Geological disaster knowledge abstract generation method and device based on knowledge graph and large model

Abstract

The invention discloses a geological disaster knowledge abstract generation method and device based on a knowledge graph and a large model. Comprising the following steps: constructing a geological disaster knowledge ontology system, converting multi-dimensional heterogeneous data into a triple through field mapping and named entity recognition, and establishing a knowledge graph with a spatial relationship; according to the county name retrieval map, statistical analysis is carried out, disaster space distribution hot spots are identified, the aggregation type is calculated and determined, and key villages and towns are positioned. A large model is used for generating a structured text, the relation between disasters and environmental factors is analyzed and revealed, and text consistency is ensured in combination with keyword matching and a dynamic correction mechanism. And integrating multiple rounds of results, and outputting county-level geological disaster knowledge abstracts. In addition, the invention further provides a device, a storage medium and computer equipment which are used for generating the geological disaster knowledge abstract. According to the method, the information integration efficiency can be effectively improved, the spatial correlation analysis is enhanced, the abstract standardization level is improved, and scientific support is provided for disaster prevention and intelligent decision making.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and geographical science, and in particular to a method and apparatus for generating geological disaster knowledge summaries based on knowledge graphs and large models. Background Technology

[0002] my country's geological environment is complex, influenced by the movements of the Pacific, Indian, and Eurasian plates, exacerbated by climate change and human engineering activities. This results in frequent, recurring geological disasters in my country. These disasters severely threaten people's lives and property, and undermine the sustainability of socio-economic development and the ecological environment. With the development of science and technology, a large amount of historical data on geological disasters has accumulated, primarily in the form of structured and unstructured text data. Structured data is widely used for downstream tasks such as geological disaster pattern analysis, stability assessment, and risk assessment, while unstructured data, mainly text reports containing a large amount of information, is primarily used as a basis for subsequent engineering management and administrative decisions. Therefore, adopting advanced technologies such as big data and artificial intelligence to improve the efficiency of geological disaster data utilization and enhance the technological level of disaster prevention and mitigation is of significant practical importance.

[0003] Geological hazards are geological phenomena formed by the combined effects of human factors and internal and external Earth forces, with landslides, debris flows, and collapses as typical examples, posing a serious threat to the natural environment and human life. With the development of computer technology, Geographic Information Systems (GIS), Global Positioning System (GPS), remote sensing, and other scientific technologies, geological hazard analysis has transformed from traditional field surveys to intelligent methods, forming a diversified approach including natural language processing for information extraction, numerical quantitative analysis, GIS spatial distribution research, and artificial intelligence question answering. However, existing analytical methods are mostly conducted independently, lacking a comprehensive application mechanism, and are limited by high data quality requirements and the need for extensive human and material resources for updates. This makes it difficult to meet the rapidly growing demand from government departments for diverse and complex geological hazard data, especially in the field of geological hazard emergency investigation, where the need for efficient integration of multi-source heterogeneous data such as hazard overview, causal analysis, and emergency measures, and the realization of intelligent analysis, is increasingly urgent.

[0004] The main drawbacks of existing technologies are: 1. Insufficient data integration capabilities for geological disasters. An efficient multi-source data integration method has not been developed, making it impossible to efficiently extract key information such as the disaster-prone environment and disaster characteristics from geological disaster emergency investigation reports. Unstructured data cannot be combined with structured data, resulting in the idleness of potentially valuable knowledge. 2. When generating text reports of multi-source heterogeneous geological disaster data, existing technologies mainly rely on manual writing or template generation. The former is costly, while the latter suffers from insufficient flexibility, rigid structure, and poor contextual coherence. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention addresses the need for integrating multi-source heterogeneous data in geological disaster emergency investigations. It constructs a knowledge graph ontology and corresponding triplet datasets to build a knowledge graph for geological disaster emergency investigation analysis. At the county level, it performs quantitative and qualitative comprehensive analysis of regional disaster data within the knowledge graph, importing the analysis results into the knowledge graph. Based on the textual characteristics of geological disasters, it integrates multimodal information using the knowledge graph and guides a large language model with prompts for engineering design, generating accurate and intelligent knowledge summaries constrained by the knowledge graph, thereby improving the accuracy and completeness of geological disaster emergency investigation knowledge summaries.

[0006] This application provides a method for generating geological hazard knowledge summaries based on knowledge graphs and large models. The technical solution adopted by the method includes the following steps: S1. Construct a hierarchical knowledge graph covering four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors. ; S2. Conduct a comprehensive quantitative and qualitative analysis of regional disaster data in the knowledge graph at the county level to obtain the analysis result triplets. And update the knowledge graph to ; S3. Guiding the large language model through prompting engineering. Generate knowledge graph Constrained, complete and well-structured natural language knowledge summaries .

[0007] Further, step S1 includes the following sub-steps: S1-1. Constructing a three-tiered geological disaster emergency investigation ontology system: "concept-object-attribute". ,but:

[0008] in: The first level is a collection of core concepts related to geological hazards. These correspond to the disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors, respectively. The second level is a collection of objects. ,satisfy , indicating the first Class concept The j-th specific object below; The third level is a collection of attributes. ,satisfy , indicating the first Class concept The j-th concrete object The kth specific attribute; S1-2, For structured datasets fields in Through field mapping functions Generate a set of triples in the form of "disaster site name - attribute - attribute value". :

[0009] in: Indicates the first Name of each disaster site, Representation field The attribute corresponding to the j-th disaster point, Representing records In the field The value in To record the total number; S1-3, For unstructured text Entity recognition function is used Extracting disaster entities Then extract the function through the relation. Construct a set of triples :

[0010] in: For the first A disaster name entity, For entities Relationship types, For entities The associated value, The total number of entities; S1-4, Introducing Administrative Division Levels , Representing province, city, county, and township respectively, using geographic coding functions Geographical location of disaster sites Mapped to township-level nodes Generate a set of spatially correlated triples :

[0011] in: Indicates the first Township-level nodes, Indicates disaster point Geographic coordinates Indicates disaster point The set of structured and unstructured triples, The total number of disaster sites; S1-5, Set the triplet set Import graph database Constructing a knowledge graph for geological disaster emergency investigation Based on the definitions and descriptions already provided in steps S1-1 to S1-5, nodes, edges, and attribute sets can be represented as follows: Node set ; edge set ; Attribute Collection ; Further, step S2 includes the following sub-steps: S2-1, Enter the name of the county-level administrative region Retrieval function through knowledge graph Extract disaster point data set ,in Indicates the first For each disaster point, construct a regional submap. ,satisfy: Node set ,in for The j-th township node under its jurisdiction; edge set ,in Indicates administrative division relationships. Indicates the attribution of disaster sites; Attribute Collection It includes both structured and unstructured attributes of disaster sites; S2-2, Set of Disaster Point Attributes Numerical properties in , Define a statistical function to represent the m-th numerical attribute of the n-th disaster point. :

[0012] in: For attributes The mean, For attributes Frequency; For category In attributes The distribution ratio in The total number of disaster sites. For attributes The number of valid values, For category The number of times it appears; S2-3, Define the kernel density estimation function ,in This is a set of geographical coordinates of the disaster site. For the bandwidth parameter, calculate each point density value : in For the improved kernel function, satisfying ; Through the regional center point Location of disaster site The relative orientational relationships are defined by the spatial distribution characteristics "East", "West", "South", and "North". Then it can be expressed as: in This is the azimuth threshold; S2-4. Calculate the degree of disaster clustering in each township. Based on disaster point density Spatial autocorrelation , can be represented as: in These are the thresholds for density and autocorrelation, respectively; weighted by a weighting function. Calculate the weight of the impact on the county level. : in These are the weighting coefficients. For townships Risk level; S2-5, Statistical results Spatial distribution characteristics and clustering type Convert to a set of triples :

[0013] in To the townships with the greatest impact, For relation type; Import graph database Update the knowledge graph .

[0014] Furthermore, step S3 includes the following sub-steps: S3-1. Design a multi-round prompt word template and define the prompt word set. ,in: The task instruction requires generating a "County-level Geological Disaster Knowledge Summary Report"; Example instructions, providing a sample structured summary; : Data command, specifying the input as the regional analysis results ; Verification command, defines the consistency threshold. ; S3-2, Set the regional analysis results ,in Indicates the first The core concepts of the disaster include analytical data on disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors; Let the ontology hierarchical sorting method be: Then they can be sorted according to the ontology hierarchy:

[0015] Input large language model Generate an initial summary:

[0016] S3-3, Based on the results of regional analysis Using large models to identify disaster development characteristics Environmental factors The correlation, and generate descriptive text. Let the correlation strength be... Then it can be expressed as:

[0017]

[0018] in These are preset parameters used to balance semantic similarity and factor importance; To utilize large models to identify semantic similarity, To utilize large models to identify the importance of factors; S3-4. Utilize large-scale models to calculate and generate text. With the original triplet Coverage:

[0019] like If the preset threshold is met, a correction command will be triggered. Guide the model to adjust the text; S3-5, Repeat S3-2 to S3-4, and generate the text sets from multiple rounds. Merge in logical order and input the large model. In the end, the final output is the final summary. :

[0020] in, This is a county-level geological disaster knowledge summary report that conforms to professional standards, is complete in content, and is logically coherent.

[0021] This application also provides a geological hazard knowledge summary generation device based on knowledge graphs and large models, including: The building blocks are used to construct a hierarchical knowledge graph covering four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors. ; The analysis unit is used to perform quantitative and qualitative comprehensive analysis of regional disaster data in knowledge graphs at the county level, and obtain analysis result triplets. And update the knowledge graph to ; Generative units are used to guide large language models through prompting engineering. Generate knowledge graph Constrained, complete and well-structured natural language knowledge summaries .

[0022] This application also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of any of the methods described above.

[0023] This application also 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 program to implement the steps of any of the methods described above.

[0024] The beneficial effects of this invention are as follows: 1. Efficient integration and knowledge structuring of multi-source data: Through the construction of a three-level ontology system of "concept-object-attribute" and knowledge graph, structured and unstructured data are uniformly transformed into triples with spatial relationships, realizing hierarchical storage and rapid retrieval of geological disaster information, significantly improving data processing efficiency and knowledge association capabilities.

[0025] 2. Precise spatial distribution characteristics: Combining numerical analysis and spatial correlation analysis, the system dynamically identifies disaster hotspots and key affected townships, providing visualized and quantitative spatial decision support for disaster risk assessment and emergency resource allocation.

[0026] 3. Intelligent summary generation and semantic association mining: Utilize a large language model to generate structured text from the knowledge graph analysis results, and combine semantic understanding capabilities to reveal the potential correlation between disaster characteristics and environmental factors, outputting a summary report that is complete, logically coherent, and conforms to the professional standards for geological disasters.

[0027] 4. Intelligent support for disaster prevention and control decision-making: Through the collaborative application of knowledge graphs and large models, the entire process from data collection to knowledge generation is automated, promoting the transformation of geological disaster emergency investigation towards standardization and intelligence, and providing scientific basis for disaster early warning, risk assessment and governance planning. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall process of this method; Figure 2 This is a conceptual knowledge system for emergency investigation and analysis of geological disasters. Detailed Implementation

[0029] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0030] like Figure 1 As shown in the technical roadmap, the geological disaster knowledge summary generation method based on knowledge graphs and large models of this invention includes the following steps: constructing a hierarchical knowledge graph covering four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and inducing factors; performing quantitative and qualitative comprehensive analysis on regional disaster data in the knowledge graph at the county level, obtaining analysis result triples and updating the knowledge graph; and generating natural language knowledge summaries that are constrained by the knowledge graph, complete in content, and structurally standardized through prompting engineering guided by a large language model.

[0031] This invention takes Province F as the research area and uses multi-source heterogeneous data, including the Province F Geological Disaster Comprehensive Prevention and Control Database and Province F Geological Disaster Emergency Investigation Reports, as a foundation. Based on the rationality and completeness of the data, 4960 geological disaster point feature data points were selected, and 219 geological disaster emergency investigation reports were obtained. Entities from the knowledge base were extracted and combined with structured data into triples, resulting in 2768 entities (nodes) and 10758 relationships (edges). Finally, 69 geological disaster emergency investigation knowledge summaries were generated. The specific implementation process is as follows: S11. Construct a three-tiered ontology system for geological disaster emergency investigation, consisting of "concept-object-attribute," as shown in Table 1. In the geological disaster emergency investigation of Province F, an ontology system is constructed containing four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors. Define the first-level concept set. ,in The disaster-prone environment (such as topography, strata lithology, and hydrogeology). These are the basic characteristics (such as disaster type, scale, and time of occurrence). The characteristics of the disaster (such as displacement, deformation, and degree of damage). Triggering factors (such as rainfall, earthquakes, and human activities). Second-level object set. Includes typical disaster targets in Province F, such as (Wuyishan area, F Province) (Landslide disasters), etc. Third-level attribute set Includes attribute values, such as (altitude) (Landslide scale), etc.

[0032] Table 1. Geological Disaster Emergency Investigation and Analysis: Concept-Object-Attribute Layer

[0033] The S12 and F Province geological disaster database contains 2,300 structured records, which are generated into triples through field mapping. For example, for the disaster point record "FZ City Gushan Landslide", the attribute corresponding to the "Disaster Type" field is extracted. The attribute value "slippery" generates a triplet. Traversing all records, a total of 7523 triples were generated.

[0034] The S13 and F Province geological disaster emergency investigation reports contain 250 unstructured texts. Sequence labeling algorithms were used to extract disaster entities. For example, from the report "A large landslide occurred in Gushan area of ​​FZ City, affected by heavy rainfall," the disaster entity "FZ City Gushan landslide" was extracted as the head entity, with the relation "affected by..." and the associated value "heavy rainfall," generating a triplet. A total of 375 disaster entity triples were extracted.

[0035] S14. Establish the administrative division level of Province F. (Province, city, county, township) This refers to the geographical location of the disaster site. Mapped to township-level nodes. For example, the geographical location of the Gushan landslide in FZ City. Through geocoding functions Mapped to township-level nodes in GS Street, FZ City Generate spatially related triples Total processing A set of three disaster points.

[0036] S15. Import the triple set into the Neo4j graph database to construct a knowledge graph for geological disaster emergency investigation in Province F. .

[0037] S21. Input the county-level administrative region name "F Province, FZ City, G District", and retrieve it using the knowledge graph retrieval function. Extract disaster point data set (A total of 50 disaster sites). Construct regional sub-maps. Node set edge set Attribute set Includes disaster location attributes.

[0038] S22, Set of disaster point attributes in area G Statistical analysis is performed on numerical attributes (such as slope height, slope gradient, and sliding surface depth). The mean, frequency, and category distribution ratio are calculated. For example, the mean slope height... meters, slope frequency Distribution ratio of slope types .

[0039] S23, Set of geographical coordinates of disaster points in area G Using kernel density estimation function Calculate density value Regional center point Spatial distribution characteristics are defined through locational relationships. For example, disaster points. Compared to Located in the west and north, with a spatial distribution characteristic of "northwest"; disaster point Located in the east and south, its spatial distribution is characterized by a "southeast" orientation.

[0040] S24. Calculate the degree of disaster clustering in each township of G area, based on the density of disaster points. Spatial autocorrelation For example, GS Street (density), (Autocorrelation), satisfies and The clustering type is "high-high". This is achieved through a weighted function. Calculate the GS street weights It was determined to be the township with the greatest impact on the county level.

[0041] S25. Convert statistical results, spatial distribution characteristics, and clustering types into sets of triples. For example, generating triples. , , .Will Import graph database and update knowledge graph .

[0042] S31, Design Tips Set .in As a "geological disaster emergency analysis expert", To "generate a county-level geological disaster knowledge summary report", Provides a structured summary sample (including analysis templates for each unit, such as disaster initiation stage, basic characteristics, disaster characteristics, triggering factors, and comprehensive analysis). Specify the input as the regional analysis results. Define consistency threshold .

[0043] S32. Sort the regional analysis results of District G, FZ City, F Province according to the ontology level, input them into the large language model, and generate an initial summary. For example, the disaster-prone environment will include information such as topography, stratigraphy, and hydrogeology, while the basic features will include information such as disaster type, scale, and occurrence time.

[0044] S33. Analyzing Disaster Development Characteristics Using Large Language Models (e.g., landslide size, timing) and environmental factors The correlation between landslide size and topographic slope (e.g., topography, lithology, hydrogeology). For example, the correlation strength between landslide size and topographic slope is... This indicates a strong correlation between the two. The analysis results of the large model will be preserved in the contextual semantics.

[0045] S34. Calculate the generated text and the original triples using a large language model. coverage .like This triggers a correction instruction. For example, if the generated text lacks the information "heavy rainfall triggered landslides," the coverage rate may be affected. Trigger correction command This guides the model to adjust the text.

[0046] S35. The text generated and verified in multiple rounds... Merge the results in logical order and output the final summary. As shown below: I. Environment for Pregnancy Located in the lower reaches of the Minjiang River, the G District of FZ City, F Province, has a terrain that is mainly a transitional zone between hills and plains. The terrain in the area is significantly undulating, with some areas having a slope greater than 30°, which provides topographical conditions for disasters such as landslides and debris flows.

[0047] II. Basic Characteristics The main geological hazards in Area G are landslides and small debris flows, with most hazards being small to medium in scale (landslide volume generally less than 10,000 cubic meters, debris flow rate less than 1,000 cubic meters per second). The density of hazard points is relatively high, mainly concentrated in hilly areas such as GS and AT subdistricts, exhibiting a northwest-southeast clustering pattern. The timing of these hazards shows a significant seasonality, with over 70% occurring during the rainy season from May to September, closely related to heavy rainfall events.

[0048] III. Disaster Characteristics Development patterns: Landslides mostly slide along the interface between bedrock and overburden, with a fast deformation rate and strong suddenness; debris flows are mostly triggered by rainstorms, flow rapidly along gullies, and have strong destructive power.

[0049] Activity status: Currently, about 60% of the disaster sites in the area are in an active state, mainly manifested as deformation signs such as cracking at the rear edge of the landslide and heaving of the soil at the front edge.

[0050] Severity of damage: The disaster poses a direct threat to residential areas, main roads and infrastructure. In the past five years, there have been 12 incidents of house damage caused by landslides, and an average of 3-5 incidents of mudslides blocking roads per year.

[0051] IV. Triggering Factors Natural factors: Rainfall: The average annual rainfall is 1400 mm. Extreme rainstorm events (daily rainfall > 150 mm) are the main triggering factors, accounting for 85% of the disaster causes.

[0052] Earthquake: The regional earthquake intensity is VII, and historical strong earthquakes have induced local soil loosening.

[0053] Human factors: Construction projects: Engineering activities such as mountain excavation and road slope cutting have damaged the stability of the original landform, and newly added disaster hazard points account for 20% of the total.

[0054] Improper drainage: Inadequate drainage systems in some areas lead to rainwater retention, causing soil saturation and instability, and exacerbating the risk of disasters.

[0055] V. Comprehensive Analysis The overall geological disaster risk level in District G is medium to high, with GS Street and AT Street designated as high-risk areas. It is recommended to strengthen monitoring and early warning systems during the rainy season, optimize drainage system design, and implement engineering remediation measures (such as anchor reinforcement and construction of intercepting ditches) for key potential hazard points to reduce the threat of disasters to urban safety.

Claims

1. A method for generating geological hazard knowledge summaries based on knowledge graphs and large models, characterized in that, Includes the following steps: S1. Construct a hierarchical knowledge graph covering four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors. ; S2. Conduct a comprehensive quantitative and qualitative analysis of regional disaster data in the knowledge graph at the county level to obtain the analysis result triplets. And update the knowledge graph to ; S3. Guiding the large language model through prompting engineering. Generate knowledge graph Constrained, complete and well-structured natural language knowledge summaries .

2. The method for generating geological disaster knowledge summaries based on knowledge graphs and large models according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1-1. Constructing a three-tiered geological disaster emergency investigation ontology system: "concept-object-attribute". ,but: in: The first level is a collection of core concepts related to geological hazards. These correspond to the disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors, respectively. The second level is a collection of objects. ,satisfy , indicating the first Class concept The j-th specific object below; The third level is a collection of attributes. ,satisfy , indicating the first Class concept The j-th concrete object The kth specific attribute; S1-2, For structured datasets fields in Through field mapping functions Generate a set of triples in the form of "disaster site name - attribute - attribute value". : in: Indicates the first Name of each disaster site, Representation field The attribute corresponding to the j-th disaster point, Representing records In the field The value in To record the total number; S1-3, For unstructured text Entity recognition function is used Extracting disaster entities Then extract the function through the relation. Construct a set of triples : in: For the first A disaster name entity, For entities Relationship types, For entities The associated value, The total number of entities; S1-4, Introducing Administrative Division Levels , Representing province, city, county, and township respectively, using geographic coding functions Geographical location of disaster sites Mapped to township-level nodes Generate a set of spatially correlated triples : in: Indicates the first Township-level nodes, Indicates disaster point Geographic coordinates Indicates disaster point The set of structured and unstructured triples, The total number of disaster sites; S1-5, Set the triplet set Import graph database Constructing a knowledge graph for geological disaster emergency investigation Based on the definitions and descriptions already provided in steps S1-1 to S1-5, nodes, edges, and attribute sets can be represented as follows: Node set ; edge set ; Attribute Collection .

3. A method for generating geological disaster knowledge summaries based on knowledge graphs and large models according to claim 1 or 2, characterized in that, Step S2 includes the following sub-steps: S2-1, Enter the name of the county-level administrative region Retrieval function through knowledge graph Extract disaster point data set ,in Indicates the first For each disaster point, construct a regional submap. ,satisfy: Node set ,in for The j-th township node under its jurisdiction; edge set ,in Indicates administrative division relationships. Indicates the attribution of disaster sites; Attribute Collection It includes both structured and unstructured attributes of disaster sites; S2-2, Set of Disaster Point Attributes Numerical properties in , Define a statistical function to represent the m-th numerical attribute of the n-th disaster point. : in: For attributes The mean, For attributes Frequency; For category In attributes The distribution ratio in The total number of disaster sites. For attributes The number of valid values, For category The number of times it appears; S2-3, Define the kernel density estimation function ,in This is a set of geographical coordinates of the disaster site. For the bandwidth parameter, calculate each point density value : in For the improved kernel function, satisfying ; Through the regional center point Location of disaster site The relative directional relationships are defined by spatial distribution characteristics "east", "west", "south", and "north". Then it can be expressed as: in This is the azimuth threshold; S2-4. Calculate the degree of disaster clustering in each township. Based on disaster point density Spatial autocorrelation , can be represented as: in These are the thresholds for density and autocorrelation, respectively; weighted by a weighting function. Calculate the weight of the impact on the county level. : in These are the weighting coefficients. For townships Risk level; S2-5, Statistical results Spatial distribution characteristics and clustering type Convert to a set of triples : in To the townships with the greatest impact, For relation type; Import graph database Update the knowledge graph .

4. A method for generating geological disaster knowledge summaries based on knowledge graphs and large models according to claim 1 or 2, characterized in that, Step S3 includes the following sub-steps: S3-1. Design a multi-round prompt word template and define the prompt word set. ,in: : Role command, specifying the model as "Geological Disaster Emergency Analysis Expert"; The task instruction requires generating a "County-level Geological Disaster Knowledge Summary Report"; Example instructions, providing a sample structured summary; : Data command, specifying the input as the regional analysis results ; Verification command, defines the consistency threshold. ; S3-2, Set the regional analysis results ,in Indicates the first The core concepts of the disaster include analytical data on disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors; Let the ontology hierarchical sorting method be: Then they can be sorted according to the ontology hierarchy: Input large language model Generate an initial summary: S3-3, Based on the results of regional analysis Using large models to identify disaster development characteristics Environmental factors The correlation, and generate descriptive text. Let the correlation strength be... Then it can be expressed as: in These are preset parameters used to balance semantic similarity and factor importance; To utilize large models to identify semantic similarity, To utilize large models to identify the importance of factors; S3-4. Utilize large-scale models to calculate and generate text. With the original triplet Coverage: like If the preset threshold is met, a correction command will be triggered. Guide the model to adjust the text; S3-5, Repeat S3-2 to S3-4, and generate the text sets from multiple rounds. Merge in logical order and input the large model. In the end, the final output is the final summary. : in, This is a county-level geological disaster knowledge summary report that conforms to professional standards, is complete in content, and is logically coherent.

5. A geological hazard knowledge summary generation device based on knowledge graphs and large models, characterized in that, include: The building blocks are used to construct a hierarchical knowledge graph covering four core concepts: disaster-prone environment, basic characteristics, disaster characteristics, and triggering factors. ; The analysis unit is used to perform quantitative and qualitative comprehensive analysis of regional disaster data in knowledge graphs at the county level, and obtain analysis result triplets. And update the knowledge graph to ; Generative units are used to guide large language models through prompting engineering. Generate knowledge graph Constrained, complete and well-structured natural language knowledge summaries .

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method described in any one of claims 1-4.

7. 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 program, it implements the steps of the method according to any one of claims 1-4.

Images