based on this external dynamic event, the system will adjust the business value and compliance importance of potential alternative data elements. Due to new regulations, the compliance importance of "customer risk exposure" has been significantly increased, and its business value may also be enhanced due to its key role in risk management, thus receiving an adjusted priority weight. The weights of other data elements may remain unchanged or be slightly adjusted.

[0082] Finally, based on the adjusted priority weights and the preset weights for alternative strategies, a weighted score is applied to "Customer Asset Changes," "Total Customer Transaction Volume," and "Customer Risk Exposure." Due to the significantly increased importance of "Customer Risk Exposure" in compliance, its overall score is likely to be the highest, thus placing it first in the priority ranking. Ultimately, the system may prioritize recommending "Customer Risk Exposure" as an alternative query and generate a corresponding second SQL query to meet the user's query needs within the compliance framework.

[0083] Specifically, the assessment of the relevance of external dynamic events to potential alternative data elements that meet compliance requirements in this invention includes the following steps: First, identify the types and impact areas of external dynamic events. Types of external dynamic events can encompass various forms, including market fluctuations, policy and regulatory changes, internal company adjustments, and sudden public events. The impact area refers to the business scope that the event may affect, such as the stock market, bond market, specific industries, or specific customer groups. This step aims to initially classify and define the scope of external events, providing precise input for subsequent transmission path analysis.

[0084] Furthermore, based on the type and impact domain of the identified external dynamic events, a search is conducted within the financial business logic transmission path library for transmission paths related to potential alternative data elements that meet compliance requirements. The financial business logic transmission path library is a pre-built knowledge base that stores paths of how various financial business events affect other business elements or data elements through different business logics and data flows. Each path may contain multiple nodes and edges representing the transmission process of the event. By searching within this library, it is possible to reveal how external events specifically affect particular data elements.

[0085] Subsequently, the analysis examines the transmission levels and latency defined in the transmission paths related to potential alternative data elements that meet compliance requirements. Transmission levels refer to the number of business processes or data transformations an event passes through from its source to its impact on the target data element; more levels generally indicate a more indirect impact. Latency refers to the time required for the event's impact to propagate to the target data element after it occurs; this can be real-time, minutes-level, hourly, or day-level. This analysis aims to quantify the depth and temporal dimension of the event's impact.

[0086] Based on this, the effective window for the impact of external dynamic events is adjusted according to the delay time defined in the transmission path. The effective window refers to when the impact of an external event begins and how long it lasts. For example, if the delay time defined in the transmission path is 1 hour, the effective window might begin 1 hour after the event occurs. This adjustment aims to ensure that the impact of the event is assessed within the correct timeframe.

[0087] Simultaneously, the indirectness of the impact of external dynamic events is assessed based on the transmission hierarchy defined in the transmission path. A higher transmission hierarchy generally indicates a more indirect impact and potentially a weaker correlation. For example, a direct impact has a lower degree of indirectness, while an impact transmitted through multiple layers has a higher degree of indirectness. This assessment aims to quantify the directness of the event's impact.

[0088] Finally, based on the adjusted effective window and the assessed degree of indirection, the relevance between external dynamic events and potential alternative data elements that meet compliance requirements is evaluated. This assessment comprehensively considers both the time dimension (effective window) and the path depth dimension (degree of indirection) to quantify the strength of the association between external events and potential alternative data elements. For example, events within the effective window and with a lower degree of indirection will have higher relevance scores.

[0089] This application's solution systematically identifies the characteristics of external dynamic events and utilizes a pre-defined financial business logic transmission path library to track the specific impact paths of these events on potential substitute data elements. This allows for the precise acquisition of the event's transmission hierarchy and latency, enabling dynamic adjustment of the event's effective window and assessment of its indirectness. It is precisely this refined analysis based on business logic and time dimensions that makes the correlation assessment between external dynamic events and potential substitute data elements more accurate and objective, thus providing a solid data foundation for subsequent prioritization of substitute data elements.

[0090] Specifically, the steps described above for feeding back the query results obtained from the database based on the second SQL query statement to the user include the following: receiving the query results fed back by the database based on the second SQL query statement; configuring the interpretation content corresponding to the second SQL query statement based on the query results; and feeding back the query results and the interpretation content to the user.

[0091] Receiving the query results from the database based on the second SQL query statement refers to the system receiving the actual data output corresponding to that query statement from the database after the second SQL query statement is executed. This query result is typically presented in the form of structured data, such as tables, lists, or JSON format, containing the final data after semantic evaluation and substitution processing of the user's initial query request.

[0092] Furthermore, configuring the explanatory content corresponding to the second SQL query statement based on the query results aims to transform the original, potentially abstract or technical query results into a natural language description that is easy for users to understand. Specifically, an explanatory text can be generated based on the data type, numerical range, and business meaning of the query results, explaining the source, meaning, and relevance to the user's original query request. For example, if the query result is the yield rate of a financial product, the explanatory content can explain how the yield rate is calculated, the time period it represents, and its significance in the current market environment. The purpose is to improve the user's understanding of the query results and avoid misunderstandings caused by the data itself being obscure or difficult to understand.

[0093] Therefore, providing query results and explanations to users means presenting processed and interpreted structured data along with corresponding natural language interpretations. This can be achieved in various ways, such as displaying the data in a combination of charts and text on the user interface, or providing it verbally via a voice assistant. The goal is to offer comprehensive, intuitive, and easy-to-understand query feedback, enabling users not only to see the data but also to understand the business logic and meaning behind it, thereby better assisting them in making decisions.

[0094] This application's solution first receives the query results from the database based on a second SQL query, ensuring the acquisition of accurate data that has undergone optimization and compliance processing. Then, by configuring corresponding explanatory content based on the query results, the raw, potentially technical data is transformed into user-friendly natural language descriptions, significantly lowering the barrier to data comprehension for users. Finally, the query results and explanatory content are presented to the user together, enabling them to not only obtain the required data but also clearly understand its background, meaning, and business value, thereby improving user experience and data utilization efficiency.

[0095] Specifically, Figure 3 This diagram illustrates the structure of a large-model inference data processing system for structured data according to an embodiment of the present invention. The system includes: The receiving module is used to receive natural language query requests sent by users based on financial transactions; The parsing module is used to parse the semantic content in the natural language query request; The first generation module is used to generate a first SQL query statement based on the semantic content; The first execution module is used to execute the first SQL query statement and obtain the query feedback information returned by the database based on the first SQL query statement; The identification module is used to identify restricted data elements in the semantic content based on the query feedback information; The second generation module is used to perform semantic content evaluation on the data elements in the first SQL query statement based on the semantic content and the restricted data elements to obtain alternative query content, wherein the alternative query content includes alternative data elements and / or query granularity. The second execution module generates a second SQL query statement based on the alternative query content and returns the query results obtained from the database based on the second SQL query statement to the user. Through modular design, the system can intelligently receive and parse the user's natural language query request, converting it into a preliminary SQL query statement. After executing the statement and obtaining feedback from the database, the system can accurately identify restricted data elements and evaluate them based on semantic content and restricted data elements to generate compliant alternative query content. Finally, the system uses the alternative query content to generate and execute the second SQL query statement and returns accurate query results that meet permission requirements to the user. This systematic approach effectively avoids the inefficient trial-and-error process of traditional solutions, significantly improving the success rate, efficiency, and reliability of data queries.

[0096] The specific steps of the large-scale model inference data processing method for structured data proposed in this application have been described in the above embodiments, and will not be repeated here. It should be emphasized that this application also provides a large-scale model inference data processing system for structured data, which implements the above method through the collaborative work of various functional modules.

[0097] Specifically, the receiving module can be configured as a front-end application programming interface (API) or a message queue listener to capture natural language query requests submitted by users through various channels, such as web interfaces, mobile applications, or chatbots. This module ensures reliable reception of query requests and performs preliminary format validation on the received requests.

[0098] The parsing module can be implemented as a Natural Language Processing (NLP) engine, integrating a lexical analyzer, a syntactic analyzer, and a named entity recognizer. This module is responsible for transforming received natural language query requests into structured semantic representations; for example, it can identify the subject, predicate, object, and modifiers such as time and location. In some implementations, the parsing module can employ various techniques such as rule-based parsing, statistical model parsing, or deep learning model parsing.

[0099] The first generation module can be configured as an SQL generator. Its core is based on preset database metadata and mapping rules, converting the structured semantic content output by the parsing module into SQL query statements that conform to database syntax. This module can utilize template matching, semantic-to-SQL mapping models, or code generation capabilities based on large models to construct the first SQL query statement.

[0100] The first execution module can be implemented as a database connection pool management component, responsible for establishing and maintaining connections with the target database, and sending the SQL query statements generated by the first generation module to the database for execution. This module is also responsible for receiving query feedback information returned by the database, including query result sets or various error messages.

[0101] The identification module can be designed as a permission analyzer. Its function is to determine whether there are any permission restrictions based on the query feedback information obtained by the first execution module. When the feedback information indicates that the query failed, the identification module will further analyze each data element in the first SQL query statement and determine the specific restricted data element through a permission detection mechanism (e.g., performing permission tests on each field).

[0102] The second generation module can be configured as an intelligent query optimizer. Its core function is to perform semantic content evaluation on the data elements in the first SQL query statement based on the original semantic content provided by the parsing module and the restricted data elements determined by the identification module. This module matches appropriate substitution strategies from a predefined substitution rule base and generates alternative query content, such as replacing data elements with non-sensitive ones or adjusting the query granularity.

[0103] The second execution module can be implemented as a database interaction component similar to the first execution module, but its input is a second SQL query statement generated by the second generation module. This module is responsible for sending the optimized SQL query statement to the database for execution and receiving the final query results. Subsequently, this module will feed back the query results along with necessary explanations (e.g., indicating that some data has been replaced due to permission restrictions) to the user, ensuring that the user receives valid and compliant information.

[0104] The large-scale model inference data processing system for structured data proposed in this application demonstrates significant progress compared to existing technologies in handling granular database access permissions in the financial field. Traditional systems, when translating natural language queries into SQL statements, often lack awareness of specific user field permissions, leading to frequent database rejections of generated SQL statements due to insufficient permissions. More importantly, existing systems struggle to accurately diagnose the root cause of problems when receiving ambiguous error feedback, resorting to inefficient trial and error or speculative field removal to attempt solutions. This not only consumes substantial computational resources and time but may also result in incomplete data returns, thereby diminishing the value of data analysis.

[0105] Therefore, the system of this application not only significantly improves the success rate and efficiency of queries, but also ensures data integrity and analytical value under restricted access conditions. In this way, this application enables large models to better adapt to the increasingly stringent access control requirements in the financial sector when processing structured data, providing users with more reliable and efficient data analysis services, and effectively overcoming the query failures and incomplete data problems caused by access restrictions in existing technologies.

[0106] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for processing large-scale model inference data for structured data, characterized in that, Includes the following steps: Receive natural language query requests from users based on financial transactions; Parse the semantic content in the natural language query request; A first SQL query statement is generated based on the semantic content; Execute the first SQL query statement and obtain the query feedback information from the database based on the first SQL query statement; Based on the query feedback information, restricted data elements in the semantic content are identified; Based on the semantic content and the restricted data elements, semantic content evaluation is performed on the data elements in the first SQL query statement to obtain alternative query content, which includes alternative data elements and / or query granularity. A second SQL query statement is generated based on the alternative query content, and the query results obtained from the database based on the second SQL query statement are returned to the user. The restricted data elements identified in the semantic content based on the query feedback information include: Determine whether there is an abnormal query status in the query feedback information, including access being denied or query error information; If an abnormal query status is detected, a field permission detection process is generated based on each field in the first SQL query statement; Each field permission detection process is executed sequentially, and the field query results returned by the database are obtained based on each field permission detection process; Identify the failure status in the field query results, and determine that the field corresponding to the field permission detection process is a restricted data element based on the failure status; The step of evaluating the semantic content of data elements in the first SQL query statement based on the semantic content and the restricted data elements to obtain alternative query content includes: Based on the semantic content and the restricted data elements, the corresponding substitution strategy is matched from the set of financial business data substitution rules; Based on the aforementioned alternative strategy, a compliance pre-check is performed on the potential alternative data elements to obtain potential alternative data elements that meet the compliance requirements. Based on the aforementioned alternative strategy and the current working context, the potential alternative data elements that meet the compliance requirements are prioritized to obtain the priority ranking result; Based on the priority ranking results, potential alternative data elements with a high degree of matching with the semantic content are selected as alternative query content.

2. The method for processing large model inference data for structured data according to claim 1, characterized in that, The process of parsing the semantic content in the natural language query request includes: Identify the query keywords and query entities in the natural language query request; Based on the query keywords and query entities, the actual business needs of the query are analyzed; Based on preset business strategies, the corresponding business objectives are matched to the actual business needs; Each piece of data in the natural language query request is mapped to the specific field in the database corresponding to the business objective.

3. The method for processing large model inference data for structured data according to claim 2, characterized in that, The step of mapping each piece of data in the natural language query request to the specific field corresponding to the business objective in the database includes: Each piece of data is matched with the database table structure of the business objective to obtain the database element of each piece of data; Analyze the business value priority of each piece of data within the stated business objectives; The order in which the database elements of each piece of data are arranged in the database table structure is adjusted based on the business value priority. The database table structure based on the arrangement order maps each piece of data to the specific field corresponding to the business objective.

4. The method for processing large model inference data for structured data according to claim 1, characterized in that, The process of generating field permission detection based on each field in the first SQL query statement includes: Identify the data columns, filter conditions, and aggregation operations for each field in the first SQL query statement; A field permission detection process is generated based on the data columns, filtering conditions, and aggregation operations.

5. The method for processing large model inference data for structured data according to claim 1, characterized in that, The priority ranking of potential alternative data elements that meet compliance requirements based on the alternative strategy and the current working context, resulting in the priority ranking results, includes: Listen for external dynamic events related to the current work context; Assess the relevance of the external dynamic events to the potential alternative data elements that meet the compliance requirements; The business value and compliance importance of the potential alternative data elements that meet compliance requirements are adjusted based on the external dynamic events to obtain the adjusted priority weights; The potential alternative data elements that meet the compliance requirements are weighted and scored based on the priority weight and the preset weight of the alternative strategy to obtain the priority ranking result.

6. The method for processing large model inference data for structured data according to claim 5, characterized in that, The assessment of the relevance of the external dynamic event to the potential alternative data elements that meet the compliance requirements includes: Identify the type of the external dynamic event and the area of ​​its impact; Based on the type of the external dynamic event and the area of ​​its impact, search the financial business logic transmission path library for transmission paths related to potential alternative data elements that meet the compliance requirements; Analyze the transmission hierarchy and delay time defined in the transmission path related to the potential alternative data elements that meet the compliance requirements; The effective window of the external dynamic event's influence is adjusted according to the delay time defined in the transmission path; Assess the degree of indirectness of the impact of the external dynamic event based on the transmission hierarchy defined in the transmission path; The correlation between the external dynamic event and the potential alternative data elements that meet the compliance requirements is assessed based on the effective window and the degree of indirection.

7. The method for processing large model inference data for structured data according to claim 1, characterized in that, The step of feeding back the query results obtained from the database based on the second SQL query statement to the user includes: Receive the query results returned by the database based on the second SQL query statement; Configure the interpretation content corresponding to the second SQL query statement based on the query results; The query results and the corresponding explanations will be provided to the user.

8. A large-scale model inference data processing system for structured data, characterized in that, The system includes: The receiving module is used to receive natural language query requests sent by users based on financial transactions; The parsing module is used to parse the semantic content in the natural language query request; The first generation module is used to generate a first SQL query statement based on the semantic content; The first execution module is used to execute the first SQL query statement and obtain the query feedback information returned by the database based on the first SQL query statement; The identification module is used to identify restricted data elements in the semantic content based on the query feedback information; The second generation module is used to perform semantic content evaluation on the data elements in the first SQL query statement based on the semantic content and the restricted data elements to obtain alternative query content, wherein the alternative query content includes alternative data elements and / or query granularity. The second execution module is used to generate a second SQL query statement based on the alternative query content, and to return the query results obtained from the database based on the second SQL query statement to the user; The identification module is also used for: Determine whether there is an abnormal query status in the query feedback information, including access being denied or query error information; If an abnormal query status is detected, a field permission detection process is generated based on each field in the first SQL query statement; Each field permission detection process is executed sequentially, and the field query results returned by the database are obtained based on each field permission detection process; Identify the failure status in the field query results, and determine that the field corresponding to the field permission detection process is a restricted data element based on the failure status; The second generation module is also used for: Based on the semantic content and the restricted data elements, the corresponding substitution strategy is matched from the set of financial business data substitution rules; Based on the aforementioned alternative strategy, a compliance pre-check is performed on the potential alternative data elements to obtain potential alternative data elements that meet the compliance requirements. Based on the aforementioned alternative strategy and the current working context, the potential alternative data elements that meet the compliance requirements are prioritized to obtain the priority ranking result; Based on the priority ranking results, potential alternative data elements with a high degree of matching with the semantic content are selected as alternative query content.

Images