Geolocation risk assessment method and apparatus based on large model, device, and medium

Abstract

The application belongs to the technical field of data processing and is suitable for the financial field, and discloses a geographic position risk assessment method, device, equipment and medium based on a large model, which comprises the following steps: acquiring an original address text of a subject to be assessed, performing address analysis on the original address text, and generating a standardized address; performing geocoding processing on the standardized address to generate longitude and latitude coordinates corresponding to the standardized address; inputting the original address text, the standardized address and the longitude and latitude coordinates into a pre-trained second large language model to perform spatial semantic consistency verification, and outputting a credibility label of the longitude and latitude coordinates after the verification is passed; acquiring a multi-disaster risk quantization vector corresponding to the longitude and latitude coordinates; inputting the standardized address, the credibility label and the multi-disaster risk quantization vector into a pre-trained third large language model to perform reasoning, and generating a risk rating result. The accuracy of geographic position risk assessment is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology and is applicable to the financial field. In particular, it relates to a method, device, equipment and medium for geolocation risk assessment based on a large model. Background Technology

[0002] In scenarios such as insurance underwriting and financial asset risk assessment, accurately determining geographical location is a crucial prerequisite for risk assessment. Insurance businesses need to assess natural disaster risks based on the location of overseas assets, while financial businesses need to conduct risk exposure analysis on collateralized assets or supply chain nodes.

[0003] Currently, geolocation risk assessment commonly employs a technical approach combining rule-based address resolution, third-party geocoding services, and single-risk-layer queries. However, overseas address resolution capabilities are weak, and rule-based methods struggle to handle multilingual content, informal place names, aliases, and outdated names, resulting in high failure rates. For example, in financial scenarios, overseas mortgaged asset addresses often contain local colloquialisms or old administrative division names, which the rule base cannot match, leading to untraceable asset locations. Secondly, geocoding accuracy is difficult to guarantee and lacks verification; third-party services often return significantly inaccurate coordinates for vague or remote addresses, sometimes even resulting in incorrect country attribution. Furthermore, risk assessment results lack interpretability, outputting only a single risk value or level, leaving business personnel unable to understand the basis and key driving factors behind the risk conclusions.

[0004] Therefore, improving the accuracy of geographic location risk assessment is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for geolocation risk assessment based on a large model, in order to solve the technical problem of how to improve the accuracy of geolocation risk assessment.

[0006] In a first aspect, the present invention provides a geographical location risk assessment method based on a large model, comprising: Obtain the original address text of the target to be evaluated, and use the pre-trained first language model to perform address parsing on the original address text to generate the standardized address of the target to be evaluated. The standardized address is geocoded to generate the latitude and longitude coordinates corresponding to the standardized address; The original address text, the standardized address, and the latitude and longitude coordinates are input into the pre-trained second language model for spatial semantic consistency verification. The model determines whether the latitude and longitude coordinates are consistent with the region described by the original address text, and outputs the credibility label of the latitude and longitude coordinates after the verification is passed. The multi-hazard risk layer is queried using the verified latitude and longitude coordinates to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The standardized address, the credibility label, and the multi-hazard risk quantification vector are input into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated.

[0007] Secondly, the present invention provides a location risk assessment device based on a large model, comprising: The acquisition module is used to acquire the original address text of the target to be evaluated, and to perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. The encoding module is used to perform geocoding on the standardized address and generate the latitude and longitude coordinates corresponding to the standardized address; The determination module is used to input the original address text, the standardized address, and the latitude and longitude coordinates into the pre-trained second language model to perform spatial semantic consistency verification, determine whether the latitude and longitude coordinates are consistent with the region described by the original address text, and output the credibility label of the latitude and longitude coordinates after the verification is passed; The query module is used to query the multi-hazard risk layer using the verified latitude and longitude coordinates, and obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The reasoning module is used to input the standardized address, the credibility label, and the multi-hazard risk quantification vector into a pre-trained third language model for reasoning, and generate a risk rating result for the target to be evaluated.

[0008] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for assessing geographical location risk based on a large model.

[0009] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for assessing geographical location risk based on a large model.

[0010] The aforementioned method, apparatus, equipment, and medium for geographic location risk assessment based on a large-scale model, implemented in this scheme, allows the client to obtain the original address text of the target to be assessed. A pre-trained first large-scale language model is used to parse the original address text, generating a standardized address for the target. The standardized address is then geocoded to generate its corresponding latitude and longitude coordinates. The original address text, the standardized address, and the latitude and longitude coordinates are input into a pre-trained second large-scale language model for spatial semantic consistency verification. This verifies whether the latitude and longitude coordinates match the region described in the original address text, and outputs a confidence label for the latitude and longitude coordinates upon successful verification. The verified latitude and longitude coordinates are used to query a multi-hazard risk layer to obtain the corresponding multi-hazard risk quantification vector. Finally, the standardized address, the confidence label, and the multi-hazard risk quantification vector are input into a pre-trained third large-scale language model for inference, generating a risk rating result for the target. In this invention, the pre-trained second large-scale language model performs spatial semantic consistency verification on the geocoding result and outputs a confidence label, automatically eliminating erroneous results where the coordinate attribution is inconsistent with the address description. Furthermore, by using a pre-trained third-generation speech model to determine risk assessment strategies based on credibility labels, the risk assessment conclusions can be adaptively adjusted according to differences in coordinate accuracy, effectively improving the accuracy of geographic location risk assessment. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of an application environment for a geographical location risk assessment method based on a large model, according to an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of a geographical location risk assessment method based on a large model in one embodiment of the present invention.

[0014] Figure 3 yes Figure 2 A schematic diagram of a specific implementation method for step S10.

[0015] Figure 4 yes Figure 2 A schematic diagram of a specific implementation method for step S20.

[0016] Figure 5 yes Figure 2A schematic diagram of a specific implementation method for step S30.

[0017] Figure 6 This is a schematic diagram of a geographical location risk assessment device based on a large model in one embodiment of the present invention.

[0018] Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0019] Figure 8 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The geographic location risk assessment method based on a large model provided in this invention can be applied to, for example... Figure 1 In the application environment, Figure 1This is a schematic diagram of an application environment for a large-scale model-based geographic location risk assessment method according to an embodiment of the present invention. The client communicates with the server via a network. The server can obtain the original address text of the target to be assessed from the client, and use a pre-trained first large-scale language model to parse the original address text, generating a standardized address for the target. The standardized address is then geocoded to generate its corresponding latitude and longitude coordinates. The original address text, the standardized address, and the latitude and longitude coordinates are input into a pre-trained second large-scale language model for spatial semantic consistency verification, determining whether the latitude and longitude coordinates match the region described in the original address text, and outputting a confidence label for the latitude and longitude coordinates upon successful verification. The verified latitude and longitude coordinates are used to query a multi-hazard risk layer to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates. The standardized address, the confidence label, and the multi-hazard risk quantification vector are input into a pre-trained third large-scale language model for inference, generating a risk rating result for the target to be assessed. In this invention, a pre-trained second-largest speech model is used to perform spatial semantic consistency verification on the geocoding results and output a credibility label, which can automatically eliminate erroneous results where the coordinate attribution is inconsistent with the address description. Furthermore, a pre-trained third-largest speech model determines a risk assessment strategy based on the credibility label, enabling the risk assessment conclusion to adaptively adjust according to differences in coordinate accuracy, effectively improving the accuracy of geographic location risk assessment. The invention will now be described in detail through specific embodiments.

[0022] Please see Figure 2 As shown, Figure 2 This is a flowchart illustrating a large-model-based geographic location risk assessment method according to an embodiment of the present invention. The large-model-based geographic location risk assessment method specifically includes the following steps: S10: Obtain the original address text of the target to be evaluated, and perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. Specifically, as follows... Figure 3 The above, Figure 3 yes Figure 2 A schematic flowchart of a specific implementation of step S20 includes the following steps S11-S13: S11: Perform semantic analysis on the original address text to obtain semantic fragments corresponding to different address levels. Specifically, in this embodiment of the invention, a pre-trained first large language model is used to understand the semantics of the original address text, rather than relying on traditional rule matching or keyword segmentation. The large language model can automatically determine which parts of the text belong to different address levels such as country, province, city, and street based on the contextual semantics, and segment continuous unstructured text into semantic fragments corresponding to each level. For example, in a financial scenario, the original address of an overseas mortgaged asset might be described as "near Pacific Mall in South Jakarta, Indonesia, next to the former site of a trading company from the Dutch colonial era." Through semantic analysis, Indonesia can be identified as the country level, South Jakarta as the city and region level, Pacific Mall as the landmark level, and "next to the former site of a trading company from the Dutch colonial era" as an auxiliary description, thereby completing the initial sorting of the mixed text.

[0023] S12: Based on the geographical knowledge built into the pre-trained first language model, the informal place names of semantic fragments at each address level are mapped to their corresponding formal place names. This specifically includes the following steps S121-S123: S121: The pre-trained first large language model is used to perform place name entity discrimination on the semantic fragments, identifying informal place name entities. Specifically, in this embodiment of the invention, the named entity recognition capability of the large language model is used to perform place name entity discrimination on semantic fragments at each address level. The large language model can combine contextual semantics to determine whether a fragment is a place name and whether the place name belongs to an informal expression. For example, in a financial scenario, after semantic analysis, the address of an overseas mortgaged asset yields semantic fragments such as Jakarta South, Pacific Place, and the former site of a Dutch colonial trading company. The large language model determines that Jakarta South is an informal place name entity at the administrative division level, Pacific Place is an informal place name entity at the landmark level, and the former site of a Dutch colonial trading company is determined to be an auxiliary description rather than a place name entity, thereby achieving accurate differentiation between place name and non-place name information in semantic fragments. This invention can accurately identify various informal place name entities from mixed text, avoiding the problem of traditional rule-based methods misjudging non-place name fragments as place names or omitting informal place names, and providing accurate target objects for subsequent place name mapping.

[0024] S122: The identified informal place name entities and their corresponding address level labels are input into the pre-trained first large language model. Utilizing the geographic knowledge graph and contextual semantics built into the pre-trained first large language model, at least one candidate formal place name that semantically matches the informal place name entity is generated. Specifically, in this embodiment of the invention, the internal geographic knowledge graph constructed during the pre-training process of the large language model, combined with the hierarchical constraints provided by the address level labels, is used to perform semantic reasoning on the informal place name entities, generating corresponding candidate formal place names. The large language model internally stores the relationships between the formal names, aliases, old names, and common names of a large number of place names worldwide. When an informal place name entity and its level label are input, the model can infer the formal place name that the informal place name might point to based on these implicit geographic knowledge graphs. For example, in a financial context, inputting the informal place name "South Jakarta" and its corresponding city / region level label into a large language model allows the model to infer, based on contextual semantics and built-in geographical knowledge, that the informal place name likely corresponds to the administrative city of South Jakarta. Similarly, inputting Pacific Place and its landmark level label generates a candidate list of multiple specific buildings within the commercial complex, such as Pacific Place Tower A and Pacific Place Tower B. This invention fully utilizes the massive geographical knowledge reserves of the large language model, enabling intelligent mapping of informal place names worldwide. It eliminates the need for manual maintenance of a large place name dictionary, significantly reducing system maintenance costs and improving mapping coverage.

[0025] S123: Obtain semantic fragments of the address levels adjacent to the informal place name entity in the original address text. Use the semantic fragments of the adjacent address levels as spatial constraints to verify the spatial inclusion relationship of the at least one candidate formal place name. Filter out candidate formal place names with mismatched geographical ranges and use the retained candidate formal place names as the formal place names corresponding to the informal place names after mapping. Specifically, in this embodiment of the invention, the hierarchical spatial inclusion relationship of the address text itself is used to verify and filter the generated candidate formal place names. There is a clear spatial inclusion relationship between the levels in the address text, that is, the lower-level address must be within the geographical range of the upper-level address. By obtaining the semantic fragments of the upper-level address levels adjacent to the informal place name entity as spatial constraints, it is verified whether each candidate formal place name is truly located within the geographical range of the upper-level address, thereby filtering out candidate results that do not conform to the spatial inclusion relationship. For example, in a financial scenario, Taikoo Square generates two candidate buildings: Taikoo Square Tower A and Taikoo Square Tower B. By using the official place name South Jakarta Administrative City, which maps to the upper-level semantic fragment of Taikoo Square's South Jakarta area, as a spatial constraint, it verifies whether both candidates are located within South Jakarta Administrative City. If Taikoo Square Tower B is actually located in Central Jakarta Administrative City, it conflicts with the spatial constraint and will be eliminated. Ultimately, Taikoo Square Tower A is retained as the official place name after Taikoo Square's mapping. This invention effectively solves the ambiguous mapping problem that may arise in large language models when spatial constraints are lacking by introducing logical verification of spatial inclusion relationships. It avoids mapping errors caused by landmarks with the same name or similar names, significantly improving the accuracy and reliability of informal place name mapping.

[0026] S13: Extract valid address elements from the semantic fragments after mapping processing, and reorganize the extracted valid address elements according to a preset geographical hierarchy structure to generate standardized addresses with a hierarchical structure. Specifically, in this embodiment of the invention, the semantic fragments after mapping processing undergo element extraction and structured reorganization. Each address level in the semantic fragments after mapping processing has been replaced with an official place name. Valid address elements such as country, first-level administrative region, city, district / county, and street are extracted from them, and reorganized from top to bottom according to a preset geographical hierarchy structure such as country-province / state-city-district / county-street-house number to generate standardized addresses with clear hierarchy and uniform format.

[0027] S20: Perform geocoding on the standardized address to generate the latitude and longitude coordinates corresponding to the standardized address. Specifically, for example... Figure 4 The above, Figure 4 yes Figure 2 A schematic flowchart of a specific implementation of step S20 includes the following steps S21-S23: S21: The pre-trained first large language model is used to perform place name entity discrimination on the semantic fragments, identifying informal place name entities. Specifically, in this embodiment of the invention, the deep semantic understanding capability of the large language model is utilized to perform place name entity discrimination on the semantic fragments at each address level obtained from the decomposition. The large language model can combine the context to distinguish which parts of the semantic fragment belong to place name information, which belong to directional descriptions or redundant descriptions, and further determine whether the place name belongs to an informal expression. This invention can accurately identify informal place name entities from mixed text, avoiding the parsing errors caused by traditional rule-based methods misclassifying auxiliary descriptions as place names.

[0028] S22: The identified informal place name entities and their corresponding address level labels are input into the pre-trained first large language model. Utilizing the built-in geographical knowledge graph and contextual semantics of the pre-trained first large language model, at least one candidate formal place name that semantically matches the informal place name entity is generated. Specifically, in this embodiment of the invention, the internal geographical knowledge graph constructed during the pre-training process of the large language model, combined with the hierarchical constraints and contextual semantics provided by the address level labels, is used to perform semantic reasoning and matching on informal place name entities. The large language model internally stores the implicit relationships between the formal names, aliases, old names, and common names of a large number of place names worldwide. When an informal place name entity and its level label are input, the model can activate relevant knowledge nodes and infer the set of candidate formal place names that the informal place name may refer to. This invention fully utilizes the massive geographical knowledge reserves of the large language model for mapping, eliminating the need for manually building and maintaining a large global place name comparison dictionary, significantly reducing the operation and maintenance costs of overseas address resolution systems in financial businesses.

[0029] S23: Obtain semantic fragments of the address levels adjacent to the informal place name entity in the original address text. Use the semantic fragments of the adjacent address levels as spatial constraints to verify the spatial inclusion relationship of the at least one candidate formal place name. Filter out candidate formal place names with mismatched geographical scopes and use the retained candidate formal place names as the formal place names corresponding to the informal place names after mapping. Specifically, in this embodiment of the invention, the natural administrative affiliation and geographical inclusion relationship between the various levels within the address text is used to perform spatial consistency verification on the generated candidate formal place names. In any address text, the next level of administrative division must be within the geographical scope of the previous level of administrative division; this spatial inclusion relationship constitutes a strict logical constraint. This invention obtains semantic fragments of the upper-level address levels adjacent to the informal place name entity, uses them as spatial constraints, and verifies one by one whether each candidate formal place name is truly located within the geographical scope of the upper-level address, thereby filtering out interference items that do not conform to the spatial inclusion relationship.

[0030] S30: Input the original address text, the standardized address, and the latitude and longitude coordinates into the pre-trained second language model for spatial semantic consistency verification. Determine whether the latitude and longitude coordinates are consistent with the region described by the original address text, and output the confidence label of the latitude and longitude coordinates after the verification passes. Specifically, as follows... Figure 5 The above, Figure 5 yes Figure 2 A schematic flowchart of a specific implementation of step S30 includes the following steps S31-S34: S31: The actual region of the latitude and longitude coordinates is determined using the pre-trained second large language model, and the determined actual region is compared with the region described in the original address text. Specifically, in this embodiment of the invention, the understanding ability of the large language model of geospatial information is utilized to reverse-parse the latitude and longitude coordinates returned by the geocoding service into the actual administrative region it falls into, and a semantic comparison is performed with the region described in the original address text. The second large language model is invoked to reverse-engineer the actual country or region to which the coordinates belong based on the latitude and longitude coordinates, and then this result is verified for consistency with the country or region described in the user's original input address.

[0031] S32: If the comparison result is inconsistent, the verification is deemed unsuccessful, and the latitude and longitude coordinates are marked as invalid coordinates. Specifically, in this embodiment of the invention, an automated error coordinate identification and blocking mechanism is established. When the comparison result shows that the actual location of the coordinates is inconsistent with the location described in the address text, it indicates that the coordinates returned by the geocoding service have a fundamental error, and the coordinates have lost their value as the basis for subsequent risk analysis. The latitude and longitude coordinates are then explicitly marked as invalid coordinates, preventing them from entering the downstream risk layer query and risk rating process. For example, in a financial scenario, an address in Paris, France, is geocoded to Germany. The comparison result is inconsistent, and the coordinates are marked as invalid. The system automatically terminates the subsequent processing of the coordinates and issues an address resolution anomaly warning to business personnel.

[0032] S33: If the comparison result is consistent, the verification is deemed successful, and the address completeness of the original address text is evaluated using the pre-trained second large language model. Specifically, in this embodiment of the invention, coordinates that pass the regional consistency verification are not directly assigned the same confidence level, but rather the fineness of the original address text itself is further quantitatively evaluated. The address completeness determines the extent to which the coordinates can represent the true location of the target. The large language model can understand the number of levels and information granularity contained in the address text, and thereby determine whether the address is a detailed address accurate to the house number or a vague address describing only the city or province.

[0033] S34: Output the corresponding level of credibility label for the latitude and longitude coordinates based on the assessment result of the address integrity. Specifically, in this embodiment of the invention, the qualitative assessment of the address integrity is transformed into a structured credibility label that can be directly referenced in the downstream risk analysis process. The credibility label is a comprehensive evaluation of the quality of this address resolution and geocoding process. It carries information on the extent to which the coordinates can be accepted, guiding the adoption of different assessment strategies in subsequent risk analysis.

[0034] S40: Use the verified latitude and longitude coordinates to query the multi-hazard risk layer and obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates. Specifically, this includes the following steps S41-S43: S41: Input the verified latitude and longitude coordinates into the preset geographic information service interface. Specifically, in this embodiment of the invention, only latitude and longitude coordinates that have passed spatial semantic consistency verification and are determined to be valid are used. The preset geographic information service interface is a unified access point for the underlying multi-hazard risk database. It encapsulates complex spatial data retrieval logic and provides standardized query services to upper-layer business modules. For example, in a financial scenario, a bank needs to conduct a natural disaster risk assessment on an overseas mortgage asset in Jakarta, Indonesia. After parsing, encoding, and verification, the asset address yields a set of latitude and longitude coordinates with a credibility label. This invention inputs the verified coordinates into the geographic information service interface, using the verification result as a prerequisite for querying, thus ensuring the data input quality for mortgage asset risk assessment in financial business.

[0035] S42: Through the geographic information service interface, query the physical hazard parameters of each hazard corresponding to the latitude and longitude coordinates in the pre-constructed multi-hazard risk grid layer. Specifically, in this embodiment of the invention, a pre-constructed grid layer technology is used to store and retrieve global natural disaster risk data. The pre-constructed multi-hazard risk grid layer pre-calculates and stores the physical hazard parameters of various natural disasters such as earthquakes, floods, and typhoons in a database according to unified spatial grid units. When a query request is received, the system only needs to locate the grid unit where the request is located based on the latitude and longitude coordinates to quickly extract the hazard parameters corresponding to that unit, without performing complex real-time geographic calculations. For example, in a financial scenario, after the coordinates of the Jakarta, Indonesia asset are input into the geographic information service interface, the system locates the grid unit where the coordinates are located in the pre-constructed grid layer and simultaneously queries the physical hazard parameters of multiple hazards corresponding to that unit, such as peak ground acceleration, flood inundation depth, and typhoon maximum wind speed. This invention achieves parallel and efficient querying of multi-hazard data through a pre-constructed grid layer, significantly shortening the response time for risk data acquisition.

[0036] S43: Extract the physical hazard parameters for each hazard type from the query, and integrate the extracted physical hazard parameters into a structured multi-hazard risk quantification vector. Specifically, in this embodiment of the invention, the scattered physical hazard parameters for each hazard type obtained from the query are standardized and integrated according to a preset data structure to generate a structured risk quantification vector that can be directly processed by downstream analysis models. This vector uniformly encapsulates physical hazard parameters at different scales into a standard set of numerical features, which not only preserves the independent risk information of each hazard type, but also provides well-formatted input data for subsequent comprehensive risk inference.

[0037] S50: Input the standardized address, the credibility tag, and the multi-hazard risk quantification vector into the pre-trained third language model for inference to generate a risk rating result for the target to be evaluated. Specifically, this includes the following steps S51-S52: S51: The standardized address, the credibility tag, and the multi-hazard risk quantification vector are combined into a structured prompt word, and the structured prompt word is input into the pre-trained third language model. Specifically, in this embodiment of the invention, the heterogeneous information produced in the aforementioned stages is uniformly encapsulated into structured prompt words, which serve as the standardized input to the third language model. The standardized address carries the spatial location information of the target object, the credibility tag carries the precision constraints of the coordinates, and the multi-hazard risk quantification vector carries the objective physical hazard parameters. Together, these three constitute the complete information set required for risk assessment. By organizing this information into structured prompt words according to a preset template, the large language model can accurately understand the meaning of each piece of information and their interrelationships, avoiding confusion and ambiguity in information input.

[0038] S52: The pre-trained third language model determines the risk assessment strategy for the target to be evaluated based on the credibility label, and performs inference based on the determined risk assessment strategy and the multi-hazard risk quantification vector to generate the risk rating result of the target to be evaluated. Specifically, in this embodiment of the invention, the credibility label is used as the core basis for adjusting the risk assessment strategy, enabling the large language model to adopt a graded inference method according to the accuracy difference of the coordinates. When the credibility label is high or medium, it indicates that the coordinates can accurately reflect the true location of the target, and the model adopts a precise assessment strategy, directly judging the risk level based on the parameters in the risk quantification vector corresponding to the coordinates. When the credibility label is low, it indicates that the coordinates can only represent the approximate area where the target is located, and the model automatically switches to a regional conservative assessment strategy, considering the historical highest risk value in the region based on the risk quantification vector.

[0039] In one embodiment of the present invention, after inputting the standardized address, the credibility label, and the multi-hazard risk quantification vector into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated, the method further includes: S61: The risk rating result, the standardized address, the credibility tag, and the multi-hazard risk quantification vector are combined into report generation prompts. Specifically, in this embodiment of the invention, various key information generated throughout the risk assessment process are systematically summarized as the data basis for generating the risk assessment report. The risk rating result is the final risk level determination, the standardized address provides a spatial location description of the target, the credibility tag indicates the data quality level of this assessment, and the multi-hazard risk quantification vector records the specific physical hazard parameters of each hazard. Combining these four types of information according to a preset template into report generation prompts allows the large language model to obtain the complete information needed to generate risk interpretation text.

[0040] S62: The report generation prompts are input into a pre-trained fourth language model for reasoning to generate a risk interpretation text for the target to be evaluated. This risk interpretation text includes a risk level explanation, key causative factors, and corresponding risk response strategies. Specifically, in this embodiment of the invention, the natural language generation capability of the large language model is used to transform the quantitative risk data in the structured prompts into risk interpretation text understandable to business personnel. During the reasoning process, the fourth language model first semantically interprets the risk rating results, explaining the position and meaning of the level within the risk spectrum; secondly, it extracts the most prominent disaster parameters from the risk quantification vector, identifies the key causative factors, and analyzes their causes; finally, combining the risk level and key causative factors, it generates targeted risk response strategies from the perspectives of risk mitigation and business response. For example, in a financial scenario, the fourth language model generates prompts based on reports of Indonesian mortgaged assets for inference. The generated risk interpretation text includes the following: The risk level explanation section states that a medium to high risk level indicates that the asset has a relatively high probability of suffering losses due to natural disasters during the loan term, requiring corresponding risk mitigation measures; the main disaster-causing factors section points out that floods are the primary source of risk for the asset, with the region's historical maximum flood depth reaching 2.0 meters, followed by earthquake risk, while the impact of typhoons is relatively limited; the risk response strategy section proposes strategies requiring borrowers to purchase property insurance covering flood liability, appropriately increase loan interest rates to cover additional risk premiums, and include the asset in a key monitoring list for regular risk reassessment.

[0041] S63: Integrate the risk rating results and the risk interpretation text to generate a risk assessment report for the target asset, and output the risk assessment report. Specifically, in this embodiment of the invention, the quantitative risk rating results and the qualitative risk interpretation text are integrated to generate a standardized and complete risk assessment report. The risk rating results provide a concise and intuitive summary of conclusions for the report, while the risk interpretation text provides detailed analytical basis and action strategies. The two complement each other to form a professional risk assessment document. For example, in a financial scenario, the medium-to-high risk level of Indonesian mortgage assets is integrated with the generated risk interpretation text. In the report template, the risk level is placed on the first page as the core summary of conclusions, followed by an explanation of the risk level, details of parameters for each type of disaster, analysis of major disaster-causing factors, and risk response strategies, ultimately generating a complete natural disaster risk assessment report for mortgage assets.

[0042] As can be seen, in the above scheme, specifically in this invention, the pre-trained second-largest speech model performs spatial semantic consistency verification on the geocoding results and outputs credibility labels, which can automatically eliminate erroneous results where coordinate attribution and address description are inconsistent. Furthermore, the pre-trained third-largest speech model determines the risk assessment strategy based on the credibility labels, enabling the risk assessment conclusions to adaptively adjust according to differences in coordinate accuracy, effectively improving the accuracy of geographic location risk assessment.

[0043] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0044] In one embodiment, a large-scale model-based geographic location risk assessment device is provided, which corresponds one-to-one with the large-scale model-based geographic location risk assessment method described in the above embodiments. For example... Figure 6 As shown, Figure 6 This is a schematic diagram of a location risk assessment device based on a large model according to an embodiment of the present invention. The device includes an acquisition module 61, an encoding module 62, a judgment module 63, a query module 64, and an inference module 65. Detailed descriptions of each functional module are as follows: The acquisition module 61 is used to acquire the original address text of the target to be evaluated, and to perform address parsing on the original address text through a pre-trained first language model to generate the standardized address of the target to be evaluated. Encoding module 62 is used to perform geocoding processing on the standardized address to generate the latitude and longitude coordinates corresponding to the standardized address; The determination module 63 is used to input the original address text, the standardized address, and the latitude and longitude coordinates into the pre-trained second language model for spatial semantic consistency verification, determine whether the latitude and longitude coordinates are consistent with the region described by the original address text, and output the credibility label of the latitude and longitude coordinates after the verification is passed; The query module 64 is used to query the multi-hazard risk layer using the verified latitude and longitude coordinates, and obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The reasoning module 65 is used to input the standardized address, the credibility label, and the multi-hazard risk quantification vector into the pre-trained third language model for reasoning, and generate the risk rating result of the target to be evaluated.

[0045] In one embodiment, the acquisition module 61 is specifically used for: Semantic analysis is performed on the original address text to obtain semantic fragments corresponding to different address levels; Based on the geographical knowledge built into the pre-trained first language model, the informal place names of semantic fragments at each address level are mapped to the corresponding formal place names. Valid address elements are extracted from the semantic fragments after mapping and then reorganized according to a preset geographical hierarchy to generate a standardized address with a hierarchical structure.

[0046] In one embodiment, the acquisition module 61 is further configured to: The semantic fragment is subjected to place name entity discrimination using the pre-trained first language model, and informal place name entities are identified. The identified informal place name entities and their corresponding address level labels are input into the pre-trained first language model. The geographical knowledge graph and contextual semantics built into the pre-trained first language model are used to generate at least one candidate formal place name that matches the semantics of the informal place name entities. Semantic fragments of the address level adjacent to the informal place name entity in the original address text are obtained. The semantic fragments of the adjacent address level are used as spatial constraints to verify the spatial inclusion relationship of the at least one candidate formal place name. Candidate formal place names with mismatched geographical ranges are filtered out, and the retained candidate formal place names are used as the formal place names corresponding to the informal place names after mapping.

[0047] In one embodiment, the determination module 63 is specifically used for: The actual region of the latitude and longitude coordinates is determined by the pre-trained second language model, and the determined actual region is compared with the region described by the original address text. If the comparison results are inconsistent, the verification is deemed to have failed, and the latitude and longitude coordinates are marked as invalid coordinates. If the comparison results are consistent, the verification is deemed successful, and the address integrity of the original address text is evaluated using the pre-trained second language model. Based on the assessment results of the address completeness, output the corresponding level of confidence label for the latitude and longitude coordinates.

[0048] In one embodiment, the query module 64 is further configured to: Input the verified latitude and longitude coordinates into the preset geographic information service interface; Through the geographic information service interface, query the physical hazard parameters of each hazard corresponding to the latitude and longitude coordinates in the pre-constructed multi-hazard risk grid layer; Extract the physical hazard parameters of each hazard type from the query, and integrate the extracted physical hazard parameters of each hazard type into a structured multi-hazard risk quantification vector.

[0049] In one embodiment, the inference module 65 is specifically used for: The standardized address, the credibility label, and the multi-hazard risk quantification vector are combined into a structured prompt word, and the structured prompt word is input into a pre-trained third language model; The pre-trained third language model determines the risk assessment strategy for the target to be assessed based on the credibility label, and performs inference based on the determined risk assessment strategy and the multi-hazard risk quantification vector to generate the risk rating result of the target to be assessed.

[0050] In one embodiment, the large-model-based geographic location risk assessment device is further used for: The risk rating result, the standardized address, the credibility label, and the multi-hazard risk quantification vector are combined to generate prompt words for the report. The report generation prompts are input into the pre-trained fourth language model for reasoning to generate the risk interpretation text of the target to be evaluated, wherein the risk interpretation text includes the risk level explanation, the main disaster-causing factors and the corresponding risk response strategies. The risk rating results and the risk interpretation text are integrated to generate a risk assessment report for the target to be assessed, and the risk assessment report is then output.

[0051] This invention provides a geolocation risk assessment device based on a large model. It uses a pre-trained second large speech model to perform spatial semantic consistency verification on geocoding results and outputs credibility labels, automatically eliminating erroneous results where coordinate attribution and address description are inconsistent. Furthermore, a pre-trained third large speech model determines a risk assessment strategy based on the credibility labels, enabling the risk assessment conclusions to adaptively adjust according to differences in coordinate accuracy, effectively improving the accuracy of geolocation risk assessment.

[0052] Specific limitations regarding the large-model-based geographic location risk assessment device can be found in the limitations of the large-model-based geographic location risk assessment method described above, and will not be repeated here. Each module in the aforementioned large-model-based geographic location risk assessment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0053] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, Figure 7 This is a schematic diagram of a computer device according to an embodiment of the present invention. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side method for geolocation risk assessment based on a large model.

[0054] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 8 As shown, Figure 8This is another schematic diagram of a computer device according to an embodiment of the present invention. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of a large-model-based geographic location risk assessment method.

[0055] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the original address text of the target to be evaluated, and perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. The standardized address is geocoded to generate the latitude and longitude coordinates corresponding to the standardized address; The original address text, the standardized address, and the latitude and longitude coordinates are input into the pre-trained second language model for spatial semantic consistency verification. The model determines whether the latitude and longitude coordinates are consistent with the region described by the original address text, and outputs the credibility label of the latitude and longitude coordinates after the verification is passed. The multi-hazard risk layer is queried using the verified latitude and longitude coordinates to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The standardized address, the credibility label, and the multi-hazard risk quantification vector are input into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated.

[0056] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain the original address text of the target to be evaluated, and perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. The standardized address is geocoded to generate the latitude and longitude coordinates corresponding to the standardized address; The original address text, the standardized address, and the latitude and longitude coordinates are input into the pre-trained second language model for spatial semantic consistency verification. The model determines whether the latitude and longitude coordinates are consistent with the region described by the original address text, and outputs the credibility label of the latitude and longitude coordinates after the verification is passed. The multi-hazard risk layer is queried using the verified latitude and longitude coordinates to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The standardized address, the credibility label, and the multi-hazard risk quantification vector are input into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated.

[0057] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0058] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0059] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0060] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0061] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A geographical location risk assessment method based on a large model, characterized in that, include: Obtain the original address text of the target to be evaluated, and perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. The standardized address is geocoded to generate the latitude and longitude coordinates corresponding to the standardized address; The original address text, the standardized address, and the latitude and longitude coordinates are input into the pre-trained second language model for spatial semantic consistency verification. The model determines whether the latitude and longitude coordinates are consistent with the region described by the original address text, and outputs the credibility label of the latitude and longitude coordinates after the verification is passed. The multi-hazard risk layer is queried using the verified latitude and longitude coordinates to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The standardized address, the credibility label, and the multi-hazard risk quantification vector are input into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated.

2. The geographical location risk assessment method based on a large model according to claim 1, characterized in that, The process of obtaining the original address text of the target to be evaluated, and parsing the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated, includes: Semantic analysis is performed on the original address text to obtain semantic fragments corresponding to different address levels; Based on the geographical knowledge built into the pre-trained first language model, the informal place names of semantic fragments at each address level are mapped to the corresponding formal place names. Valid address elements are extracted from the semantic fragments after mapping and then reorganized according to a preset geographical hierarchy to generate a standardized address with a hierarchical structure.

3. The location risk assessment method based on a large model according to claim 2, characterized in that, The geographical knowledge built into the pre-trained first large language model maps informal place names of semantic fragments at each address level to corresponding formal place names, including: The semantic fragment is subjected to place name entity discrimination using the pre-trained first language model, and informal place name entities are identified. The identified informal place name entities and their corresponding address level labels are input into the pre-trained first language model. The geographical knowledge graph and contextual semantics built into the pre-trained first language model are used to generate at least one candidate formal place name that matches the semantics of the informal place name entities. Semantic fragments of the address level adjacent to the informal place name entity in the original address text are obtained. The semantic fragments of the adjacent address level are used as spatial constraints to verify the spatial inclusion relationship of the at least one candidate formal place name. Candidate formal place names with mismatched geographical ranges are filtered out, and the retained candidate formal place names are used as the formal place names corresponding to the informal place names after mapping.

4. The geographical location risk assessment method based on a large model according to claim 1, characterized in that, The original address text, the standardized address, and the latitude and longitude coordinates are input into a pre-trained second language model for spatial semantic consistency verification. This determines whether the latitude and longitude coordinates match the region described in the original address text. Upon successful verification, a credibility label for the latitude and longitude coordinates is output, including: The actual region of the latitude and longitude coordinates is determined by the pre-trained second language model, and the determined actual region is compared with the region described by the original address text. If the comparison results are inconsistent, the verification is deemed to have failed, and the latitude and longitude coordinates are marked as invalid coordinates. If the comparison results are consistent, the verification is deemed successful, and the address integrity of the original address text is evaluated using the pre-trained second language model. Based on the assessment results of the address completeness, output the corresponding level of confidence label for the latitude and longitude coordinates.

5. The location risk assessment method based on a large model according to claim 1, characterized in that, The step of querying the multi-hazard risk layer using verified latitude and longitude coordinates to obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates includes: Input the verified latitude and longitude coordinates into the preset geographic information service interface; Through the geographic information service interface, query the physical hazard parameters of each hazard corresponding to the latitude and longitude coordinates in the pre-constructed multi-hazard risk grid layer; Extract the physical hazard parameters of each hazard type from the query, and integrate the extracted physical hazard parameters of each hazard type into a structured multi-hazard risk quantification vector.

6. The location risk assessment method based on a large model according to claim 1, characterized in that, The standardized address, the credibility tag, and the multi-hazard risk quantification vector are input into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated, including: The standardized address, the credibility label, and the multi-hazard risk quantification vector are combined into a structured prompt word, and the structured prompt word is input into a pre-trained third language model; The pre-trained third language model determines the risk assessment strategy for the target to be assessed based on the credibility label, and performs inference based on the determined risk assessment strategy and the multi-hazard risk quantification vector to generate the risk rating result of the target to be assessed.

7. The location risk assessment method based on a large model according to claim 6, characterized in that, After inputting the standardized address, the credibility label, and the multi-hazard risk quantification vector into a pre-trained third language model for inference to generate a risk rating result for the target to be evaluated, the method further includes: The risk rating result, the standardized address, the credibility label, and the multi-hazard risk quantification vector are combined to generate prompt words for the report. The report generation prompts are input into the pre-trained fourth language model for reasoning to generate the risk interpretation text of the target to be evaluated, wherein the risk interpretation text includes the risk level explanation, the main disaster-causing factors and the corresponding risk response strategies. The risk rating results and the risk interpretation text are integrated to generate a risk assessment report for the target to be assessed, and the risk assessment report is then output.

8. A location risk assessment device based on a large model, characterized in that, include: The acquisition module is used to acquire the original address text of the target to be evaluated, and to perform address parsing on the original address text using a pre-trained first language model to generate the standardized address of the target to be evaluated. The encoding module is used to perform geocoding on the standardized address and generate the latitude and longitude coordinates corresponding to the standardized address. The determination module is used to input the original address text, the standardized address, and the latitude and longitude coordinates into the pre-trained second language model to perform spatial semantic consistency verification, determine whether the latitude and longitude coordinates are consistent with the region described by the original address text, and output the credibility label of the latitude and longitude coordinates after the verification is passed; The query module is used to query the multi-hazard risk layer using the verified latitude and longitude coordinates, and obtain the multi-hazard risk quantification vector corresponding to the latitude and longitude coordinates; The reasoning module is used to input the standardized address, the credibility label, and the multi-hazard risk quantification vector into a pre-trained third language model for reasoning, and generate a risk rating result for the target to be evaluated.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the large-model-based geographic location risk assessment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the large-model-based geographic location risk assessment method as described in any one of claims 1 to 7.

Images