A large model-based building address intelligent correction method and system

Abstract

This invention discloses a building address intelligent correction method and system based on a large model. The method includes: a client submitting data to be verified, including unit, floor, waiting room name, and type, to a verification server; the verification server integrating the data, predefined business rules, and verification instructions to construct structured verification prompts; calling the large model service through an application programming interface (API), which parses the prompts and performs analysis and reasoning based on semantic associations and logical constraints; the large model encapsulates the judgment results into JSON format and returns them; the verification server parses the results, generates verification conclusions and specific correction suggestions, and feeds them back to the client. This invention utilizes the semantic understanding and logical reasoning capabilities of a large language model to construct a building address data quality verification expert system. Through a carefully designed prompt engineering, knowledge such as building physics common sense, waiting room type definitions, and verification rules is injected into the large model to achieve intelligent verification of building address data and generation of correction suggestions.

Description

Technical Field

[0001] This invention relates to the field of smart building technology, and in particular to a method and system for intelligent building address correction based on a large model. Background Technology

[0002] In smart building systems, building address data forms the foundation for core functions such as elevator scheduling, visitor navigation, and property management. Building address data typically includes multiple fields such as unit name, floor name, waiting hall name, and waiting hall type, and these fields have complex logical relationships.

[0003] For example, waiting hall types include four types: lobby, underground parking, office floor, and commercial complex. Each type has a clear physical constraint on the floor range: the lobby is usually located on the ground floor, the underground parking is located on the basement level, and the office floor is located on the upper floors above ground, etc.

[0004] The disadvantages of existing technologies are as follows: (1) Low efficiency of manual review: The amount of building address data is large, and manual review of each item is time-consuming and labor-intensive, and errors are easily missed.

[0005] (2) Difficulty in maintaining the rule engine: Traditional rule-based verification methods require exhaustively listing all abnormal scenarios. The number of rules is huge and it is difficult to cover boundary situations, such as the "sky lobby" being located on a high floor, which is a reasonable scenario.

[0006] (3) Lack of semantic understanding: Traditional methods cannot understand the semantic relationship between fields, such as the correspondence between "B2 elevator hall" and "underground parking lot", and the correspondence between "35th floor elevator hall" and "office floor".

[0007] (4) Unable to provide intelligent suggestions: After an error is found, it can only be marked, but cannot provide reasonable correction suggestions.

[0008] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0009] The main objective of this invention is to provide a building address intelligent correction method and system based on a large model, which aims to solve the problems of inefficiency, incomplete coverage, lack of semantic understanding, and inability to provide correction suggestions caused by existing reliance on manual review and static rule engines when errors occur in building address data.

[0010] To achieve the above objectives, the present invention provides a building address intelligent correction method based on a large model, the building address intelligent correction method based on a large model comprising the following steps: The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields. After the large model service completes the analysis, it encapsulates the judgment results in a preset format to obtain JSON verification results and returns them to the verification server. The verification server parses and processes the JSON verification result, and returns a clear verification conclusion and intelligent correction suggestions to the client.

[0011] Optionally, in the aforementioned intelligent building address correction method based on a large model, the client submits the building address data to be verified to the verification server, specifically including: The client will contain verification data including unit name, floor name, waiting room name, and waiting room type, forming building address data in a predefined format; A network request is initiated through the application programming interface (API) to send the building address data to the API endpoint specified by the verification server.

[0012] Optionally, in the aforementioned intelligent building address correction method based on a large model, the verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompts, specifically including: After receiving the building address data, the verification server parses and extracts the unit name, floor name, waiting room name, and waiting room type fields from the building address data; The verification server assembles the extracted field values, rules defining the physical meaning of the waiting hall type, dictionaries defining the floor feature word recognition method, scenario rules defining the mandatory error reporting logic, and a reasonable scenario whitelist for handling special cases according to the preset prompt word template framework, in a modular structure to generate structured verification prompt words.

[0013] Optionally, the intelligent building address correction method based on a large model, wherein the large model service parses the verification prompt words and comprehensively considers the semantic relationships and logical constraints between fields to perform semantic understanding and logical reasoning, specifically including: The large model service receives and parses the verification prompt words, and defines the role and task of the large model as a building data quality verification expert. Based on the domain knowledge injected into the verification prompt words, feature word analysis is performed on the floor name to determine whether it belongs to the underground floor, the ground floor, or the upper floor, and the definition of the waiting hall type is associated with it to infer and judge the semantic consistency between the fields.

[0014] Optionally, the intelligent building address correction method based on a large model, wherein after the large model service completes the analysis, it encapsulates the judgment result according to a preset format to obtain a JSON verification result and returns it to the verification server, specifically including: After completing semantic and logical analysis, the large model service generates a preliminary Boolean judgment on whether the data has a problem, a brief description of the problem type, and a detailed logical analysis of the reasons. The large model service, based on the preset output format specifications, encapsulates the preliminary Boolean judgment, the brief description, the logical analysis reasons, and the correction suggestions generated based on the analysis into a JSON validation result that conforms to the JavaScript object notation format, and returns it to the validation server.

[0015] Optionally, in the aforementioned intelligent building address correction method based on a large model, the verification server parses and post-processes the JSON verification result, returning a clear verification conclusion and intelligent correction suggestions to the client, specifically including: The verification server receives JSON verification results that conform to the JavaScript object notation format and parses out the problem presence flag, problem type, reason, and suggestion fields of the JSON verification results; The verification server generates a clear conclusion text indicating whether the verification passed or failed based on the problem existence flag, and combines the clear conclusion text with the problem type, the reason, and the suggestion into a final response message, which is then returned to the client through the original application interface path.

[0016] Optionally, the intelligent building address correction method based on a large model further includes: When constructing the verification prompt words, a preset number of learning examples are integrated, wherein the learning examples include a set of input-output pairs, which are used to guide the large model service to respond according to the specified output format and logical reasoning mode.

[0017] Optionally, the building address intelligent correction method based on the large model includes core definitions regarding the matching relationship between waiting hall type and floor range in the predefined business rule constraints. For example, the common floor for the lobby type is the ground floor; the common floor for the underground parking type is the underground floor; the common floor for the office floor type is the upper floor above ground; and the common floor for the commercial complex type is the lower floor or podium.

[0018] Optionally, the intelligent building address correction method based on the large model further includes mandatory error reporting scenario rules in the predefined business rule constraints; the mandatory error reporting scenario rules include at least the following: when the waiting hall type is identified as an underground parking lot and the associated floor name is indicated as an above-ground floor, an error must be reported.

[0019] In addition, to achieve the above objectives, the present invention also provides a building address intelligent correction system based on a large model, wherein the building address intelligent correction system based on a large model includes: a client, a verification server, and a large model service; The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields. After the large model service completes the analysis, it encapsulates the judgment result in a preset format to obtain a JSON verification result and returns it to the verification server. The verification server parses and processes the JSON verification result, and returns a clear verification conclusion and intelligent correction suggestions to the client.

[0020] In this invention, the client submits building address data to be verified to the verification server. This building address data includes unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompts. The verification server then calls a large model service through an application programming interface (API) to retrieve these prompts. The large model service parses the prompts, comprehensively considers the semantic relationships and logical constraints between fields, and performs semantic understanding and logical reasoning. After analysis, the large model service encapsulates the results into a JSON verification result according to a preset format and returns it to the verification server. The verification server parses and post-processes the JSON verification result, returning a clear verification conclusion and intelligent correction suggestions to the client. This invention utilizes the semantic understanding and logical reasoning capabilities of a large language model to construct a building address data quality verification expert system. Through a carefully designed prompt engineering process, knowledge such as building physics, waiting room type definitions, and verification rules is injected into the large model, enabling intelligent verification of building address data and generation of correction suggestions. Attached Figure Description

[0021] Figure 1 This is a flowchart of a preferred embodiment of the intelligent building address correction method based on a large model of the present invention; Figure 2 This is a schematic diagram illustrating the interaction between the client, the verification server, and the large model service in a preferred embodiment of the intelligent building address correction system based on a large model according to the present invention. Figure 3 This is a flowchart of the verification process in a preferred embodiment of the intelligent building address correction method based on a large model of the present invention. Figure 4 This is a flowchart of batch verification in a preferred embodiment of the intelligent building address correction method based on a large model of the present invention; Figure 5 This is a schematic diagram of the error data detection interface in a preferred embodiment of the intelligent building address correction method based on a large model of the present invention; Figure 6 This is a schematic diagram of the error marking interface of the location list in a preferred embodiment of the intelligent building address correction method based on a large model of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] The preferred embodiment of the intelligent building address correction method based on a large model described in this invention, such as... Figure 1and Figure 2 As shown, the intelligent building address correction method based on a large model includes the following steps: Step S10: The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type.

[0024] Specifically, the client will take the data to be verified, including the unit name, floor name, waiting room name, and waiting room type, and form a building address data in a predefined format; then, through the application programming interface (API), it will initiate a network request to send the building address data to the API endpoint specified by the verification server.

[0025] Four input fields are defined: unit name (unit_name), floor name (floor_name), hall name (hall_name), and hall type (hall_type). Hall types include lobby, underground parking, office floor, and commercial complex. Table 1 defines the physical meaning of the four hall types. Table 1: Four Types of Waiting Lots

[0026] Step S20: The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words.

[0027] Specifically, after receiving the building address data, the verification server parses and extracts the unit name, floor name, waiting room name, and waiting room type fields from the building address data. Based on the preset prompt word template framework, the verification server assembles the extracted field values, the rules defining the physical meaning of the waiting room type, the dictionary defining the floor feature word recognition method, the scenario rules defining the mandatory error reporting logic, and the reasonable scenario whitelist for handling special cases according to the modular structure to generate structured verification prompt words.

[0028] By incorporating knowledge from areas such as building physics, waiting room type definition, feature word dictionary, and verification rules into a carefully designed prompt word structure, the general-purpose model possesses professional building address verification capabilities.

[0029] As shown in Table 2, the rules for identifying floor feature words are defined as follows: Table 2: Floor Feature Word Recognition Rules

[0030] When constructing the verification prompts, a preset number of learning examples are integrated. These learning examples include a set of input-output pairs to guide the large model service to respond according to a specified output format and logical reasoning pattern. Through carefully selected input-output examples, the large model is guided to output verification results in a specified JSON format (JavaScript Object Notation), ensuring that the output results can be parsed and processed by the program.

[0031] The predefined business rules constraints include core definitions regarding the matching relationship between waiting hall types and floor ranges. For example, the common floor for the lobby type is the ground floor; the common floor for the underground parking type is the basement; the common floor for the office floor type is the upper floors above ground; and the common floor for the commercial complex type is the lower floors or the podium.

[0032] The predefined business rule constraints also include mandatory error reporting scenario rules; the mandatory error reporting scenario rules include at least the following: when the waiting hall type is identified as an underground parking lot and the associated floor name indicates an above-ground floor, an error must be reported.

[0033] As shown in Table 3, the scenarios that must report an error are defined as follows: Table 3: Scenarios Where Errors Must Be Reported

[0034] Define a special scenario where no errors are reported: Commercial complex + B1 level: The basement level of a shopping mall is usually a commercial area, which is a normal configuration.

[0035] Step S30: The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields.

[0036] Specifically, the large model service receives and parses the verification prompt words, defines the role and task of the large model as a building data quality verification expert (clarifies that its core capability is to perform logical verification based on the common sense of building in the physical world); based on the domain knowledge injected in the verification prompt words, it performs feature word analysis on the floor name to determine whether it belongs to the underground floor, the ground floor or the high floor, and associates it with the definition of the waiting hall type, and performs reasoning judgment on the semantic consistency between fields.

[0037] Step S40: After the large model service completes the analysis, it encapsulates the judgment results in a preset format to obtain JSON verification results and returns them to the verification server.

[0038] Specifically, after completing semantic and logical analysis, the large model service generates a preliminary Boolean judgment on whether the data has a problem, a brief description of the problem type, and a detailed logical analysis of the reasons. According to the preset output format specifications, the large model service encapsulates the preliminary Boolean judgment, the brief description, the logical analysis reasons, and the correction suggestions generated based on the analysis into a JSON validation result that conforms to the JavaScript object notation format and returns it to the validation server.

[0039] Define the JSON output structure: has_problem: A boolean value indicating whether a problem exists; problem_type: A brief description of the problem type; Reason: Detailed logical analysis; suggestion: to revise the suggestion.

[0040] Step S50: The verification server parses and post-processes the JSON verification result, and returns the clear verification conclusion and intelligent correction suggestions to the client.

[0041] Specifically, the verification server receives a JSON verification result conforming to the JavaScript object notation format and parses out the problem presence flag, problem type, reason, and suggestion fields of the JSON verification result; the verification server then generates a clear conclusion text indicating whether the verification passed or failed based on the problem presence flag, and combines the clear conclusion text with the problem type, reason, and suggestion into a final response message, which is returned to the client through the original application interface path.

[0042] like Figure 3 As shown, the first step is to construct the prompt word. The verification server doesn't simply forward the raw data; instead, it initiates a crucial preprocessing step. Based on a carefully designed prompt word engineering architecture, it integrates and assembles the received raw data fields with a built-in domain knowledge base (including the physical definition of waiting room types, floor feature word recognition rules, and various mandatory error reporting logic scenarios) to construct a rigorously structured and clearly defined prompt word. This prompt word is essentially a complete package of working instructions containing the task background, expert roles, specific data, and verification requirements.

[0043] The second step involves calling the large language model for inference and verifying the results. The server sends the constructed prompts to the large language model service via the Application Programming Interface (API). Upon receiving the instruction, the large language model (LLM) activates its core semantic understanding and logical reasoning capabilities. It first parses all the information and constraints in the prompts, and then, like a professional building data expert, begins in-depth analysis of the input data.

[0044] The third step: Feature word identification and logical judgment. During the analysis, the large model performs a crucial sub-step: feature word identification. It parses fields such as floor names based on the feature word identification dictionary provided in the prompts. For example, it determines that B2 belongs to the underground level, floor 01 belongs to the ground floor, and floor 35 belongs to the high-rise level. Subsequently, the model logically associates and matches the feature word identification results with the definition of the waiting hall type (e.g., whether the underground parking type matches the underground level feature B2).

[0045] Step 4: Whitelist Priority Verification. Before performing regular rule verification, the system introduces an optimization step: checking if the data matches the whitelist. The large model or verification server logic will determine whether the current data combination belongs to the predefined reasonable scenario whitelist (e.g., a commercial complex type appearing on layer B1). If so, the data is directly deemed problem-free, skipping subsequent detailed rule verification and directly proceeding to the result return stage. This mechanism effectively avoids false positives for special but reasonable scenarios.

[0046] Step 5: Rule Validation and Result Generation. If the data does not match the whitelist (i.e., the process does not proceed), the large model will continue to rigorously check according to the mandatory error scenarios in the validation rule module. If logical conflicts or semantic inconsistencies are found (e.g., the underground parking type corresponds to the above-ground floor 08), the data is determined to have a problem. At this point, the large model will generate a problem and suggestions, that is, not only indicating the problem type and cause, but also generating specific and actionable correction suggestions based on domain knowledge (e.g., suggesting changing the type to 'office floor'). If all rule validations pass, the data is determined to be problem-free.

[0047] Step 6: Formatting the Return Result. Finally, the large model encapsulates the final judgment result (whether there is a problem), problem details, and correction suggestions strictly according to the output format defined in the prompt (i.e., the specified JSON structure), and returns this structured result to the verification server. The server then parses and post-processes it, and finally feeds back the clear and easy-to-understand verification conclusion and intelligent correction suggestions to the client that initially made the request, thus completing a complete intelligent correction loop.

[0048] Guide the large model's output format and inference logic through carefully selected examples: (1) Example 1: Normal data.

[0049] Input: Unit = "Building A", Floor = "01", Hall Name = "Building A Lobby", Type = "Lobby".

[0050] Output: has_problem=false, reason=“The lobby is located on the ground floor (floor 01), which meets the definition.”

[0051] (2) Example 2: The underground parking lot is located above ground.

[0052] Input: Unit = "Basement", Floor = "8th Floor", Hall Name = "8th Floor Elevator Hall", Type = "Underground Parking".

[0053] Output: has_problem=true, problem_type="underground parking is located on a ground floor", suggestion="suggestion change to office floor".

[0054] (3) Example 3: The lobby floor is too high.

[0055] Input: Unit = "Building 1", Floor = "35th Floor", Hall Name = "35th Floor Elevator Hall", Type = "Lobby".

[0056] Output: has_problem=true, problem_type="Lobby type floor is too high", suggestion="It is recommended to change it to office floor".

[0057] (4) Example 4: The office floor is located underground.

[0058] Input: Unit = "Office Building", Floor = "B2", Hall Name = "B2 Elevator Hall", Type = "Office Floor".

[0059] Output: has_problem=true, problem_type="office floor is underground", suggestion="suggestion change to underground parking".

[0060] like Figure 4 As shown, the first step is receiving and initiating a batch task. The process begins with the verification server receiving batch data. The client submits a list or file containing multiple building address records to the verification server. After recognizing that this is a batch task, the verification server initiates the corresponding batch processing flow.

[0061] The second step involves iterating and validating each record. The core operation on the server side is to traverse each data record. It sequentially retrieves individual records from the dataset, which typically contain fields such as unit name, floor name, waiting room name, and waiting room type. For each retrieved data record, the server calls the large model validation module. This module represents the single-data-record processing flow: it constructs structured hints for the data record, calls the large model to perform feature word recognition, whitelist checks, rule-based logical reasoning, and other analyses, ultimately obtaining a structured validation result containing fields such as has_problem, problem_type, reason, and suggestion.

[0062] Step 3: Result Collection and Loop Check. After each call to the large model to obtain the validation result of a single data entry, the validation server immediately collects the result and temporarily stores it in memory or temporary storage. Then, the system performs a conditional check: whether there is still data, i.e., checking if the currently traversed data is the last record in the dataset. If so (there is another data entry), the process returns to the step of traversing each data entry and continues processing the next record, forming a loop. This loop continues until all records in the dataset have been processed.

[0063] Step 4: Results Summarization and Report Generation. Once all data has been traversed and validated (i.e., the result is negative and there is no more data), the process enters the post-processing stage. The validation server will summarize the problem data, filter out records with the has_problem flag set to true from all collected results, and may perform statistical analysis and classification.

[0064] Finally, based on the summarized issue data and the verification status of all records, the system generates a verification report. This report is typically a structured document or data file, and may include: an overall verification overview (such as total number, number of issues, pass rate), a detailed list of issues (including specific information and correction suggestions for each issue), and possible statistical analysis. This report is ultimately returned to the client for users to view, analyze, and perform batch corrections.

[0065] As shown in Table 4, the achieved results are as follows: Table 4: Comparison of the effects achieved by the existing technical solution and the solution of the present invention

[0066] Verification example: Example 1: Normal data.

[0067] Input data:

[0068] Verification result:

[0069] Example 2: Error data.

[0070] Input data:

[0071] Verification result:

[0072] The prompt word template structure is as follows:

[0073] Practical application effects such as Figure 5 and Figure 6 As shown, Figure 5 The error data detection interface indicates that the system detected the waiting hall type as "underground parking," but the floor "01" is the ground floor, and the name "East Lobby" indicates this is the lobby area. An error message automatically pops up with correction suggestions. It is recommended to change the type to "Lobby," or check if the floor and name are entered incorrectly. Figure 6 This is the error marking interface for the location list. In the location list, the system automatically marks data with problems and displays "Error reminder: Waiting hall type is defined as...". Users can click to view detailed error reasons and correction suggestions.

[0074] The technical effects that this invention can bring are as follows: (1) Significantly improved verification efficiency and automation: By replacing manual review with a large model and designing a batch processing flow, the system achieves second-level automated verification of massive building address data. The processing speed is several orders of magnitude faster than manual review, completely solving the problem of time-consuming and labor-intensive manual review.

[0075] (2) Enhanced accuracy and semantic understanding: By leveraging the powerful semantic understanding and logical reasoning capabilities of the large model, combined with carefully injected domain knowledge (feature words, business rules), the system can understand the deep logical relationships between fields (such as the matching of B2 and parking lot) like an expert, which significantly improves the discrimination accuracy of complex scenarios and boundary situations (such as the sky lobby) and overcomes the shortcomings of traditional rule engines in terms of incomplete coverage.

[0076] (3) System flexibility and maintainability optimization: The core technical logic is encapsulated in an adjustable prompt word project, rather than a hard-coded rule base. When business rules change or new scenarios need to be expanded (such as adding a waiting hall type), only the prompt word template needs to be optimized or the knowledge module needs to be updated. There is no need to make large-scale modifications to the program code, which greatly reduces the system maintenance cost and iteration difficulty.

[0077] (4) Closed-loop correction and intelligent decision support: The system output is not only a binary judgment of pass / fail, but also provides specific and actionable intelligent correction suggestions (such as suggesting that the type be changed to office floor). This forms a complete data governance closed loop of problem discovery, cause location and solution provision, upgrading the system from a simple inspection tool to an intelligent expert that assists decision-making, directly improving the efficiency and accuracy of data correction.

[0078] Furthermore, such as Figure 2 As shown, based on the above-mentioned intelligent building address correction method based on a large model, the present invention also provides an intelligent building address correction system based on a large model, wherein the intelligent building address correction system based on a large model includes: a client, a verification server, and a large model service; The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields. After the large model service completes the analysis, it encapsulates the judgment result in a preset format to obtain a JSON verification result and returns it to the verification server. The verification server parses and processes the JSON verification result, and returns a clear verification conclusion and intelligent correction suggestions to the client.

[0079] In summary, this invention provides a building address intelligent correction method and system based on a large model. The method includes: a client submitting building address data to be verified to a verification server, the building address data including unit name, floor name, waiting room name, and waiting room type; the verification server integrating the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; the verification server calling the large model service through an application programming interface (API) to retrieve the verification prompt words, the large model service parsing the verification prompt words and comprehensively considering the semantic relationships and logical constraints between fields to perform semantic understanding and logical reasoning; after completing the analysis, the large model service encapsulates the judgment result according to a preset format to obtain a JSON verification result and returns it to the verification server; the verification server parses and post-processes the JSON verification result, returning a clear verification conclusion and intelligent correction suggestions to the client. This invention utilizes the semantic understanding and logical reasoning capabilities of a large language model to construct an expert system for verifying the quality of building address data. Through carefully designed prompt word engineering, knowledge such as building physics common sense, waiting hall type definition, and verification rules are injected into the large model to achieve intelligent verification of building address data and generation of correction suggestions.

[0080] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0081] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0082] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A building address intelligent correction method based on a large model, characterized in that, The intelligent building address correction method based on a large model includes: The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields. After the large model service completes the analysis, it encapsulates the judgment results in a preset format to obtain JSON verification results and returns them to the verification server. The verification server parses and processes the JSON verification result, and returns a clear verification conclusion and intelligent correction suggestions to the client.

2. The intelligent building address correction method based on a large model according to claim 1, characterized in that, The client submits the building address data that needs to be verified to the verification server, specifically including: The client will contain verification data including unit name, floor name, waiting room name, and waiting room type, forming building address data in a predefined format; A network request is initiated through the application programming interface (API) to send the building address data to the API endpoint specified by the verification server.

3. The intelligent building address correction method based on a large model according to claim 1, characterized in that, The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompts, specifically including: After receiving the building address data, the verification server parses and extracts the unit name, floor name, waiting room name, and waiting room type fields from the building address data; The verification server assembles the extracted field values, rules defining the physical meaning of the waiting hall type, dictionaries defining the floor feature word recognition method, scenario rules defining the mandatory error reporting logic, and a reasonable scenario whitelist for handling special cases according to the preset prompt word template framework, in a modular structure to generate structured verification prompt words.

4. The intelligent building address correction method based on a large model according to claim 1, characterized in that, The large model service parses the verification prompt words and, taking into account the semantic relationships and logical constraints between fields, performs semantic understanding and logical reasoning, specifically including: The large model service receives and parses the verification prompt words, and defines the role and task of the large model as a building data quality verification expert. Based on the domain knowledge injected into the verification prompt words, feature word analysis is performed on the floor name to determine whether it belongs to the underground floor, the ground floor, or the upper floor, and the definition of the waiting hall type is associated with it to infer and judge the semantic consistency between the fields.

5. The intelligent building address correction method based on a large model according to claim 1, characterized in that, After the large model service completes the analysis, it encapsulates the judgment results into a JSON verification result according to a preset format and returns it to the verification server. Specifically, this includes: After completing semantic and logical analysis, the large model service generates a preliminary Boolean judgment on whether the data has a problem, a brief description of the problem type, and a detailed logical analysis of the reasons. The large model service, based on the preset output format specifications, encapsulates the preliminary Boolean judgment, the brief description, the logical analysis reasons, and the correction suggestions generated based on the analysis into a JSON validation result that conforms to the JavaScript object notation format, and returns it to the validation server.

6. The intelligent building address correction method based on a large model according to claim 5, characterized in that, The verification server parses and processes the JSON verification result, returning a clear verification conclusion and intelligent correction suggestions to the client, specifically including: The verification server receives JSON verification results that conform to the JavaScript object notation format and parses out the problem presence flag, problem type, reason, and suggestion fields of the JSON verification results; The verification server generates a clear conclusion text indicating whether the verification passed or failed based on the problem existence flag, and combines the clear conclusion text with the problem type, the reason, and the suggestion into a final response message, which is then returned to the client through the original application interface path.

7. The intelligent building address correction method based on a large model according to claim 3 or 4, characterized in that, The intelligent building address correction method based on a large model also includes: When constructing the verification prompt words, a preset number of learning examples are integrated, wherein the learning examples include a set of input-output pairs, which are used to guide the large model service to respond according to the specified output format and logical reasoning mode.

8. The intelligent building address correction method based on a large model according to claim 1, characterized in that, The predefined business rules constraints include core definitions regarding the matching relationship between waiting hall types and floor ranges. For example, the common floor for the lobby type is the ground floor; the common floor for the underground parking type is the basement; the common floor for the office floor type is the upper floors above ground; and the common floor for the commercial complex type is the lower floors or the podium.

9. The intelligent building address correction method based on a large model according to claim 8, characterized in that, The predefined business rule constraints also include mandatory error reporting scenario rules; the mandatory error reporting scenario rules include at least the following: when the waiting hall type is identified as an underground parking lot and the associated floor name indicates an above-ground floor, an error must be reported.

10. A building address intelligent correction system based on a large model, characterized in that, The intelligent building address correction system based on a large model includes: a client, a verification server, and a large model service; The client submits the building address data that needs to be verified to the verification server. The building address data includes the unit name, floor name, waiting room name, and waiting room type. The verification server integrates the building address data, predefined business rule constraints, and explicit verification instructions to construct verification prompt words; The verification server calls the large model service through the application programming interface to retrieve the verification prompt words. The large model service parses the verification prompt words and performs semantic understanding and logical reasoning by comprehensively considering the semantic relationships and logical constraints between fields. After the large model service completes the analysis, it encapsulates the judgment result in a preset format to obtain a JSON verification result and returns it to the verification server. The verification server parses and processes the JSON verification result, and returns a clear verification conclusion and intelligent correction suggestions to the client.

Images