Natural Language-Based Intelligent Analysis Methods, Systems, and Media for Aviation Business Data

Abstract

This invention relates to the field of data processing technology and discloses a method, system, and medium for intelligent analysis of aviation business data based on natural language. The method includes: extracting query indicators, query dimensions, query time ranges, and query conditions; generating target semantic frames and caliber locking results based on indicator caliber data, dimension mapping data, granularity transformation data, and time anchor data; generating an initial query plan; determining the flight segment attribution result corresponding to the record based on business records and business snapshots when preset attribution conditions are met; adjusting the initial query plan according to the flight segment attribution result to obtain the target query plan; generating and executing a structured query statement to obtain statistical results; and performing attribution consistency verification on the statistical results. Statistical results are output when the attribution consistency verification passes. This application improves the accuracy of natural language query understanding in aviation business, the reliability of cross-granularity statistics, and the credibility of the final results.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method, system, and medium for intelligent analysis of aviation business data based on natural language. Background Technology

[0002] As airlines continue to deepen their digital operations, business data such as orders, tickets, flights, settlements, service requests, and cost logs are constantly accumulating. The demand from both management and non-technical personnel for "directly querying data and conducting business analysis using natural language" is increasing. Related technologies have evolved from traditional fixed reports and manually written query statements to data analysis methods that combine large language models, prompts, metadata, and structured query statements. These technologies lower the barrier to entry for database use and improve data acquisition efficiency, demonstrating significant application value and practical necessity in the aviation business context.

[0003] In existing technologies, such as invention patent CN117555986A – an intelligent data analysis method and device based on a large language model; this solution, and similar patent CN118861081A, focus on natural language problem parsing, generation of query codes or structured query statements, execution feedback, and result output, mainly addressing the question of "whether query statements can be generated and executed relatively accurately." However, in aviation business data analysis scenarios, business data often comes from different business systems, and the statistical objects may span different granularities such as log records, orders, tickets, flight segments, and routes, accompanied by different indicator calibers and time calibers. Even if existing technologies can generate and execute query statements, problems such as inconsistencies between query semantics and statistical calibers, insufficient verifiability of statistical results, and unstable accuracy of results under specific dimensions may still occur.

[0004] Therefore, it is necessary to design a natural language-based intelligent analysis method, system, and medium for aviation business data to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes an intelligent analysis method, system and medium for aviation business data based on natural language, aiming to solve the problems of insufficient accuracy, verifiability and stability of query results in the current aviation business natural language query process, due to the dispersed data sources of multiple business systems, inconsistent statistical granularity and easy deviation of indicator and time caliber.

[0006] This invention proposes a natural language-based intelligent analysis method for aviation business data, comprising: Receive natural language query text and extract query metrics, query dimensions, query time range, and query conditions; Based on preset indicator data, dimension mapping data, granularity conversion data, and time anchor data, the natural language query text is subjected to semantic constraint parsing to generate target semantic frames and caliber locking results. An initial query plan is generated based on the target semantic frame and the caliber locking result. When the query indicators and the query dimensions meet the preset attribution conditions, the flight segment attribution result corresponding to the record is determined based on the business record and business snapshot. The initial query plan is then adjusted based on the flight segment attribution result to obtain the target query plan. When the query indicators and the query dimensions do not meet the preset attribution conditions, the initial query plan is used as the target query plan. A structured query statement is generated based on the target query plan. The structured query statement is executed to obtain statistical results, and the statistical results are subjected to attribution consistency verification. The statistical results are output when the attribution consistency verification passes, and the rollback results are output when the attribution consistency verification fails.

[0007] Furthermore, the business records include request logs and business action records, and the business snapshots include order snapshots, ticket snapshots, and flight segment snapshots; based on the request time corresponding to the record, a valid business snapshot matching the request time is selected from the business snapshots as the request time snapshot corresponding to the record.

[0008] Furthermore, when determining the attribution result of the flight segment based on the requested snapshot, the process includes: The target segment is determined based on the segment association information in the record; if the target segment is not determined, it is determined sequentially based on the segment corresponding to the business action, the first valid segment to take off after the request time, and the earliest valid segment.

[0009] Furthermore, when determining the target segment based on the flight segment corresponding to the business action, the earliest valid flight segment to take off after the request time, and the earliest valid flight segment, this includes: A standard segment identifier is generated based on the marketing flight number, the operating flight number, the departure airport, the arrival airport, and the flight date. The target segment is then determined by merging the valid segments based on the standard segment identifier.

[0010] Furthermore, the attribution consistency check includes record coverage check, consistency check of the sum of statistical values ​​before and after attribution, and flight segment validity check; if any check fails, the rollback result is a log-level statistical result.

[0011] Furthermore, when multiple merged valid segments exist, the merged valid segment that has not been identified as the target segment by other records is selected; when no target segment is selected, the merged valid segment that took off first after the request time is selected; when no target segment is selected, the earliest valid merged valid segment is selected; when the sum of the statistical values ​​before and after attribution exceeds the preset minimum statistical unit, log-level statistical results are output.

[0012] Furthermore, the semantic constraint parsing includes: extracting indicator words, dimension words, time words, and condition words from the natural language query text; determining candidate indicators and corresponding statistical granularities based on the indicator caliber data; standardizing the dimension words based on the dimension mapping data; determining the target time caliber based on the time anchor point data; and generating candidate semantic results.

[0013] Furthermore, based on the granularity conversion data, the statistical granularity and query dimension in the candidate semantic results are filtered for relevance, eliminating non-convertible combinations, and the target semantic frame and the caliber locking result are generated according to the filtered candidate indicators, standardized dimensions and target time caliber.

[0014] Compared with existing technologies, the advantages of this invention are as follows: By extracting query indicators, query dimensions, query time range, and query conditions from natural language query text, and combining them with indicator caliber data, dimension mapping data, granularity transformation data, and time anchor data for semantic constraint parsing, it is possible to uniformly constrain the meaning of indicators, dimension mapping relationships, statistical granularity relationships, and time caliber in user expressions before query generation, reducing the risk of erroneous queries caused by ambiguous natural language expressions, mixed use of synonyms, and statistical caliber bias; by first generating an initial query plan, and then determining the flight segment attribution result and adjusting the initial query plan based on business records and business snapshots when preset attribution conditions are met, it is possible to target log records, orders, tickets, flight segments, and flight information in aviation business. The presence of multi-granularity relationships between lines allows for targeted correction of data attribution relationships in specific query scenarios, avoiding potential misattribution, omissions, or duplicate statistics that may occur during direct table joins. This improves the accuracy of statistical results in specific dimensions, especially flight segment and route dimensions. Furthermore, by directly using the initial query plan as the target query plan when preset attribution conditions are not met, the integrity of the overall processing flow is ensured, preventing continuous operation from being affected by whether attribution processing is triggered. By performing attribution consistency checks after obtaining statistical results from structured query statements and outputting rollback results when checks fail, it prevents the direct output of results with attribution conflicts, statistical biases, or caliber anomalies, thus improving the verifiability, stability, and output security of statistical results.

[0015] On the other hand, this application also provides a natural language-based intelligent analysis system for aviation business data, used to apply the above-mentioned natural language-based intelligent analysis method for aviation business data, including: The data collection unit is configured to receive natural language query text and extract query metrics, query dimensions, query time range, and query conditions. The processing unit is configured to perform semantic constraint parsing on the natural language query text based on preset indicator caliber data, dimension mapping data, granularity conversion data, and time anchor data, and generate target semantic frames and caliber locking results. The execution unit is configured to generate an initial query plan based on the target semantic frame and the caliber locking result; when the query indicators and the query dimensions meet preset attribution conditions, determine the flight segment attribution result corresponding to the record based on the business record and business snapshot, and adjust the initial query plan according to the flight segment attribution result to obtain a target query plan; when the query indicators and the query dimensions do not meet the preset attribution conditions, the initial query plan is used as the target query plan. The output unit is configured to generate a structured query statement according to the target query plan, execute the structured query statement to obtain statistical results, and perform attribution consistency verification on the statistical results; output the statistical results when the attribution consistency verification passes, and output the rollback results when the attribution consistency verification fails.

[0016] On the other hand, this application also provides a computer-readable storage medium in which computer-executable instructions stored in the computer-readable storage medium are executed by a processor of an electronic device, enabling the electronic device to perform the above-described method.

[0017] It is understandable that the above-mentioned intelligent analysis methods, systems and storage media for aviation business data based on natural language have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating the intelligent analysis method for aviation business data based on natural language provided in this embodiment of the invention; Figure 2 A functional block diagram of an intelligent analysis system for aviation business data based on natural language, provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] In some embodiments of this application, see Figure 1 As shown, this application proposes a natural language-based intelligent analysis method for aviation business data, including: S100: Receives natural language query text and extracts query metrics, query dimensions, query time range, and query conditions.

[0021] S200: Based on preset indicator data, dimension mapping data, granularity conversion data, and time anchor data, semantic constraint parsing is performed on natural language query text to generate target semantic frames and definition locking results.

[0022] S300: Generate an initial query plan based on the target semantic frame and caliber locking results. When the query metrics and query dimensions meet the preset attribution conditions, determine the flight segment attribution results corresponding to the records based on business records and business snapshots, and adjust the initial query plan according to the flight segment attribution results to obtain the target query plan. When the query metrics and query dimensions do not meet the preset attribution conditions, the initial query plan is used as the target query plan.

[0023] S400: Generates a structured query statement based on the target query plan, executes the structured query statement to obtain statistical results, and performs attribution consistency verification on the statistical results. Outputs the statistical results if the attribution consistency verification passes, and outputs a rollback result if the attribution consistency verification fails.

[0024] Specifically, in this embodiment, the business database includes at least an order data table, a ticket data table, a flight segment data table, a service request log table, a business action record table, and a rule data table related to statistical metrics. The rule data table includes indicator metric data, dimension mapping data, granularity conversion data, and time anchor data. Indicator metric data defines the statistical object, aggregation method, default filtering metric, and default statistical granularity corresponding to each query indicator. Dimension mapping data defines the correspondence between different expressions in natural language and standard business dimensions. Granularity conversion data defines the paths that allow associations between different data granularities. Time anchor data defines the business time meaning corresponding to different query indicators during statistics. Preset attribution conditions are used to determine whether a query belongs to a query scenario requiring flight segment attribution processing. In this embodiment, the preset attribution conditions can be set as follows: the query indicator corresponds to a request log-level or cost log-level statistical object, and the query dimension includes a flight segment dimension or a route dimension. This condition stems from the actual organization of aviation business data, namely, the need for cross-granularity mapping between log-level records and flight segment-level analysis targets, while general order-level summary queries do not require additional attribution processing.

[0025] In step S100, after receiving the natural language query text, the query text is segmented, part-of-speech identified, and business terminology identified to extract query metrics, query dimensions, query time range, and query conditions. Query metrics refer to the statistical objects the user desires, such as call cost, number of requests, average cost, or composite operating indicators. Query dimensions refer to the grouping method of the statistical results, such as flight segment, route, airport, channel, or date. Query time range refers to the statistical period entered by the user, such as "last month," "last 7 days," or a specific date range. Query conditions refer to additional filtering conditions besides the statistical dimensions and time range, such as "domestic routes," "economy class only," or "refund and change requests only." To avoid omissions due to colloquial expressions, during the extraction process, the same natural language segment is allowed to be mapped to candidate metrics or candidate conditions simultaneously, and the mapping results are sent to the semantic constraint parsing process for unified adjudication, thereby avoiding premature discarding of valid information in the initial extraction stage.

[0026] In step S200, semantic constraint parsing is performed on the natural language query text based on indicator caliber data, dimension mapping data, granularity conversion data, and time anchor data to generate a target semantic frame and caliber locking result. The target semantic frame records the target indicator, target dimension, target time caliber, and target filtering conditions determined after parsing. The caliber locking result records the data source, statistical granularity, allowed association paths, and default filtering rules corresponding to the target semantic frame. In specific implementation, candidate indicators and their corresponding statistical granularities are first determined based on indicator caliber data. Then, dimension terms in natural language are mapped to standard business dimensions based on dimension mapping data. Afterward, the time range expressed by the user is converted into the target time caliber based on time anchor data. For example, "last month's call cost" is locked to the previous month's interval statistically based on the request occurrence time, rather than statistically based on the order creation time. If the same query text can correspond to multiple candidate semantic results, the target semantic frame is selected based on the preset semantic matching score. In one implementation, the semantic matching score threshold can be set to 0.80. The threshold can be obtained through offline verification results of historical annotated query samples. That is, the score that significantly reduces the false matching rate while ensuring a stable correct matching rate is taken as the threshold. If multiple candidate semantic results are all below this threshold, a clarification prompt can be triggered or the default business standard can be adopted to ensure that the determination process of the target semantic frame is reproducible and consistent.

[0027] In step S300, an initial query plan is generated based on the target semantic frame and the caliber locking result. The initial query plan includes at least a target data table, a related data table, filtering conditions, grouping fields, and aggregation fields. When the query metrics and query dimensions meet the preset attribution conditions, the flight segment attribution result corresponding to the record is further determined based on business records and business snapshots, and the initial query plan is adjusted using the flight segment attribution result to obtain the target query plan. Business records include request logs and business action records, and business snapshots include order snapshots, ticket snapshots, and flight segment snapshots. Request logs represent a service request or a system call, business action records represent business actions related to the request such as reservations, ticketing, rescheduling, and refunds, and business snapshots represent the business status corresponding to the request time. Using the request time as the matching benchmark, a snapshot corresponding to the request time and with a valid status is selected from the business snapshots as the request time snapshot. Here, "valid status" means that the snapshot has taken effect at the request time and has not been overwritten by subsequent cancellation records. If the request log already contains explicit flight segment association information, the target flight segment is directly determined accordingly. If not included, the target segment is determined based on the order of the segment corresponding to the business action, the earliest valid segment departing after the request time, and the earliest valid segment. For code-shared or coexisting marketing and operating flight numbers, a standard segment identifier is generated based on the marketing flight number, operating flight number, departure airport, arrival airport, and flight date. Candidate segments are then merged using this standard segment identifier before determining the target segment. Through this method, the grouping key, join key, and filtering path in the initial query plan can be replaced or supplemented with query constraints consistent with the target segment, thus forming the target query plan. If the query metrics and dimensions do not meet the preset attribution conditions, segment attribution processing is unnecessary; the initial query plan is directly used as the target query plan to ensure overall process closure and avoid introducing unnecessary processing overhead to ordinary queries.

[0028] In step S400, a structured query statement is generated based on the target query plan and submitted to the database for execution to obtain statistical results. The structured query statement can be a database-executable query statement, and its generation process is automatically completed based on the data tables, join relationships, filtering conditions, grouping fields, and aggregation fields determined in the target query plan. After obtaining the statistical results, attribution consistency verification is performed on the statistical results. Attribution consistency verification includes at least record coverage verification, consistency verification of the sum of statistical values ​​before and after attribution, and segment validity verification. Record coverage verification is used to determine whether there are any missing records in the request log set participating in attribution before and after attribution. Consistency verification of the sum of statistical values ​​before and after attribution is used to determine whether the sum of the original statistical values ​​before attribution is consistent with the sum of the statistical values ​​after attribution and segment allocation. Segment validity verification is used to determine whether the segment determined as the target segment is a valid segment within the business status window corresponding to the request time. For queries that do not meet the preset attribution conditions and therefore have not undergone segment attribution processing, a basic consistency confirmation method is preferred. The basic consistency confirmation method includes at least confirming the completeness of the target data table, filtering conditions, grouping fields, and aggregation fields in the target query plan. Confirm that the structured query statement can be executed in the target database. Confirm that the target fields in the statistical results are complete and without missing sources. When the above conditions are met, the branch processing is deemed successful. This ensures that this step has clear execution results under different query branches. Regarding threshold settings, in one implementation, the allowable threshold for the difference between the sum of statistical values ​​before and after attribution can be set to a preset minimum statistical unit or an integer multiple thereof. This threshold is derived from the accuracy of financial settlement, the accuracy of log accumulation, or the enterprise's internal statistical rules. For example, when the statistical object is cost data, the preset minimum statistical unit can be set to 0.01 yuan. When the statistical object is request count data, the counts before and after attribution must be strictly consistent, and no difference is allowed. If the attribution consistency check passes, the statistical results are output. If the attribution consistency check fails, the rollback result is output. In a preferred implementation, the rollback result is a log-level statistical result. For queries that do not meet the preset attribution conditions and have not undergone segment attribution processing, the basic consistency confirmation method can be used to confirm the integrity of the query plan and the statistical results before directly outputting the statistical results.

[0029] Specifically, in a particular application scenario, the user's input natural language query text is "Statistics on the cost of intelligent customer service calls for domestic routes last month, displayed by flight segment". The query metric is identified as call cost, the query dimension as flight segment, the query time range as last month, and the query condition as domestic routes. Subsequently, based on the metric data, the cost field in the corresponding service request log table is determined. Based on the dimension mapping data, "domestic routes" is mapped to routes where both departure and arrival airports are within the domestic area. Based on the time anchor data, "last month" is locked as a natural month interval statistically analyzed based on the request occurrence time. Based on the granularity conversion data, it is confirmed that request log-level data can be associated with the flight segment dimension through request logs, business action records, and flight segment snapshots. Then, an initial query plan is generated. Since this query has both cost-related metrics and a flight segment dimension, satisfying the preset attribution conditions, the request time and request identifier are further read from each record in the request log, and the request time snapshot is selected from the associated order snapshot, ticket snapshot, and flight segment snapshot. For records where the flight segment can be directly located, the target flight segment is directly determined. For records that cannot be directly located, they are determined sequentially based on the flight segment corresponding to the business action, the earliest valid flight segment to take off after the request time, and the earliest valid flight segment. If code sharing exists, merging is completed using standard flight segment identifiers. Based on the attribution results, the grouping fields and association paths in the initial query plan are adjusted to generate the target query plan, which in turn generates and executes a structured query statement. Finally, attribution consistency is verified on the statistical results. If the difference in the total cost before and after attribution does not exceed the preset minimum statistical unit, and all attribution flight segments are in a valid state, the call cost results statistically analyzed by flight segment are output. Otherwise, log-level call cost results are output for further user verification.

[0030] Understandably, this embodiment achieves automatic conversion from natural language to database queries. Through a branching processing structure, the same analysis framework can cover both general query scenarios and provide targeted enhancements for complex aviation query scenarios involving cross-granularity and prone to errors. This improves the stability and determinism of statistical results for flight segment and route dimensions without significantly increasing the overhead of general query processing. Especially in situations involving code sharing, frequent changes in business status, and temporal misalignment between log records and business objects, this embodiment, through request-time snapshot matching and standard flight segment identifier merging, can suppress statistical fluctuations caused by state drift and inconsistent dimension mapping. This results in higher consistency of output for the same query under the same basic data conditions, enhancing the reproducibility of statistical results in complex aviation business scenarios.

[0031] In some embodiments of this application, business records include request logs and business action records, and business snapshots include order snapshots, ticket snapshots, and flight segment snapshots. Based on the request time corresponding to the record, a valid business snapshot matching the request time is selected from the business snapshots as the request time snapshot corresponding to the record.

[0032] In some embodiments of this application, determining the flight segment attribution result based on the snapshot at the request time includes: determining the target flight segment according to the flight segment association information in the record. If the target flight segment is not determined, the target flight segment is determined sequentially based on the flight segment corresponding to the business action, the first valid flight segment to take off after the request time, and the earliest valid flight segment.

[0033] Specifically, in this embodiment, business records are used to represent raw event data in aviation business analysis. Request logs include at least a request identifier, request time, order identifier, ticket identifier, flight segment association information, request type, and corresponding cost or count value. Business action records include at least an action identifier, action time, action type, associated order identifier, associated ticket identifier, and the flight segment corresponding to the action. Business snapshots are used to represent the state of a business object at a specific point in time. Order snapshots record order status, ticket snapshots record ticketing, refund, or rebooking status, and flight segment snapshots record flight, airport, date, and validity status. To ensure that attribution processing uses a unified input, a unified association key is established for request logs, business action records, and business snapshots. This unified association key can be a combination of order identifier, ticket identifier, passenger identifier, and flight date, or a preset association index.

[0034] When selecting a snapshot at a specific request time, the preferred method is to select a snapshot whose time falls within the valid timeframe. For each business snapshot, the snapshot's effective time and expiration time are recorded. For a given request log, among the associated order snapshots, ticket snapshots, and flight segment snapshots, snapshots that meet the condition that "the snapshot's effective time is no later than the request time and its expiration time is later than the request time" are selected as candidate valid business snapshots. If multiple candidates exist, the snapshot with the effective time closest to and no later than the request time is selected as the valid business snapshot. Considering the potential time synchronization errors and database storage delays between different business systems, a time tolerance window can be set, such as 300 seconds. The time tolerance window can be determined based on historical statistical results of time synchronization errors, log storage delays, and interface transmission delays. If no corresponding snapshot is found within the time tolerance window, the most recent valid snapshot before the request time is selected first. If it still does not exist, the record is marked as a missing snapshot for subsequent statistics and verification. Thus, the selection of snapshots at the request time has a clear judgment order and stable reproducibility.

[0035] The request-time snapshot consists of order snapshots, ticket snapshots, and flight segment snapshots, used to represent the actual business status at the time the request occurs. A valid business snapshot here does not simply refer to the most recent snapshot in the database, but rather to a snapshot that is effective within the business status window corresponding to the request time and has not been overridden, replaced, or invalidated by subsequent business actions. For example, if a ticket has already been rebooked before the request time, the snapshot of the old flight segment corresponding to the original flight segment is no longer considered a valid business snapshot. If a flight segment has been canceled or replaced before the request time, that flight segment is no longer considered a valid candidate flight segment. For scenarios involving order splitting, ticket rebooking, or connecting flight rebooking, the request-time snapshot is determined primarily according to the principle of "ticket snapshot takes precedence over order snapshot, and consistency between flight segment snapshots and ticket snapshots takes precedence," to reduce the risk of cross-segment misallocation when attributing solely based on order-level status.

[0036] When determining the flight segment attribution result based on the snapshot at the request time, the first step is to determine whether the request log contains flight segment association information that can directly point to a flight segment. Flight segment association information can be a flight segment identifier, a combination of flight number and flight date, a combination of departure and arrival airports, or other fields that can uniquely map to a flight segment. When valid and verifiable flight segment association information exists in the request log, the corresponding flight segment is directly identified as the target flight segment. "Valid and verifiable" here means that the association information can find a unique and valid corresponding flight segment in the snapshot at the request time, and the status of this valid flight segment belongs to a preset set of valid statuses. The set of valid statuses can include reserved, ticketed, rebooked, and pending statuses, but does not include refunded, cancelled, or overridden statuses.

[0037] When no segment association information for directly locating the target segment is found in the request log, the sequential attribution process begins. First, the target segment is determined based on the action type and corresponding segment in the business action record. For example, ticket confirmation, refund / change confirmation, or check-in processing directly point to the corresponding segment. If the business action record still cannot determine the target segment, the earliest valid segment departing after the request time is selected from the set of valid segments corresponding to the snapshot at the request time. If no valid segment departing after the request time exists, the earliest valid segment is further selected as the target segment. Here, "earliest departure after the request time" means the planned departure time is no earlier than the request time and is the earliest planned departure time among the valid segments that meet the conditions. "Earliest valid segment" refers to the segment that is first when sorted in ascending order of planned departure time. If multiple candidate segments are identical in planned departure time, further adjudication can be made according to segment sequence number, ticket sequence number, or snapshot generation time order to ensure a unique attribution result.

[0038] Specifically, in a rebooking request scenario, the request log only records the order identifier, request time, and "rebooking successful" request type, without directly recording the target segment identifier. In this case, a snapshot of the request time is first extracted based on the order identifier and request time. If the order corresponds to two valid segments before the request time, with the first being the original segment and the second being the rebooked segment, and the business action record contains a rebooking confirmation action associated with the request identifier, and the corresponding segment points to the second segment, then the second segment is determined as the target segment, and the cost value corresponding to the request log is attributed to the second segment. If there is no rebooking confirmation action, but both valid segments are in a valid state, then the first valid segment to depart after the request time is selected. If the planned departure times of both valid segments are earlier than the request time, then the earliest valid segment is selected. Thus, even if the request log does not explicitly carry the segment identifier, a unique attribution result can still be obtained based on the evolution of the business state.

[0039] Understandably, by using request time as the attribution anchor and replacing current state data with a snapshot at the time of request for attribution, the reverse contamination of historical request attribution results by subsequent changes in business state can be suppressed. Simultaneously, by setting a hierarchical attribution order of "direct association priority, action correspondence priority, and time sequence priority," a high degree of consistent attribution capability can be maintained even in cases of incomplete log information, frequent changes in business state, or the coexistence of connecting flights and multiple flight segments. The unique advantage of this embodiment lies in enhancing the resilience of the underlying attribution chain to temporal disturbances and state drift, making statistical results more reproducible in scenarios involving repeated queries, delayed recalculation, and historical review.

[0040] In some embodiments of this application, when determining the target segment based on the flight segment corresponding to the business action, the first valid flight segment to take off after the request time, and the earliest valid flight segment, the process includes: generating a standard flight segment identifier based on the marketing flight number, the operating flight number, the departure airport, the arrival airport, and the flight date, and determining the target flight segment after merging the valid flight segments based on the standard flight segment identifier.

[0041] In some embodiments of this application, attribution consistency verification includes record coverage verification, consistency verification of the sum of statistical values ​​before and after attribution, and segment validity verification. If any verification fails, the rollback result is the log-level statistical result.

[0042] Specifically, in this embodiment, when the target segment cannot be uniquely determined directly based on the segment corresponding to the business action, or when multiple valid segments that may correspond to the same transportation fact exist in the snapshot at the request time, a standard segment identifier is first generated based on the marketing flight number, the carrier flight number, the departure airport, the arrival airport, and the flight date. Then, valid segments are merged based on the standard segment identifier to reduce the problem of duplicate candidates caused by code-sharing, joint flights, airline number changes, or differences in records from multiple systems. Here, the marketing flight number is the flight number displayed to passengers, the carrier flight number is the flight number actually performing the flight mission, and the flight date preferably refers to the natural date corresponding to the planned departure. The standard segment identifier is used to uniquely represent the same transportation segment. In one implementation, the marketing flight number and the carrier flight number are first standardized, including removing format differences, unifying the airline code representation, and eliminating the influence of leading zeros. Then, a unified airport code is adopted for the departure and arrival airports. Subsequently, a standard segment identifier is generated according to the method of "carrier priority, marketing supplement, airport pair, and flight date joint constraint". This allows the same actual flight segment, which is recorded as multiple candidates in different business systems, to be converged into a single merged object.

[0043] When merging valid flight segments, a standard flight segment identifier is first generated for each candidate flight segment in the set of valid flight segments corresponding to the snapshot at the request time. Candidate flight segments with the same standard flight segment identifier are then merged into the same merged flight segment set. Auxiliary attributes are retained within each merged flight segment set, including at least the business action correlation strength, planned takeoff time, snapshot effective time, and state stability. After merging, if only one merged flight segment set remains, its corresponding flight segment is directly determined as the target flight segment. If multiple merged flight segment sets still exist, the subsequent sequential decision-making process begins.

[0044] Attribution consistency checks are performed after the statistical results are generated. Record coverage checks determine whether all original records entering the attribution process are reflected in the attribution results. This can be achieved by comparing the set of original record identifiers participating in the statistics before attribution with the set of record identifiers participating in the summary after attribution. The consistency check of the total statistical value before and after attribution determines whether the attribution process introduces lost or duplicate statistical values. This can be achieved by comparing the total statistical value of the original records before attribution with the total statistical value summarized by target flight segment after attribution. When the statistical object is monetary data, the difference is allowed not to exceed the preset minimum statistical unit or an integer multiple thereof. The preset minimum statistical unit can be determined based on the enterprise's settlement accuracy, the minimum currency unit of measurement, and the log accumulation accuracy; for example, 0.01 can be used when retaining two decimal places for monetary amounts. When the statistical object is frequency data, a difference of 0 is preferred. Flight segment validity checks determine whether the target flight segment being attributed belongs to the valid status set within the business status window corresponding to the request time. Only when all the above checks pass will the current statistical result be output. If any validation fails, the statistical results after segment attribution are not directly output. Instead, log-level statistical results are generated as the rollback result. The log-level statistical results retain the original record granularity and are basically summarized according to the user's query conditions, without introducing segment merging and attribution adjustments. A rollback reason flag can also be generated at the same time to indicate the type of validation that failed, so as to facilitate subsequent verification.

[0045] Specifically, in a code-share flight scenario, the snapshot at the time of the request contains two valid flight segment records: one using a marketing flight number and the other using a carrier flight number. Both have the same departure airport, arrival airport, and flight date. The marketing flight number, carrier flight number, departure airport, arrival airport, and flight date are extracted from both records, and after normalization, a standard flight segment identifier is generated. The two records are then merged into the same merged flight segment set. If the business action corresponding to the request does not directly point to a specific flight segment, the merged flight segment set is further evaluated to determine if it meets the subsequent adjudication conditions. After the statistics are completed, the total amount before attribution is compared with the total amount summarized by target flight segment after attribution. If the difference does not exceed 0.01, and the target flight segment after attribution is within the valid status set at the time of the request, the result of the statistics by flight segment is output. If the difference is greater than 0.01, or the target flight segment has an abnormal status, log-level statistics are output. Therefore, even in complex scenarios involving code sharing and the coexistence of multiple systems, it is possible to avoid duplicate attributions and error summaries caused by different flight representation methods.

[0046] Understandably, by introducing a standard segment identifier merging mechanism before determining the target segment and setting up an attribution consistency verification mechanism centered on record coverage, statistical value conservation, and segment legality before outputting results, it is possible to simultaneously constrain the two key aspects of "whether the attribution object is unique" and "whether the attribution result is credible." This embodiment further improves the deduplication stability of the attribution layer and the conservation reliability of the result layer, enabling the same transportation fact to be uniformly identified even when multiple flight representation methods coexist. Furthermore, it automatically reverts to a more conservative data output level when there are minor anomalies in the attribution results, thereby reducing the risk of systematic deviations under complex aviation operation data conditions.

[0047] In some embodiments of this application, when multiple merged valid flight segments exist, the merged valid flight segment that has not been identified as the target flight segment by other records is selected. If no target flight segment is selected, the merged valid flight segment that took off earliest after the requested time is selected. If no target flight segment is still selected, the earliest valid merged valid flight segment is selected. When the difference between the sum of the statistical values ​​before and after attribution exceeds a preset minimum statistical unit, log-level statistical results are output.

[0048] Specifically, in this embodiment, after the standard segment identifier merging is completed, if multiple merged valid segments still exist for the same record to be attributed, the target segment is determined according to a preset order. Here, "merged valid segments" refers to candidate segments formed after merging standard segment identifiers within the snapshot range at the request time, and which are in an attributable state within the business status window corresponding to the request time. "Other records" refers to other records to be attributed that belong to the same statistical batch or the same sliding processing window as the current record to be attributed. To avoid multiple records in the same batch being concentrated in the same candidate segment, a target segment occupancy status table is maintained during the attribution process to record whether each merged valid segment has been determined as the target segment by other records.

[0049] First, among multiple merged valid flight segments, the merged valid flight segment that has not been identified as the target flight segment by other records is selected. "Not identified as the target flight segment by other records" means that within the current statistical batch or preset sliding processing window, the merged valid flight segment has not yet been written into the target flight segment occupancy status table. If multiple merged valid flight segments meet this condition, their association strength with the current record can be further compared, with the higher association strength taking priority. In this embodiment, association strength is determined according to preset grading rules. Preferably, the association strength of an action directly pointing to a flight segment is higher than the association strength of matching ticket identifiers, the association strength of matching ticket identifiers is higher than the association strength of matching order identifiers, and the association strength of matching order identifiers is higher than the association strength of only matching passenger identifier and flight date. When the association strength levels are the same, subsequent timing decisions are made. If no target flight segment is still selected, the merged valid flight segment that departed earliest after the requested time is selected. If it still does not exist, the earliest valid merged valid flight segment is selected. Here, "earliest departure after the requested time" means that the planned departure time is no earlier than the requested time and is the earliest among the candidates that meet the conditions. "Earliest valid" preferentially refers to the earliest planned departure time. If the planned departure time is unavailable, the one with the earlier snapshot effective time takes priority. Through the above order of decision-making, a unique target segment can be obtained when multiple merged valid segments coexist, avoiding situations where the same record corresponds to multiple target segments or different operation batches produce different attribution results.

[0050] After determining the target flight segment and generating attribution results, the difference between the sum of statistical values ​​before and after attribution is calculated, and this is used to determine whether to output log-level statistical results. The sum of statistical values ​​before attribution is the cumulative result of the statistical values ​​of the original records participating in this attribution process before attribution. The sum of statistical values ​​after attribution is the cumulative result of the statistical values ​​obtained by summarizing the target flight segments after the target flight segment attribution of the same batch of original records is completed. Statistical values ​​can be cost values, call counts, request counts, or other accumulative business values. When the statistical object is monetary data, the difference between the sum of statistical values ​​before and after attribution is preferably calculated using absolute values ​​and compared with a preset minimum statistical unit. When the statistical object is frequency data, a difference of 0 is preferred. The preset minimum statistical unit can be determined based on the accuracy of enterprise financial settlement, log storage accuracy, rounding rules, and the accuracy of database numerical fields. For example, when the statistical object is a monetary amount stored with two decimal places, the preset minimum statistical unit can be 0.01. When the statistical object is a monetary amount stored in cents, the preset minimum statistical unit can be one minimum unit of measurement. If the difference between the sum of the statistical values ​​before and after attribution exceeds the preset minimum statistical unit, it is determined that the consistency check of the sum of the statistical values ​​before and after attribution has failed, and thus it is determined that there is an unacceptable cumulative deviation in the current attribution processing result. In this case, the statistical results after attribution are not output, but the log-level statistical results are output.

[0051] Specifically, in a scenario involving inter-trip request statistics, the same request log corresponds to two merged valid flight segments in the snapshot at the time of the request. The first merged valid flight segment has already been identified as the target segment by another request log in the same batch, while the second merged valid flight segment is not yet occupied. In this case, the target segment occupancy status table is queried first. It is found that the second merged valid flight segment meets the condition that it has not been identified as the target segment by other records, so it is directly selected as the target segment without proceeding to subsequent timing decisions. If both merged valid flight segments are already occupied by other records, their planned departure times are compared, and the merged valid flight segment that departs earliest after the request time is selected. If both planned departure times are earlier than the request time, the merged valid flight segment with the earliest planned departure time is selected. After all records are attributed, the total amount of the original amount before attribution is 1280.36, and the total amount after attribution by target segment is 1280.38. When the preset minimum statistical unit is 0.01, since the difference of 0.02 exceeds the preset minimum statistical unit, the segment statistical results after attribution are not output, but the log-level statistical results are output as the rollback result.

[0052] Understandably, by prioritizing the exclusion of candidate segments already occupied by other records when multiple merged valid segments coexist, not only can the unique determination of the target segment be achieved, but the phenomenon of repeated absorption of locally popular segments during statistical processing can also be suppressed. Simultaneously, by comparing the sum of statistical values ​​before and after attribution with a preset minimum statistical unit, and directly downgrading the output to log-level statistical results when the limit is exceeded, the spread of abnormal results can be promptly blocked at an early stage when subtle cumulative deviations occur. This embodiment further enhances the balance of attribution distribution and the sensitivity to statistical value conservation in multi-candidate segment scenarios, making the results in complex connecting flights, multi-segment flights, and dense request scenarios more stable and auditable.

[0053] In some embodiments of this application, semantic constraint parsing includes: extracting indicator words, dimension words, time words, and condition words from natural language query text; determining candidate indicators and corresponding statistical granularities based on indicator caliber data; standardizing and mapping dimension words based on dimension mapping data; determining target time caliber based on time anchor data; and generating candidate semantic results.

[0054] In some embodiments of this application, the statistical granularity and query dimension in the candidate semantic results are filtered for relevance based on granularity conversion data, eliminating non-convertible combinations, and the target semantic frame and caliber locking result are generated based on the filtered candidate indicators, standardized dimensions and target time caliber.

[0055] Specifically, in this embodiment, semantic constraint parsing is performed after receiving the natural language query text. To ensure the parsing results are reproducible and can directly serve the generation of subsequent query plans, indicator caliber data, dimension mapping data, granularity conversion data, and time anchor data are pre-maintained. Indicator caliber data records the standard name, statistical object, aggregation method, default statistical granularity, allowed statistical dimensions, and default filtering rules for each indicator. Dimension mapping data records the standard name, synonyms, hierarchical relationships, and encoding mapping relationships of business dimensions. Granularity conversion data records allowed association paths, prohibited association combinations, and intermediate bridging objects between different statistical granularities. Time anchor data records the business time meaning corresponding to different indicators, such as request occurrence time, order creation time, ticketing time, flight date, departure time, and settlement time. The above data is preferably stored in the form of a rule table or configuration library and can be managed by version number to ensure that the same query yields consistent parsing results under the same rule version.

[0056] First, extract indicator terms, dimension terms, time terms, and condition terms from the natural language query text. This extraction process can be achieved using a combination of business dictionary matching, syntactic segmentation, and entity recognition. Indicator terms are words that characterize the user's statistical goals; dimension terms characterize the objects being analyzed in groups; time terms characterize the statistical time range or benchmark; and condition terms characterize the filtering range. When the same term may correspond to multiple business meanings, a decision is not made immediately; instead, multiple candidate interpretations are retained for subsequent constraint filtering. Next, candidate indicators and their corresponding statistical granularities are determined based on indicator caliber data. Statistical granularity refers to the natural statistical level of an indicator in the underlying data, such as log level, order level, ticket level, flight segment level, or route level. The extracted indicator terms are matched with the standard names and synonyms in the indicator caliber data to obtain at least one candidate indicator, and the corresponding default statistical granularity and allowed statistical dimensions are read. For cases with multiple candidate indicators, term matching scores and context consistency scores can be calculated to retain candidate indicators whose scores reach a preset matching threshold. In one implementation, the preset matching threshold can be set to 0.80, which can be determined based on the offline verification results of historical labeled query samples.

[0057] Next, the dimension terms are standardized based on the dimension mapping data, and the target time caliber is determined based on the time anchor data. Standardization mapping refers to converting non-standard business expressions in natural language into internally unified standard dimensions. For time terms, not only the time range is determined, but also the corresponding business time field is determined. Here, the "target time caliber" refers to the actual time benchmark used in the statistics. For example, for "last month's cost," the time benchmark for call cost can be locked as the request occurrence time benchmark, while for ticket amount, it can be locked as the ticket issuance time benchmark. If the user explicitly provides a time benchmark, it is adopted first. If the user does not explicitly provide one, it is determined according to the default time rule of the candidate indicator in the time anchor data. After the above processing, the candidate indicator, standardized dimension, target time caliber, and conditional terms are combined to generate at least one candidate semantic result.

[0058] Furthermore, based on the granularity transformation data, the statistical granularity and query dimension in the candidate semantic results are screened for relevance to eliminate non-relevance combinations. The core of relevance screening is to determine whether the statistical granularity corresponding to the candidate indicator can be stably mapped to the query dimension under preset business rules. In specific implementation, the allowed conversion paths from the statistical granularity of the candidate indicator to the query dimension can be queried in the granularity transformation data. When a valid path exists, the candidate semantic result is retained. When no valid path exists, a prohibited association marker exists, or multiple conflicting paths exist and cannot be converged into a unique association path by preset priority rules, the candidate semantic result is eliminated. Among them, the prohibited association marker is preferably pre-configured in the granularity transformation data. Conflicting paths refer to paths where the target statistical object sets corresponding to different paths are inconsistent, or paths that will lead to an increase in the risk of duplicate accumulation under the same query conditions. Preferably, the preset priority rules are determined in the order of direct association paths taking precedence over bridging association paths, and low duplicate accumulation risk paths taking precedence over high duplicate accumulation risk paths. After the screening is completed, the target semantic frame and caliber locking result are generated based on the screened candidate indicators, standardized dimensions, and target time caliber. The target semantic frame includes at least the target metric, target dimension, target time caliber, and target filtering conditions. The caliber locking result includes at least the target data source, statistical granularity, allowed association paths, default filtering rules, and time field selection results. If multiple candidate semantic results remain after filtering, they can be sorted according to rule priority, matching score, or historical usage frequency, and the one with the highest priority can be selected as the target semantic frame. If the score difference between candidates is lower than a preset ambiguity threshold, a clarification prompt can be triggered. In one implementation, the ambiguity threshold can be set to 0.10, and its source can be the statistical results of the score distribution differences between correct and incorrect intents in historical query statements.

[0059] Specifically, when a user inputs "Statistics on the cost of displaying intelligent customer service calls for domestic routes by segment last month," the system first extracts "call cost" as the metric term, "segment" as the dimension term, "last month" as the time term, and "domestic routes" as the condition term. Then, based on the metric caliber data, the system determines the cost field in the service request log corresponding to "call cost," with a default statistical granularity of log level. Based on the dimension mapping data, "domestic routes" is mapped to the condition that both the departure and arrival airports are located within China, and "display by segment" is mapped to the segment dimension. Based on the time anchor data, "last month" is locked to the natural month interval corresponding to the request occurrence time. Next, based on the granularity conversion data, the system determines whether the log-level metric is allowed to be associated with the segment dimension. If a valid association path exists from the request log through business records to the segment snapshot, the candidate semantic result is retained, and a target semantic frame and caliber locking result are generated. If no valid path exists, the corresponding candidate semantic result is eliminated, thus preventing it from entering the subsequent query plan generation stage.

[0060] Understandably, by introducing pre-constraint links before query plan generation, candidate interpretations that can form query statements but cannot form stable statistical calibers can be eliminated at the semantic level, thereby reducing the hidden error of "executed statements but unreliable results". This embodiment moves error control forward to the semantic parsing stage, completing the convergence processing of indicator meaning, time meaning, and granular path before generating structured query statements. Therefore, it can suppress systematic biases caused by synonymous expressions, time base confusion, and cross-granularity misassociations, making it easier to obtain unique, stable, and verifiable analytical calibers in complex aviation business scenarios.

[0061] Based on another preferred embodiment described above, see [link to preferred embodiment]. Figure 2 As shown, this embodiment provides a natural language-based intelligent analysis system for aviation business data, used to apply the above-mentioned natural language-based intelligent analysis method for aviation business data, including: The data collection unit is configured to receive natural language query text and extract query metrics, query dimensions, query time range, and query conditions.

[0062] The processing unit is configured to perform semantic constraint parsing on natural language query text based on preset indicator data, dimension mapping data, granularity transformation data, and time anchor data, and generate target semantic frames and caliber locking results.

[0063] The execution unit is configured to generate an initial query plan based on the target semantic frame and the caliber locking result. When the query metrics and query dimensions meet the preset attribution conditions, it determines the flight segment attribution result corresponding to the record based on the business record and business snapshot, and adjusts the initial query plan according to the flight segment attribution result to obtain the target query plan. When the query metrics and query dimensions do not meet the preset attribution conditions, the initial query plan is used as the target query plan.

[0064] The output unit is configured to generate a structured query statement based on the target query plan, execute the structured query statement to obtain statistical results, and perform attribution consistency verification on the statistical results. It outputs the statistical results if the attribution consistency verification passes, and outputs a rollback result if the attribution consistency verification fails.

[0065] On the other hand, this application also provides a computer-readable storage medium in which the electronic device can perform the above-described method when the computer-executable instructions stored in the computer-readable storage medium are executed by the processor of the electronic device.

[0066] It is understandable that the above-mentioned intelligent analysis methods, systems and storage media for aviation business data based on natural language have the same beneficial effects, and will not be elaborated further here.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A natural language-based intelligent analysis method for aviation business data, characterized in that, include: Receive natural language query text and extract query metrics, query dimensions, query time range, and query conditions; Based on preset indicator data, dimension mapping data, granularity conversion data, and time anchor data, the natural language query text is subjected to semantic constraint parsing to generate target semantic frames and caliber locking results. An initial query plan is generated based on the target semantic frame and the caliber locking result. When the query indicators and the query dimensions meet the preset attribution conditions, the flight segment attribution result corresponding to the record is determined based on the business record and business snapshot. The initial query plan is then adjusted based on the flight segment attribution result to obtain the target query plan. When the query metrics and the query dimensions do not meet the preset attribution conditions, the initial query plan will be used as the target query plan. A structured query statement is generated based on the target query plan. The structured query statement is executed to obtain statistical results, and the statistical results are subjected to attribution consistency verification. The statistical results are output when the attribution consistency verification passes, and the rollback results are output when the attribution consistency verification fails. 2.The natural language based aviation business data intelligent analysis method according to claim 1, characterized in that, The business records include request logs and business action records, and the business snapshots include order snapshots, ticket snapshots, and flight segment snapshots. Based on the request time corresponding to the record, a valid business snapshot matching the request time is selected from the business snapshots and used as the request time snapshot corresponding to the record. 3.The natural language based aviation business data intelligent analysis method according to claim 2, characterized in that, When determining the attribution result of the flight segment based on the requested snapshot, the following is included: The target segment is determined based on the segment association information in the record; if the target segment is not determined, it is determined sequentially based on the segment corresponding to the business action, the first valid segment to take off after the request time, and the earliest valid segment. 4.The natural language based aviation business data intelligent analysis method according to claim 3, characterized in that, When determining the target segment based on the flight segment corresponding to the business action, the earliest valid flight segment to take off after the request time, and the earliest valid flight segment, the following applies: A standard segment identifier is generated based on the marketing flight number, the operating flight number, the departure airport, the arrival airport, and the flight date. The target segment is then determined by merging the valid segments based on the standard segment identifier. 5.The natural language based aviation business data intelligent analysis method according to claim 3, characterized in that, The attribution consistency check includes record coverage check, consistency check of the sum of statistical values ​​before and after attribution, and flight segment validity check; if any check fails, the rollback result is a log-level statistical result. 6.The natural language based aviation business data intelligent analysis method according to claim 4, characterized in that, When multiple merged valid segments exist, select the merged valid segment that has not been identified as the target segment by other records; when no target segment is selected, select the merged valid segment that took off first after the request time; when no target segment is selected, select the earliest valid merged valid segment; when the sum of the statistical values ​​before and after attribution exceeds the preset minimum statistical unit, output log-level statistical results. 7.The natural language based aviation business data intelligent analysis method according to claim 1, wherein, The semantic constraint parsing includes: extracting indicator words, dimension words, time words, and condition words from the natural language query text; determining candidate indicators and corresponding statistical granularities based on the indicator caliber data; standardizing the dimension words based on the dimension mapping data; determining the target time caliber based on the time anchor point data; and generating candidate semantic results. 8.The natural language based aviation business data intelligent analysis method according to claim 7, characterized in that, Based on the granularity transformation data, the statistical granularity and query dimension in the candidate semantic results are filtered for relevance, eliminating non-transformable combinations, and the target semantic frame and the caliber locking result are generated according to the filtered candidate indicators, standardized dimensions and target time caliber.

9. A natural language based intelligent analysis system for aviation business data, for applying the natural language based intelligent analysis method for aviation business data according to any one of claims 1-8, characterized in that, include: The data collection unit is configured to receive natural language query text and extract query metrics, query dimensions, query time range, and query conditions. The processing unit is configured to perform semantic constraint parsing on the natural language query text based on preset indicator caliber data, dimension mapping data, granularity conversion data, and time anchor data, and generate target semantic frames and caliber locking results. The execution unit is configured to generate an initial query plan based on the target semantic frame and the caliber locking result; when the query indicators and the query dimensions meet the preset attribution conditions, determine the flight segment attribution result corresponding to the record based on the business record and business snapshot; and adjust the initial query plan according to the flight segment attribution result to obtain the target query plan. When the query metrics and the query dimensions do not meet the preset attribution conditions, the initial query plan will be used as the target query plan. The output unit is configured to generate a structured query statement according to the target query plan, execute the structured query statement to obtain statistical results, and perform attribution consistency verification on the statistical results; output the statistical results when the attribution consistency verification passes, and output the rollback results when the attribution consistency verification fails.

10. A computer-readable storage medium, characterized in that, When the computer-executable instructions stored in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is capable of performing the method as described in any one of claims 1 to 8.

Images