A service request system and service request method for low-altitude flight services

Abstract

The application provides a business request system and method for low-altitude flight services. The system includes a data input module, a rule engine module, a large model intent recognition module, an explainability tracking module, and a request routing module. The data input module receives business request text. The rule engine module performs rule matching to determine whether a preset business type is hit. If so, a first recognition result is output. If not, the large model intent recognition module calls a large language model for intent recognition to output a second recognition result. The explainability tracking module generates first audit link information based on the first recognition result or second audit link information based on the second recognition result. The request routing module generates target business intent information based on the recognition result and the audit link information and sends it to a downstream business system. The application can meet the regulatory review needs of the low-altitude economy field.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a business request system and method for low-altitude flight services. Background Technology

[0002] In the fields of artificial intelligence and large-scale intelligent agents, intent recognition is a key step in understanding user input and routing it to the corresponding business modules.

[0003] With the development of large language model technology, prompt-based intent recognition methods have gradually become one of the mainstream solutions. This approach concatenates the user's original query with predefined prompts, for example: "Please identify the intent of the following user questions, with candidate categories including: route query, flight activity declaration, airspace application..." The concatenated text is then fed into a large language model for inference, and the model directly outputs the intent category label. The entire process relies on the large model's ability to understand natural language and follow instructions, eliminating the need to train a dedicated classification model. Therefore, it has certain advantages when dealing with small samples or flexibly changing intent categories.

[0004] However, the low-altitude economy is a heavily regulated sector. Businesses such as airspace approval and flight application have strict compliance requirements for decision-making transparency and full-process auditability. The aforementioned pure big-model reasoning is a black-box decision-making logic. Once an intention identification misjudgment occurs, it is impossible to fully trace the basis for the judgment, making it difficult to meet the regulatory review needs of the low-altitude economy sector. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a business request system and a business request method for low-altitude flight services, which can meet the regulatory review requirements of the low-altitude economy.

[0006] According to one aspect of the present invention, a service request system for low-altitude flight services is provided, the system comprising a data input module, a rule engine module, a large model intent recognition module, an interpretability tracking module, and a request routing module; The data input module is used to receive business request text; The rule engine module is used to perform rule matching based on the business request text to determine whether a preset business type is matched; if matched, the corresponding first recognition result is output and the first recognition result is sent to the interpretability tracking module; if not matched, the business request text is sent to the large model intent recognition module. The large model intent recognition module is used to call the large language model to perform intent recognition on the business request text when the rule engine module fails to match, output the corresponding second recognition result, and send the second recognition result to the interpretability tracking module. The interpretability tracking module is configured to generate first audit link information based on the first identification result, and send the first identification result and the first audit link information together to the request routing module; or, generate second audit link information based on the second identification result, and send the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; The request routing module is used to generate target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and send the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system provides corresponding low-altitude flight services based on the business intent information.

[0007] According to another aspect of the present invention, a service request method for low-altitude flight services is provided. The method is applied to the aforementioned service request system, which includes a data input module, a rule engine module, a large model intent recognition module, an interpretability tracing module, and a request routing module. The method includes: The data input module receives the business request text. The rule engine module performs rule matching based on the business request text to determine whether a preset business type is matched. If matched, the corresponding first recognition result is output and sent to the interpretability tracking module. If not matched, the business request text is sent to the large model intent recognition module. When the rule engine module fails to match the large model intent recognition module, the large language model is invoked to perform intent recognition on the business request text, outputting the corresponding second recognition result, and sending the second recognition result to the interpretability tracking module. The interpretability tracking module generates first audit link information based on the first identification result and sends the first identification result and the first audit link information together to the request routing module; or, it generates second audit link information based on the second identification result and sends the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; The request routing module generates target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and sends the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system provides corresponding low-altitude flight services based on the business intent information.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device including the above-described service request system.

[0009] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the above-described business request method.

[0010] This invention provides a business request system for low-altitude flight services. The system includes a data input module, a rule engine module, a large-scale model intent recognition module, an interpretability tracking module, and a request routing module. The rule engine module is positioned as the primary decision-making channel, and the large-scale model intent recognition module as the secondary decision-making channel. The two are automatically routed based on rule matching. A hierarchical collaborative system of "rule priority, model fallback" is constructed, allowing deterministic intents to completely bypass model reasoning. This architecture simultaneously ensures accuracy and response speed, fundamentally eliminating probabilistic uncertainty for standardized expressions and meeting the rigid reliability requirements of serious scenarios such as low-altitude economic flight applications.

[0011] Furthermore, the business request system designs the interpretability tracing module as an independent and integrated functional module rather than a post-event log system. It generates audit link information uniformly across both rule engine hits and large model inference paths, filling the gap in decision-making transparency in existing solutions. This design ensures that the complete evidence for each intent identification is transparent and verifiable, providing reliable technical support for problem localization, effect auditing, and compliance review. Attached Figure Description

[0012] Further details, features, and advantages of the invention are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of a service request system for low-altitude flight services provided according to an exemplary embodiment of the present invention is shown. Figure 2 This diagram illustrates an automatic flow distribution system for a hierarchical collaborative system according to an exemplary embodiment of the present invention. Figure 3 A flowchart of a service request method for low-altitude flight services according to an exemplary embodiment of the present invention is shown. Figure 4 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0013] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0014] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0015] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0016] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0017] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0018] This invention provides a service request system for low-altitude flight services, such as... Figure 1 The diagram shown illustrates a business request system that includes a data input module, a rules engine module, a large model intent recognition module, an interpretability tracing module, and a request routing module.

[0019] The implementation principles of each module will be introduced below.

[0020] The data input module is used to receive business request text. The rules engine module is used to perform rule matching based on the business request text to determine whether a preset business type is matched. If matched, the corresponding first recognition result is output and sent to the interpretability tracking module. If not matched, the business request text is sent to the large model intent recognition module. The large model intent recognition module is used to call the large language model to perform intent recognition on the business request text when the rule engine module fails to match, output the corresponding second recognition result, and send the second recognition result to the interpretability tracing module. The interpretability tracking module is used to generate first audit link information based on the first identification result, and send the first identification result and the first audit link information together to the request routing module; or, generate second audit link information based on the second identification result, and send the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; The request routing module is used to generate target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and send the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system can provide corresponding low-altitude flight services based on the business intent information.

[0021] In one possible implementation, the data input module serves as the system's unified entry point, responsible for receiving and preprocessing various user-input business request texts. This module supports multimodal input formats, including direct text input, speech-to-text transcription, and structured text converted from user interface selections. Internally, the module includes a basic validity verification layer that performs length checks, encoding format standardization, sensitive character filtering, and empty input interception on the user-input business request text, ensuring that data entering subsequent processes meets basic quality requirements.

[0022] The rules engine module is responsible for accurately identifying deterministic intents. It performs a quantitative evaluation of the user-input business request text against stored rules, outputting a matching degree value for each preset business type. The matching degree value is then compared to a preset threshold, and a decision-making process is initiated. If a preset business type is matched, it indicates that the current business request text is a standardized, well-expressed, and clearly defined business request, which can be accurately and unambiguously identified using fixed rules without requiring semantic reasoning from a large model. In this case, the corresponding first identification result is output and sent to the interpretability tracking module. If a match is not found, it indicates that the current business request text is an unconventional, open-ended, complex, and ambiguous business request, which cannot be accurately identified using fixed rules. In this case, the threshold determiner no longer generates a rule identification result and directly forwards the original business request text to the large model intent recognition module, where the large model relies on semantic understanding and contextual reasoning capabilities to perform deep recognition of complex intents.

[0023] The large-scale model intent recognition module is the system's generalization understanding channel, responsible for handling complex, ambiguous, or non-standardized inputs that the rule engine fails to reliably match. Based on the received business request text, the large-scale model intent recognition module generates prompt words, calls the large language model to perform semantic reasoning on these prompt words, and the large language model (also known as the large model) relies on massive semantic knowledge and contextual understanding capabilities to parse the business request text, identify the user's business intent, output the corresponding second recognition result, and send the second recognition result to the interpretability tracking module.

[0024] The interpretability tracking module, as an independent functional component, runs throughout the entire intent recognition decision-making process and is a key design element for achieving end-to-end transparency and auditability. This module collects specific decision-making process information for each path the business request text traverses. Then, it integrates all collected decision-making process information with the identifiers corresponding to the business request text to form standardized audit link information, resulting in first audit link information for the rule engine module and second audit link information for the large-scale model intent recognition module. Subsequently, the first audit link information and its corresponding first recognition result are sent together to the request routing module, or the second audit link information and its corresponding second recognition result are sent together to the request routing module.

[0025] The request routing module is responsible for encapsulating the identification results and audit link information in a structured form as target business intent information, and routing it as a business request to the corresponding low-altitude flight business system, so that the downstream low-altitude flight business system can adapt to subsequent low-altitude flight services and the operation and maintenance personnel can perform quality monitoring.

[0026] The business request system provided in this embodiment of the invention includes a data input module, a rule engine module, a large model intent recognition module, an interpretability tracing module, and a request routing module. The rule engine module is positioned as the primary decision-making channel, and the large model intent recognition module is positioned as the secondary decision-making channel. The two are automatically routed based on the determination of whether a rule is matched. A hierarchical collaborative system of "rule priority, model fallback" is constructed, which allows deterministic intents to completely bypass model reasoning. From the architectural level, it simultaneously ensures accuracy and response speed, fundamentally eliminating the probabilistic uncertainty for standardized expressions, and meeting the rigid requirements for absolute reliability in serious scenarios such as low-altitude economic flight declarations.

[0027] Furthermore, this invention designs the interpretability tracking module as an independent and integrated functional module, rather than a post-event log system. It generates audit link information uniformly across both rule engine hits and large model inference paths, filling the gap in decision-making transparency in existing solutions. This design ensures that the complete evidence for each intent identification is transparent and verifiable, providing reliable technical support for problem localization, effect auditing, and compliance review. This capability is particularly crucial in regulated sectors such as the low-altitude economy.

[0028] The following will refer to Figure 2 The diagram shown illustrates the automatic routing of the hierarchical collaborative system, introducing the execution process of the two paths: rule engine hit and large model inference.

[0029] First, we introduce the execution process of the rule engine's hit path. This path uses a multi-factor weighted fusion strategy to determine the degree of matching between the business request text and each business intent matching rule, in order to improve the accuracy of rule matching.

[0030] In this path, the rules engine module is used for: Iterate through the preset trigger condition text corresponding to each preset business type, and determine the matching degree value corresponding to each preset business type based on the text similarity, keyword hit rate and semantic vector similarity between the business request text and the current preset trigger condition text. Among them, text similarity is used to identify business request texts containing standardized expressions; keyword hit rate is used to represent different business types through multiple keywords with different weights; semantic vector similarity is used to identify business request texts that are semantically identical to the standardized expressions and have sentence rewriting or entity synonym substitution. If there is a preset business type with a matching degree value greater than a preset threshold, then obtain the first business type corresponding to the maximum matching degree value and output the corresponding first identification result. The first identification result includes the business identifier of the first business type. If there is no preset business type with a matching degree value greater than the preset threshold, it is determined that the match is not found, and the business request text is sent to the large model intent recognition module.

[0031] In one possible implementation, the rule engine module can consist of three sub-components: a deterministic rule base, a similarity calculation, and a threshold determiner.

[0032] The deterministic rule base component stores predefined business intent matching rules. Each rule consists of a trigger condition and a corresponding business type identifier. Trigger conditions support multiple expression formats, including regular expressions, keyword combinations, entity constraints, and structured templates, accurately describing the textual features of high-frequency, standardized business intents. The rule base supports hot updates; operations and maintenance personnel can add, delete, and modify rules through the management interface without restarting the service or retraining the model, ensuring precise and controllable impact of changes.

[0033] The similarity calculation component is responsible for performing a quantitative evaluation of the matching between the user-input business request text and each rule in the rule base, and outputting the matching degree value corresponding to each preset business type.

[0034] Specifically, text similarity can be used to identify business request texts that contain standardized expressions.

[0035] After normalizing the user-inputted business request text, it is sequentially matched with the preset trigger condition text of each rule in the rule base. Text similarity is calculated using a proportional similarity formula based on edit distance. This score reflects the closeness between the user input and the rule template at the character level, demonstrating a high ability to identify standardized, deterministic expressions. The specific formula is as follows:

[0036] in, Indicates text similarity. This is the normalized business request text. The text is a preset trigger condition. This indicates the edit distance between the business request text and the preset trigger condition text. The string length representing the business request text. This indicates the string length of the preset trigger condition text. This indicates the maximum string length between the business request text and the preset trigger condition text.

[0037] Keyword hit rate can be used to represent different business types using multiple keywords with different weights.

[0038] A set of core keywords and corresponding weights can be predefined for each rule. After segmenting the user-input business request text, the hit rate of each keyword is calculated. Keyword weights are pre-set based on their distinguishability to the corresponding business intent, with high-distinguishing keywords assigned high weights and general keywords assigned low weights. This score effectively captures the key semantic signals of intent recognition, compensating for the lack of semantic focus in pure literal matching. The specific formula is as follows:

[0039] in, Indicates keyword hit rate. n To preset the number of keywords corresponding to the business type, For the first corresponding to the preset business type i One keyword, For the first i Preset weights for each keyword; For the hit indicator function, when the keyword The value is 1 if it appears in the business request text, and 0 otherwise.

[0040] Semantic vector similarity is used to identify business request texts that are semantically identical to the normalized expression, or that contain sentence rewriting or entity synonym substitution.

[0041] To overcome the limitations of text similarity and keyword hit rate in covering synonyms and sentence variations, a lightweight semantic vector similarity calculation can be introduced. The business request text and the preset trigger condition text are mapped to fixed-dimensional semantic vectors using a pre-trained sentence vector model, and cosine similarity is used to calculate the distance between them in the semantic space.

[0042] This score measures the semantic relationship between user input and rule templates. Figure 1 Consistency allows the rule engine module to have stronger generalization matching capabilities for sentence variations such as "check the flight route" and "what's the weather like today". Unlike the high latency of large language model inference, the sentence vector model used here is a lightweight encoder, with inference time within milliseconds, which does not affect the overall real-time performance of the rule engine module.

[0043] The specific formula is as follows:

[0044] in, Represents semantic vector similarity. The sentence vector representation of the business request text. This is a sentence vector representation of the text that triggers the preset conditions.

[0045] By employing a multi-factor weighted fusion mechanism, the similarity scores from the three dimensions mentioned above are weighted and fused to obtain a matching degree value:

[0046] in, This represents the degree of matching, and its value can range from (0,1). , , Let be the weight coefficients of each factor, satisfying .

[0047] The weighting coefficient can be flexibly configured according to the business scenario. For example, in scenarios with extremely high requirements for standardized expression (such as standardized flight declarations), the weighting coefficient can be increased. To enhance the accuracy of literal matching; in scenarios where user expression is highly diverse, it can increase... To improve semantic generalization ability.

[0048] Furthermore, the threshold determiner can determine the threshold value based on a preset threshold. Matching degree value Make a judgment and complete the decision-making process. If This indicates that the rules engine module has successfully identified the user's business intent and directly outputs the corresponding business identifier as the first identification result, completely bypassing the large model inference process. This indicates that the rules engine module cannot reliably match the business request text entered by the current user. The system automatically routes it to the large model intent recognition module, and continues to make decisions through the large model inference path.

[0049] The decision-making process of the aforementioned rule engine module breaks through the traditional rule engine's binary matching mode that relies solely on regular expressions or keywords. It proposes a multi-factor weighted fusion similarity measurement and evaluation method. While maintaining millisecond-level response speed, it endows the rule engine with fault-tolerant matching capabilities for synonyms and sentence variations, significantly improving the recall rate and robustness of the rule engine. Simultaneously, configurable weight coefficients allow the same engine to adapt to the precision and generalization balance requirements of different business scenarios.

[0050] Furthermore, within the rule engine's hit path, the first identification result can be transmitted to the interpretability tracking module. Accordingly, the interpretability tracking module is used for: Upon receiving the first recognition result sent by the rule engine module, the first decision channel identifier, trigger condition information, and similarity information corresponding to the first recognition result are obtained and recorded as the first tracking information. The trigger condition information includes the rule identifier and condition content corresponding to the preset trigger condition text, and the similarity information includes text similarity, keyword hit rate, semantic vector similarity, and maximum matching degree value. Generate a decision path summary of the first tracking information as the first audit link information.

[0051] In one possible implementation, when the interpretability tracking module receives the first identification result sent by the rule engine module, it can mark the current decision channel as the first decision channel identifier, and obtain the aforementioned trigger condition information and similarity information from the rule engine module. It can automatically record complete information such as rule tag, condition content, similarity information, and priority. It can also record the final routing decision and end-to-end processing latency of the path hit by the rule engine, and store them in the interpretability tracking module as the first tracking information, providing reliable technical support for problem localization, effect auditing, and compliance review.

[0052] To improve transmission efficiency, the aforementioned first tracking information can be aggregated into a structured decision path summary based on the identifier corresponding to the business request text. This decision path summary may include the first decision channel identifier and the key decision basis for the rule engine's hit path. This decision path summary, as the first audit link information, is transmitted to the request routing module along with the business identifier of the identified first business type.

[0053] Based on this, the request routing module is used for: Obtain the first audit link information, and the business identifier of the first business type carried by the first identification result; Configure the first confidence level corresponding to the first identification result to a preset certain value; Based on the first audit link information, the business identifier of the first business type, and the first confidence level, generate the first business intent information corresponding to the business request text; The first business intent information is sent to the low-altitude flight business system corresponding to the first business type.

[0054] In one possible implementation, after receiving the first audit link information and the first identification result sent by the interpretability tracing module, the request routing module can parse and obtain the service identifier of the first service type carried in the first identification result, and use this service identifier as the user intent code for the final determination by the system. The first confidence level corresponding to the first identification result is configured to a preset value, such as "1", reflecting that the reliability of this identification result is the highest.

[0055] The first audit link information, the business identifier of the first business type, and the first confidence level are used as three core fields. These are encapsulated in a structured form as the first business intent information corresponding to the business request text, and sent to the low-altitude flight business system corresponding to the first business type as the final intent recognition result.

[0056] This concludes the introduction to the execution process of the rule engine's hit path. The following section will introduce the reasoning path of the large model.

[0057] In this path, the large model intent recognition module is used for: Call the large language model to determine the business scenario of the business request text and evaluate at least one estimated business scenario corresponding to the business request text. Obtain the target prompt word template corresponding to the estimated business scenario, combine the target prompt word template with the business request text, and input it into the large language model for intent recognition. Receive the second recognition result returned by the large language model, which includes the response text, the business identifier of the second business type, and the model confidence score.

[0058] In one possible implementation, the large model intent recognition module can consist of three sub-components: a prompt word template library, a model invocation agent, and a confidence evaluator.

[0059] The prompt word template library maintains a modular set of prompt word templates, with each business scenario corresponding to an independent template configuration. This means it stores prompt word templates for different business scenarios. Unlike existing solutions that use a single, massive prompt word, this embodiment of the invention adopts a scenario-based splitting strategy. Each prompt word template covers only one set of related business intents, avoiding interference between intents and the maintenance difficulties caused by prompt word overload. The prompt word templates support independent debugging and version management, allowing operations and maintenance personnel to optimize prompt words for specific scenarios without affecting the recognition performance in other scenarios.

[0060] The model invocation proxy is responsible for standardized interaction with the large language model service, encapsulating low-level operations such as request construction, model invocation, and result parsing. When the rule engine module misses a match, the large model intent recognition module can receive the business request text sent by the rule engine module, and then use the model invocation proxy to determine the business scenario of the business request text, evaluating at least one predicted business scenario corresponding to the business request text.

[0061] When the model calls the proxy to evaluate and obtain a single predicted business scenario, it can directly retrieve the prompt word template corresponding to that single business scenario, concatenate the prompt word template with the business request text to form a complete request message, and then send it to the large language model for inference.

[0062] A user may have multiple business intentions in a single query, which may cause the model invocation agent to evaluate multiple estimated business scenarios. The model invocation agent can then retrieve the prompt word template corresponding to each estimated business scenario, concatenate multiple prompt word templates with the business request text to obtain a composite prompt word, form a complete request message, and send it to the large language model for inference.

[0063] After the large language model completes inference, it returns the response text corresponding to the business request text, the business identifier of at least one second business type, and the model confidence score as the second identification result.

[0064] Optionally, the confidence evaluator can assess the quality of the output of the large language model. Based on this, the large model intent recognition module is also used to: determine the quality score of the second recognition result based on the format legality of the response text, the attribution of the business identifier of the second business type in the low-altitude flight business system, and the model confidence.

[0065] The confidence evaluator comprehensively considers the legality of the response text output format, the attribution verification of the business identifier of the second business type in the candidate set of the business system, and the model confidence returned by the large model to generate a final confidence index, which serves as the quality score of the second identification result. When this quality score is lower than a preset quality threshold, the system can trigger a manual review process or record it as a sample to be optimized, providing a basis for subsequent rule supplementation and model tuning.

[0066] The aforementioned large-scale model intent recognition module, through contextualized prompt word templates and an independent confidence evaluation mechanism, retains the generalization and understanding capabilities of the large-scale model while achieving modular management of prompt words and controllable quality of model output. This avoids the coupling problem and unpredictable side effects of a single giant prompt word.

[0067] Furthermore, within the large model inference path, the second recognition result can be transmitted to the interpretability tracking module. Correspondingly, the interpretability tracking module is used for: Upon receiving the second recognition result sent by the large model intent recognition module, the second decision channel identifier, the version identifier of the target prompt word template, the version identifier of the large language model, the response text, and the quality score corresponding to the second recognition result are obtained and recorded as the second tracking information. Generate a decision path summary of the second tracking information as the second audit link information.

[0068] In one possible implementation, when the interpretability tracking module receives the second recognition result sent by the large model intent recognition module, it can mark the current decision channel as the second decision channel identifier, and obtain the version identifier of the target prompt word template, the version identifier and quality score of the large language model, as well as the original response text carried by the second recognition result from the large model intent recognition module. It can automatically record complete information such as the prompt word template version (llm-prompt-version), the large model version (llm-version), the original inference output (llm-output), and the quality score (llm-score). It can also record the final routing decision and end-to-end processing latency of the large model inference path, and store them in the interpretability tracking module as second tracking information, providing reliable technical support for problem localization, effect auditing, and compliance review.

[0069] To improve transmission efficiency, the aforementioned second tracking information can be aggregated into a structured decision path summary based on the identifier corresponding to the business request text. This decision path summary may include the second decision channel identifier and key decision-making criteria for the large model inference path. This decision path summary, as the second audit link information, is transmitted to the request routing module along with the business identifier of the identified second business type.

[0070] Based on this, the request routing module is used for: Obtain the second audit link information, and the business identifier of the second business type carried by the second identification result; Configure the second confidence level corresponding to the second identification result as a quality score; Based on the second audit link information, the business identifier of the second business type, and the second confidence level, generate the second business intent information corresponding to the business request text; The second business intent information is sent to the low-altitude flight business system corresponding to the second business type.

[0071] In one possible implementation, after receiving the second audit link information and the second identification result sent by the interpretability tracing module, the request routing module can parse and obtain the service identifier of the second service type carried in the second identification result, and use this service identifier as the user intent code for the final system determination. The second confidence level corresponding to the second identification result is configured as the aforementioned quality score. Distinguished from the first confidence level determined by the rule engine's hit path, the second confidence level of the large model inference path is dynamic, fluctuating dynamically with the large model's semantic inference results, allowing downstream business systems to intuitively perceive the credibility of the current large model identification result.

[0072] The second audit link information, the business identifier of the second business type, and the second confidence level are used as three core fields. These are encapsulated in a structured form as the second business intent information corresponding to the business request text, and sent to the low-altitude flight business system corresponding to the second business type as the final intent recognition result.

[0073] The aforementioned request routing module outputs the identification results and decision path summaries in one integrated manner, enabling downstream business systems not only to obtain the identified business identifiers, but also to perceive the reliability of the identification results and the decision-making channels, providing structured information support for differentiated business processing and system optimization.

[0074] The following will use the flight application scenario in the low-altitude economic field as an example to illustrate the specific implementation process of the embodiments of the present invention.

[0075] A city's low-altitude flight service platform processes approximately 5,000 drone flight application requests daily. The system deploys the business request system provided by this invention, and the collaborative workflow of each module is as follows.

[0076] Step 1: Data Input Module Reception and Verification. The user submits a query request via mobile terminal: "Request to conduct aerial photography flight in Area B from 10:00 AM to 12:00 PM tomorrow." After receiving this text, the data input module performs basic validity checks: it checks that the input length is normal, the encoding format conforms to the UTF-8 standard, and there are no sensitive characters. After successful verification, the original query text is output to the rule engine module.

[0077] Step Two: Rule Engine Module Matching and Judgment. After receiving the input, the rule engine module traverses the deterministic rule base. The rule base pre-sets a "Flight Plan Declaration" rule, triggered by the co-occurrence of the keyword combination "application" and "flight," and associated with a time window and regional entity constraints. The similarity calculation engine performs a multi-factor weighted fusion evaluation on this input: the text similarity is calculated to be 0.92, the keyword hit rate is 1.0 (both core keywords "application" and "flight" are hit), and the semantic vector similarity is 0.95. After fusion according to preset weights, the overall similarity score is calculated. If this score is greater than the preset threshold of 0.85, the rule engine determines a successful match and outputs the intent label "FLIGHT_PLAN_SUBMIT," completely bypassing the large model intent recognition module, with an end-to-end time of approximately 5 milliseconds.

[0078] Step 3: Large Model Intent Recognition Module (not triggered in this case). Since the rule engine has successfully matched, the dynamic routing mechanism skips the large model intent recognition module, thus avoiding model call overhead.

[0079] Step 4: The interpretability tracking module records the decision path. The interpretability tracking module automatically records complete information about this decision: the processing channel is labeled "Rule Engine," the triggering rule is identified as "FP_001," the similarity scores for each dimension are as follows, the overall score is as follows, and the response time is 5 milliseconds. This information is aggregated into a structured decision path based on the request identifier and transmitted synchronously with the recognition results.

[0080] Step 5: The request routing module returns a structured result. The request routing module encapsulates the following output: the business intent category is "FLIGHT_PLAN_SUBMIT", the confidence score is 1.0 (a definite value when the rule engine hits), and the decision path summary is "Rule Hit - FP_001 - Overall Similarity 0.961". The downstream flight approval system automatically creates flight plan work orders and allocates airspace resources based on this, without any manual intervention.

[0081] When another user inputs a complex, non-standard request, "We have an urgent task that requires temporary operation near the no-fly zone in Area C. How do we apply for this?", the rule engine calculates that its highest overall similarity with all rules is only 0.47, below the threshold of 0.85. The system automatically routes the request to the large model's intent recognition module. The large model uses the "Emergency Work Permit Consultation" prompt template for inference and outputs the intent "EMERGENCY_FLIGHT_INQUIRY" with a confidence level of 0.91 and a response time of approximately 180 milliseconds. The interpretability tracking module synchronously records the model's inference trajectory, and the request routing module returns the intent category, confidence level, and decision path.

[0082] Through the aforementioned collaborative mechanism, the rule engine module identifies approximately 70% of the high-frequency standardized application requests on the platform within milliseconds, achieving 100% accuracy in deterministic intent; the remaining complex requests are handled by the large model module for deep understanding, balancing efficiency and generalization ability.

[0083] This embodiment can achieve the following beneficial effects: This invention provides a business request system for low-altitude flight services. The system includes a data input module, a rule engine module, a large-scale model intent recognition module, an interpretability tracking module, and a request routing module. The rule engine module is positioned as the primary decision-making channel, and the large-scale model intent recognition module as the secondary decision-making channel. The two are automatically routed based on rule matching. A hierarchical collaborative system of "rule priority, model fallback" is constructed, allowing deterministic intents to completely bypass model reasoning. This architecture simultaneously ensures accuracy and response speed, fundamentally eliminating probabilistic uncertainty for standardized expressions and meeting the rigid reliability requirements of serious scenarios such as low-altitude economic flight applications.

[0084] Furthermore, the business request system designs the interpretability tracing module as an independent and integrated functional module rather than a post-event log system. It generates audit link information uniformly across both rule engine hits and large model inference paths, filling the gap in decision-making transparency in existing solutions. This design ensures that the complete evidence for each intent identification is transparent and verifiable, providing reliable technical support for problem localization, effect auditing, and compliance review.

[0085] Based on the same inventive concept, this embodiment of the invention provides a service request method for low-altitude flight services. This method is applied to the aforementioned service request system, which includes a data input module, a rule engine module, a large-scale model intent recognition module, an interpretability tracing module, and a request routing module. The implementation of this method is similar to that described above, and will not be repeated in this embodiment.

[0086] Reference Figure 3 The diagram shows a service request method for low-altitude flight services, which includes the following steps 301-305: Step 301: Receive the service request text through the data input module; Step 302: The rule engine module performs rule matching based on the business request text to determine whether a preset business type is matched; if matched, the corresponding first recognition result is output and sent to the interpretability tracking module; if not matched, the business request text is sent to the large model intent recognition module. Step 303: When the rule engine module fails to match the intent recognition of the large model, the large language model is invoked to perform intent recognition on the business request text, output the corresponding second recognition result, and send the second recognition result to the interpretability tracking module. Step 304: Based on the first identification result, the interpretability tracking module generates first audit link information and sends the first identification result and the first audit link information together to the request routing module; or, based on the second identification result, generates second audit link information and sends the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; Step 305: The request routing module generates target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and sends the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system provides corresponding low-altitude flight services based on the business intent information.

[0087] Optionally, step 302 includes: Iterate through the preset trigger condition text corresponding to each preset business type, and determine the matching degree value corresponding to each preset business type based on the text similarity, keyword hit rate and semantic vector similarity between the business request text and the current preset trigger condition text; The text similarity is used to identify the business request text containing standardized expressions; the keyword hit rate is used to represent different business types through multiple keywords with different weights; and the semantic vector similarity is used to identify the business request text that has the same semantics as the standardized expression, or that has sentence rewriting or entity synonym substitution. If there is a preset service type with a matching degree value greater than a preset threshold, then obtain the first service type corresponding to the maximum matching degree value and output the corresponding first identification result. The first identification result includes the service identifier of the first service type. If no preset service type with a matching degree value greater than the preset threshold exists, it is determined that the match has not been found, and the service request text is sent to the large model intent recognition module.

[0088] Optionally, step 304 includes: Upon receiving the first recognition result sent by the rule engine module, the first decision channel identifier, trigger condition information, and similarity information corresponding to the first recognition result are obtained and recorded as the first tracking information. The trigger condition information includes the rule identifier and condition content corresponding to the preset trigger condition text, and the similarity information includes the text similarity, the keyword hit rate, the semantic vector similarity, and the maximum matching degree value. Generate a decision path summary of the first tracking information as the first audit link information.

[0089] Optionally, step 305 includes: Obtain the first audit link information and the service identifier of the first service type carried by the first identification result; Configure the first confidence level corresponding to the first identification result to a preset value; Based on the first audit link information, the business identifier of the first business type, and the first confidence level, the first business intent information corresponding to the business request text is generated; The first business intent information is sent to the low-altitude flight business system corresponding to the first business type.

[0090] Optionally, the large model intent recognition module stores prompt word templates corresponding to different business scenarios; Step 302 includes: The large language model is invoked to determine the business scenario of the business request text and to evaluate at least one estimated business scenario corresponding to the business request text. Obtain the target prompt word template corresponding to the estimated business scenario, combine the target prompt word template with the business request text, and input it into the large language model for intent recognition. Receive the second recognition result returned by the large language model. The second recognition result includes the response text, the business identifier of the second business type, and the model confidence level.

[0091] Optionally, step 302 further includes: Based on the format validity of the response text, the attribution of the business identifier of the second business type in the low-altitude flight business system, and the model confidence level, the quality score of the second identification result is determined.

[0092] Optionally, step 304 includes: Upon receiving the second recognition result sent by the large model intent recognition module, the second decision channel identifier corresponding to the second recognition result, the version identifier of the target prompt word template, the version identifier of the large language model, the response text, and the quality score are obtained and recorded as the second tracking information. Generate a decision path summary of the second tracking information as the second audit link information.

[0093] Optionally, step 305 includes: Obtain the second audit link information and the service identifier of the second service type carried by the second identification result; Configure the second confidence level corresponding to the second identification result as the quality score; Based on the second audit link information, the business identifier of the second business type, and the second confidence level, second business intent information corresponding to the business request text is generated; The second business intent information is sent to the low-altitude flight business system corresponding to the second business type.

[0094] This embodiment can achieve the following beneficial effects: This invention provides a business request system for low-altitude flight services. The system includes a data input module, a rule engine module, a large-scale model intent recognition module, an interpretability tracking module, and a request routing module. The rule engine module is positioned as the primary decision-making channel, and the large-scale model intent recognition module as the secondary decision-making channel. The two are automatically routed based on rule matching. A hierarchical collaborative system of "rule priority, model fallback" is constructed, allowing deterministic intents to completely bypass model reasoning. This architecture simultaneously ensures accuracy and response speed, fundamentally eliminating probabilistic uncertainty for standardized expressions and meeting the rigid reliability requirements of serious scenarios such as low-altitude economic flight applications.

[0095] Furthermore, the business request system designs the interpretability tracing module as an independent and integrated functional module rather than a post-event log system. It generates audit link information uniformly across both rule engine hits and large model inference paths, filling the gap in decision-making transparency in existing solutions. This design ensures that the complete evidence for each intent identification is transparent and verifiable, providing reliable technical support for problem localization, effect auditing, and compliance review.

[0096] An exemplary embodiment of the present invention also provides an electronic device equipped with a service request system for low-altitude flight services provided in an embodiment of the present invention. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform a method according to an embodiment of the present invention.

[0097] An exemplary embodiment of the present invention also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0098] An exemplary embodiment of the present invention also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of the present invention.

[0099] refer to Figure 4The present invention will now be described in the form of a structural block diagram of an electronic device 400 that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0100] like Figure 4 As shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0101] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, output unit 407, storage unit 408, and communication unit 409. Input unit 406 can be any type of device capable of inputting information to electronic device 400. Input unit 406 can receive input digital or text information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 407 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 408 may include, but is not limited to, disks and optical discs. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, Wi-Fi devices, WiMax devices, cellular communication devices, and / or the like.

[0102] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above. For example, in some embodiments, the above-described service request method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 can be configured to perform the above-described service request method by any other suitable means (e.g., by means of firmware).

[0103] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0104] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0105] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0106] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0107] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0108] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A service request system for low-altitude flight services, characterized in that, The system includes a data input module, a rule engine module, a large model intent recognition module, an interpretability tracing module, and a request routing module; The data input module is used to receive business request text; The rule engine module is used to perform rule matching based on the business request text to determine whether a preset business type is matched. If a match is found, the corresponding first identification result is output and sent to the interpretability tracking module; If no match is found, the business request text is sent to the large model intent recognition module; The large model intent recognition module is used to call the large language model to perform intent recognition on the business request text when the rule engine module fails to match, output the corresponding second recognition result, and send the second recognition result to the interpretability tracking module. The interpretability tracking module is configured to, upon receiving a first identification result sent by the rule engine module, obtain the first decision channel identifier, trigger condition information, and similarity information corresponding to the first identification result from the rule engine module, record them as first tracking information, generate a decision path summary of the first tracking information as first audit link information, and send the first identification result and the first audit link information together to the request routing module; or, upon receiving a second identification result sent by the large model intent recognition module, obtain the second decision channel identifier, target prompt word template version identifier, large language model version identifier, response text, and quality score corresponding to the second identification result from the large model intent recognition module, record them as second tracking information, generate a decision path summary of the second tracking information as second audit link information, and send the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; The request routing module is used to generate target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and send the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system provides corresponding low-altitude flight services based on the business intent information.

2. The system according to claim 1, characterized in that, The rule engine module is used for: Iterate through the preset trigger condition text corresponding to each preset business type, and determine the matching degree value corresponding to each preset business type based on the text similarity, keyword hit rate and semantic vector similarity between the business request text and the current preset trigger condition text; The text similarity is used to identify the business request text containing standardized expressions; the keyword hit rate is used to represent different business types through multiple keywords with different weights; and the semantic vector similarity is used to identify the business request text that has the same semantics as the standardized expression, or that has sentence rewriting or entity synonym substitution. If there is a preset service type with a matching degree value greater than a preset threshold, then obtain the first service type corresponding to the maximum matching degree value and output the corresponding first identification result. The first identification result includes the service identifier of the first service type. If no preset service type with a matching degree value greater than the preset threshold exists, it is determined that the match has not been found, and the service request text is sent to the large model intent recognition module.

3. The system according to claim 2, characterized in that, The triggering condition information includes the rule identifier and condition content corresponding to the preset triggering condition text, and the similarity information includes the text similarity, the keyword hit rate, the semantic vector similarity, and the maximum matching degree value.

4. The system according to claim 3, characterized in that, The request routing module is used for: Obtain the first audit link information and the service identifier of the first service type carried by the first identification result; Configure the first confidence level corresponding to the first identification result to a preset value; Based on the first audit link information, the business identifier of the first business type, and the first confidence level, the first business intent information corresponding to the business request text is generated; The first business intent information is sent to the low-altitude flight business system corresponding to the first business type.

5. The system according to claim 1, characterized in that, The large model intent recognition module stores prompt word templates corresponding to different business scenarios; The large model intent recognition module is used for: The large language model is invoked to determine the business scenario of the business request text and to evaluate at least one estimated business scenario corresponding to the business request text. Obtain the target prompt word template corresponding to the estimated business scenario, combine the target prompt word template with the business request text, and input it into the large language model for intent recognition. Receive the second recognition result returned by the large language model. The second recognition result includes the response text, the business identifier of the second business type, and the model confidence level.

6. The system according to claim 5, characterized in that, The large model intent recognition module is also used for: Based on the format validity of the response text, the attribution of the business identifier of the second business type in the low-altitude flight business system, and the model confidence level, the quality score of the second identification result is determined.

7. The system according to claim 6, characterized in that, The request routing module is used for: Obtain the second audit link information and the service identifier of the second service type carried by the second identification result; Configure the second confidence level corresponding to the second identification result as the quality score; Based on the second audit link information, the business identifier of the second business type, and the second confidence level, second business intent information corresponding to the business request text is generated; The second business intent information is sent to the low-altitude flight business system corresponding to the second business type.

8. A service request method for low-altitude flight services, characterized in that, The method is applied to the business request system as described in any one of claims 1-7, the system comprising a data input module, a rule engine module, a large model intent recognition module, an interpretability tracing module, and a request routing module; The method includes: The data input module receives the business request text. The rule engine module performs rule matching based on the business request text to determine whether a preset business type is matched. If matched, the corresponding first recognition result is output and sent to the interpretability tracking module. If not matched, the business request text is sent to the large model intent recognition module. When the rule engine module fails to match the large model intent recognition module, the large language model is invoked to perform intent recognition on the business request text, outputting the corresponding second recognition result, and sending the second recognition result to the interpretability tracking module. The interpretability tracking module, upon receiving a first identification result from the rule engine module, retrieves the first decision channel identifier, trigger condition information, and similarity information corresponding to the first identification result from the rule engine module, records them as first tracking information, generates a decision path summary of the first tracking information as first audit link information, and sends the first identification result and the first audit link information together to the request routing module; or, upon receiving a second identification result from the large model intent recognition module, retrieves the second decision channel identifier, target prompt word template version identifier, large language model version identifier, response text, and quality score corresponding to the second identification result from the large model intent recognition module, records them as second tracking information, generates a decision path summary of the second tracking information as second audit link information, and sends the second identification result and the second audit link information together to the request routing module; wherein, the first audit link information is used to characterize the decision process information corresponding to the rule engine module, and the second audit link information is used to characterize the decision process information corresponding to the large model intent recognition module; The request routing module generates target business intent information corresponding to the business request text based on the first identification result and the first audit link information, or based on the second identification result and the second audit link information; and sends the target business intent information to the corresponding low-altitude flight business system so that the low-altitude flight business system provides corresponding low-altitude flight services based on the business intent information.

9. An electronic device, characterized in that, The electronic device includes the system as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method as described in claim 8.

Images