Typhoon disaster knowledge graph generation method and system based on large language model

Abstract

The invention discloses a typhoon disaster knowledge graph generation method and system based on a large language model, and relates to the technical field of knowledge graph construction, and the method comprises the steps: obtaining typhoon disaster data, carrying out the cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, and obtaining a typhoon disaster knowledge graph. The cognition and expression of the typhoon disaster knowledge comprises a typhoon evolution process and a typhoon disaster effect; based on the cognition and expression result of typhoon disaster knowledge, a knowledge extraction instruction is generated, and the typhoon disaster data comprises the wind speed, the wind direction, the central air pressure, the movement track, the sea wave height, the rainfall and the disaster situation; the knowledge extraction instruction is input into a pre-established large language model LLMs for training, and a TyphoonKGent is obtained; and inputting the typhoon disaster data into the Typhoon KGent to obtain the structured knowledge graph extracted by the Typhoon KGent, thereby realizing the construction of evaluating the typhoon disaster knowledge graph based on a data and manual combination method.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph construction, and specifically to a method and system for generating a typhoon disaster knowledge graph based on a large language model. Background Art

[0002] With the rapid development of technologies such as sensors, the Internet of Things, and geographic information, people have accumulated a large amount of data on typhoon disasters, but these data are usually fragmented and isolated, making it difficult to obtain the required disaster knowledge from them. Knowledge graph is a large-scale semantic network that uses graph models to describe knowledge and model the relationship between all things in the world. It can represent complex knowledge in the real world in computers in a way that is closer to human thinking. Applying knowledge graph technology to the field of typhoon disasters can not only promote the systematic integration of meteorological disaster data and build a geographic knowledge graph (GeoKG) for the typhoon field, but also use knowledge reasoning capabilities to support more accurate and effective disaster prevention and mitigation services. However, the method of using machine learning or deep learning methods to build GeoKG still has many shortcomings, which can be summarized as follows: 1) Limitations of rule templates. In the GIS field, rule matching is performed with the help of place name dictionaries. Although this method is simple to operate and has high accuracy, it has poor flexibility and is difficult to adapt to complex and changing real-world scenarios. For example, in typhoon disaster management, new observation data or meteorological models continue to emerge, and the rules and templates formulated by experts have limited coverage, resulting in reduced prediction accuracy. 2) Requirements for large-scale corpus annotation. Training machine learning models usually requires a large amount of annotated data, which increases the time and cost of data preparation, and the quality of annotated data is easily affected by subjective factors, resulting in increased data noise and affecting the performance of extraction. 3) Insufficient generalization ability. Traditional methods rely on labeled data sets in specific fields. When faced with unseen data or new scenarios, data annotation needs to be re-annotated, and model performance is significantly reduced, limiting the scope of application of knowledge graphs, especially in rapidly changing environments, such as rainstorm disaster monitoring and response. 4) Complex construction process. The construction of knowledge graphs involves multiple steps such as entity recognition, relationship extraction, and knowledge fusion. Each step requires independent tools or algorithms, which increases the possibility of errors, and the connection problems between each link may also affect the quality of the final knowledge graph. Therefore, current research urgently needs to explore a new GeoKG construction method to simplify the process and achieve end-to-end construction. Summary of the invention

[0003] In order to solve the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a method and system for generating a typhoon disaster knowledge graph based on a large language model.

[0004] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a method for generating a typhoon disaster knowledge graph based on a large language model, the method comprising the following steps: Acquire typhoon disaster data, and carry out cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and typhoon disaster effects; generate knowledge extraction instructions based on the cognition and expression results of typhoon disaster knowledge, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster conditions; The knowledge extraction instructions are input into the pre-established large language model LLMs for training to obtain TyphoonKGent; the typhoon disaster data is input into TyphoonKGent, and the structured knowledge graph extracted by TyphoonKGent is output.

[0005] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the typhoon disaster knowledge-related data characterizes the cognition and expression process of typhoon disaster knowledge through information aggregation of the entity layer, stage layer, state layer and attribute layer.

[0006] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the entity layer is used to express component objects of typhoon disasters in the real world; The stage layer is used to express the specific changes, actions or records of the typhoon object in time and space, and is divided into four stages: formation stage, development stage, continuation stage and decay stage; The state layer is used to decompose the typhoon object in the entity layer into different states. The decomposition is achieved by analyzing the key attributes of the object, including the intensity, speed and landing place of the typhoon; the change of the object is represented by a set of states, and the development process of the event itself is expressed by the different states of the object; The attribute layer is used to define the timestamp, geographic location, specific attributes and behavior mode of the typhoon object in the entity layer.

[0007] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the generating of knowledge extraction instructions is based on one or more of spatiotemporal knowledge completion, coreference resolution, entity alignment, and life cycle.

[0008] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: the spatiotemporal knowledge completion is used to parse the LLMs and complete the ambiguous spatiotemporal information in the text; The coreference resolution is used to automatically identify and parse pronouns and demonstrative expressions in texts for LLMs; The entity alignment is used to automatically identify and merge multiple representations of the same entity from different sources in LLMs; The life cycle determination is used to automatically generate corresponding time points and state changes for LLMs according to different stages of typhoon disaster evolution.

[0009] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the process of inputting the knowledge extraction instruction into the pre-established large language model LLMs for training includes: Low-rank adapted weight matrices are used for low-rank approximation training of large language models (LLMs).

[0010] In a second aspect, in order to achieve the above-mentioned purpose, the present invention discloses a typhoon disaster knowledge graph generation system based on a large language model, comprising: An instruction generation module is used to obtain typhoon disaster data, and to carry out cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and typhoon disaster effects; based on the cognition and expression results of typhoon disaster knowledge, a knowledge extraction instruction is generated, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster situation; The graph generation module is used to input knowledge extraction instructions into the pre-established large language model LLMs for training to obtain TyphoonKGent; typhoon disaster data is input into TyphoonKGent, and the structured knowledge graph extracted by TyphoonKGent is output.

[0011] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, a typhoon disaster knowledge graph generation method based on a large language model as described above is adopted.

[0012] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a computer-readable storage medium is disclosed, in which a computer program is stored. When the computer program is loaded and executed by a processor, the method for generating a typhoon disaster knowledge graph based on a large language model as described above is adopted.

[0013] Beneficial effects of the present invention: The present invention realizes the automatic conversion from unstructured text to structured knowledge graph by using pre-trained LLMs for fine-tuning and combining with specific knowledge extraction instructions. This method not only greatly simplifies the complex steps in the traditional knowledge graph construction process, but also can adapt well to changing data and scenarios. In addition, due to the use of pre-trained large language models and a small amount of fine-tuning to achieve specific tasks, the demand for large-scale annotated data sets is greatly reduced, and the time and cost of data preparation are reduced. Ultimately, the typhoon disaster knowledge graph constructed by the present invention can support more accurate and effective disaster prevention and mitigation services. By conducting a comprehensive and systematic analysis of typhoon disasters, it helps decision makers take timely measures to reduce the impact of disasters, providing strong support for typhoon disaster management. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work. Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a hierarchical expression framework of the typhoon disaster evolution process constructed by the present invention; Figure 3 It is a hierarchical expression framework of typhoon disaster effects constructed by the present invention; Figure 4 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Embodiment 1: like Figure 1 As shown, a method for generating a typhoon disaster knowledge graph based on a large language model comprises the following steps: S101: Acquire typhoon disaster data, and carry out cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and typhoon disaster effects; generate knowledge extraction instructions based on the cognition and expression results of typhoon disaster knowledge, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster conditions; like Figure 2 As shown in Figure 1, the hierarchical expression of the evolution process of typhoon disasters covers the entity layer, stage layer, state layer and attribute layer. Among them, the entity layer contains the specific disaster objects involved in the typhoon disaster knowledge graph, such as Typhoon Mangkhut, Typhoon Kanu, Typhoon Dusurui, etc. The stage layer is used to express the specific changes, actions or records of typhoon objects in time and space, including the formation period, development period, continuation period and decline period. 1) Formation period: From the initial formation of low-pressure circulation to the maximum average wind speed near the center reaching level eight, it usually takes about two days, three or four days at the slowest, and only a few hours at the fastest, showing characteristics such as tropical disturbances and low-pressure areas; 2) Development period: The typhoon continues to develop in intensity until the central air pressure reaches the lowest value and the wind speed reaches the maximum value, showing characteristics such as tropical depression and increased wind speed; 3) Sustaining period: The typhoon no longer develops in intensity, the air pressure no longer decreases, and the wind speed no longer increases, but the range gradually expands, showing heavy rain and maximum sustained wind speed; 4) Decay period: The wind speed in the center of the typhoon decreases, the air pressure increases, or it enters the mid-latitudes and transforms into a temperate cyclone, showing characteristics such as residual low pressure and wind speed attenuation. The four evolutionary stages of typhoon disasters together constitute its complete life cycle. In each stage, the typhoon will contain multiple states, such as state 1, state 2, state 3, etc. Their collection is the state layer, which can be used to represent the state evolution process of the typhoon in this stage. The attribute layer goes into more detail and defines the specific characteristics of typhoon disasters in different states, including timestamps, action patterns, geographic locations, specific attributes, etc. These are key identifiers for distinguishing and identifying various types of disaster information. Through this hierarchical knowledge expression strategy, the aim is to store typhoon texts in the form of structured triples in TyphoonKG.

[0017] like Figure 3As shown in the figure, for the text of disaster effects, the present invention proposes a knowledge aggregation model of "disaster-bearing body category layer-disaster-affected instance layer-attribute layer" from the perspective of disaster science and geography. According to the preset standards, the disaster-bearing body category layer is divided into three categories: "people", "property", and "resources and environment". Among them, the "people" disaster-bearing body includes categories such as missing persons, disaster-affected persons, and dead persons; the "property" disaster-bearing body covers infrastructure, housing and buildings, industrial assets, public service facilities, agriculture, forestry, animal husbandry, and fishery, and service industries; and the "resources and environment" disaster-bearing body covers categories such as land resources, ecological and environmental resources, water resources, biological resources, and mineral resources. Each disaster-bearing body category can derive multiple disaster-affected instances, and each disaster-affected instance has unique attributes different from other instances. For example, the public service facility disaster-bearing body includes "primary schools, middle schools, universities, etc.", "broadcast studios, printing shops, etc.", "libraries, cultural corridors, etc.", "hospitals, clinics, etc.", "stadiums, stadiums, etc.", "nursing homes, welfare institutions, etc.", and "airports, bus stations, highways, etc.". Among them, the disaster-affected instance "hospital, clinic" contains attributes such as distance, number of doctors, number of injured people, and location. These disaster instances and attributes together constitute a comprehensive knowledge representation of the disaster effects.

[0018] Secondly, the generation of typhoon disaster knowledge extraction instructions is completed, including the adaptation of LLMs to the typhoon disaster knowledge graph construction task, improving the ability of LLMs to construct the typhoon disaster knowledge graph, and generating knowledge extraction instructions.

[0019] (1) Role-playing: Role-playing, as a behavioral simulation technology, is widely used to experience and understand the behavior patterns, thinking methods and communication skills of a specific role. In the TyphoonKG construction task, the knowledge contained in LLMs is usually unstructured, while the knowledge graph requires explicit and structured knowledge representation. Through the role-playing strategy, the present invention can effectively connect these two different data forms, thereby promoting the application of LLMs in TyphoonKG construction, as shown in the first row of Table 1. This role positioning not only gives LLMs clear task goals, but also enables them to actively process implicit information, infer causal relationships, and perform well in dealing with uncertain data in the complex knowledge extraction process. This strategy is far beyond the reach of traditional methods, because traditional methods usually rely on static rules and cannot adapt to changing text content and complex contextual requirements.

[0020] (2) Output format: The knowledge graph is a structured knowledge network in which triples are the smallest constituent units. The basic form of a triple is<h, t, r> , where h represents the subject entity, t represents the object entity, and r represents the relationship between the two. However, considering that the typhoon disaster texts processed by the present invention usually present multidimensional heterogeneity, covering various information such as the origin, development and disaster impact of typhoons. These descriptions involve multiple dimensions, such as spatial location, time process and disaster effects. Therefore, in order to fully express a piece of typhoon knowledge, the format output shown in the second row of Table 1 is defined. When facing different typhoon texts, different format outputs need to be selected.

[0021] In the TyphoonKG construction task, the four major links of spatiotemporal knowledge completion, coreference resolution, entity alignment and lifecycle determination are intertwined and constitute the core of knowledge graph construction. In order to give full play to the capabilities of LLMs, the present invention has carried out detailed instruction design for these complex tasks, so that it can gradually replace the shortcomings of traditional methods in the automation process.

[0022] (1) Spatiotemporal knowledge completion and construction of a multi-dimensional spatiotemporal framework: Spatiotemporal knowledge completion is a key link in improving the accuracy of spatiotemporal dimensions in TyphoonKG. Traditional methods often rely on manual annotation and rule matching when processing spatiotemporal information. This method is inefficient when facing complex time and space descriptions, and it is difficult to ensure consistency and accuracy. By introducing LLMs, the present invention designs special instructions, as shown in the third row of Table 1, so that LLMs can parse and complete the ambiguous spatiotemporal information in the text. For example, when encountering time expressions such as "the next day" or "two days later" in the text, the model can automatically infer the corresponding specific date and be consistent with the logic of the context. Similarly, for geographical descriptions such as "coastal areas", the model can refine the geographical location based on contextual information, thereby constructing a more complete and detailed spatiotemporal framework. This method not only improves the accuracy of spatiotemporal reasoning, but also lays a solid foundation for further causal analysis and event association. Traditional methods are often unable to handle such complex spatiotemporal information in multiple dimensions and levels, which is one of the reasons why LLMs show advantages in the construction of TyphoonKG.

[0023] (2) Coreference resolution, parsing complex reference relationships: Coreference resolution is a key task in text understanding, especially in the description of geographical events. Traditional coreference resolution methods usually rely on preset rules or limited context windows and cannot handle complex reference relationships. In the TyphoonKG task, the present invention designs a special coreference resolution instruction, as shown in the fourth row of Table 1, which enables LLMs to automatically identify and parse pronouns and indicative expressions in the text and accurately map them to specific geographical entities or locations. For example, the model can recognize that the referent of "it" is the tropical disturbance mentioned above, while "they" refers to the "Joint Typhoon Warning Center". Through this process, the correlation between different entities in the knowledge graph is significantly enhanced, which improves the overall consistency and coherence of the graph. In contrast, traditional methods are prone to parsing errors or omissions when processing complex texts, making it difficult to ensure the accuracy and consistency of the results.

[0024] (3) Entity alignment, solving the consistency problem of heterogeneous data sources: The complexity of the entity alignment process is mainly reflected in the integration of heterogeneous data sources. Different sources may use different naming methods or description methods for the same entity. Traditional entity alignment methods mostly rely on manual rules or preset matching algorithms, which is time-consuming and inefficient when processing a large amount of heterogeneous data, and often cannot accurately align. In the process of building TyphoonKG, the present invention designs detailed entity alignment instructions, as shown in the fifth row of Table 1, so that LLMs can automatically identify and merge multiple representations pointing to the same entity in different sources. For example, for the various names of Typhoon "Kano" at different stages of its evolution, such as "tropical depression" and "tropical storm", the model can automatically classify these expressions as the same entity. This not only eliminates redundant information in the knowledge graph, but also improves the overall consistency of the graph. When dealing with such tasks, traditional methods often find it difficult to effectively solve the problems of inconsistent naming and entity ambiguity, resulting in a large number of duplicate or conflicting nodes in the graph construction process.

[0025] (4) Life cycle determination, capturing the dynamic evolution of typhoons: Life cycle determination is a core link in the construction of TyphoonKG, which aims to accurately depict the dynamic evolution process of natural disasters such as typhoons. For life cycle determination, traditional methods usually rely on predefined models or expert systems to infer the life cycle stage of disasters. Although this method performs well in certain specific situations, it is prone to insufficient adaptability when facing different types of typhoons or disasters with greater variability. To this end, the present invention designs a life cycle determination instruction, as shown in the sixth row of Table 1, so that LLMs can automatically generate corresponding time points and state changes according to different stages of typhoon disaster evolution, thereby more accurately capturing the details of the typhoon life cycle.

[0026] Finally, the instructions generated by the six modules of "role-playing", "structured output", "spatiotemporal knowledge completion", "co-reference resolution", "entity alignment" and "life cycle determination" are combined together to obtain a complete knowledge extraction instruction.

[0027] Table 1. Instructions for generating typhoon disaster knowledge extraction S102: Input the knowledge extraction instruction into the pre-established large language model LLMs for training to obtain TyphoonKGent; input the typhoon disaster data into TyphoonKGent, and output the structured knowledge graph extracted by TyphoonKGent.

[0028] Specifically, the generated knowledge extraction instructions are Instruction , the original data collected as Input , the manually annotated triples are used as Output , the three are combined into a training sentence <Instruction,Input,Output> The data is input into the pre-established large language model LLMs, and the parameters of LLMs are changed through the parameter efficient fine-tuning method Lora, and finally TyphoonKGent is trained. TyphoonKGent is a model rooted in the typhoon field and is good at extracting structured knowledge from unstructured text. It can receive unstructured text of typhoon disasters as output, and return the extracted structured knowledge to the user as output. The user can then build TyphoonKG based on the output of TyphoonKGent. Finally, the result evaluation of the typhoon disaster knowledge graph is completed.

[0029] For instruction fine-tuning, the present invention selects two LLMs, LLaMA model and ChatGLM model, for instruction fine-tuning. It was proposed by Google Research in 2021, bridging the gap between the next word prediction goal inside the model and the external human instruction goal. Therefore, instruction fine-tuning is regarded as a special form of supervised fine-tuning. Now, it has become one of the key methods to improve the controllability of LLMs and the alignment with user instructions. In addition, in order to reduce training resources, the present invention adopts Lora fine-tuning, which is one of the most commonly used parameter-efficient fine-tuning methods and has been verified by numerous studies, such as training in law, medicine, geographical sciences and other fields. Specifically, LoRA uses a matrix with a smaller parameter amount for low-rank approximation training when freezing the original LLMs parameters. For the pre-trained weight matrix , LoRA limits its update method, that is, the incremental parameter matrix of all parameter fine-tuning It is expressed as a low-rank approximation of two matrices A and B with smaller parameters: (1) in, and is the weight matrix of LoRA low-rank adaptation, the rank of the low-rank matrix The dimension d, at this time, the fine-tuning parameter amount is from of , becomes B and A ,thus ;in, ; Given input , output after adding LoRA : (2) in, , Represents the space of real matrices of different sizes.

[0030] During training, the original parameters Being frozen means Although it participates in forward propagation and backward propagation, its corresponding gradient is not calculated , and its parameters will not be updated. During inference, directly merge equation (2) into Therefore, there is no inference delay compared to the original LLMs.

[0031] Knowledge Graph Evaluation In terms of accuracy assessment, considering that reliable data sources are usually regarded as credible sources of knowledge, manual comparison is used to compare the content in the knowledge graph with the original content in the data source; specifically, the present invention subdivides the accuracy (Concept / Entity / Attribute / Relation (CEAR) Accuracy) into the following aspects: concept accuracy (detecting whether the description of each concept is consistent with the actual meaning and unambiguous), entity accuracy (detecting whether the entity is correctly identified and mapped to the corresponding concept or category), attribute accuracy (detecting whether the entity can be accurately represented) and relationship accuracy (detecting whether the relationship connection between entities is correct), as shown in Formula 3.

[0032] (3) In the formula, CEAR is the initials of the four words Concept, Entity, Attribute, and Relation. The accuracy is calculated as follows: the number of correct CEARs (Number of CorrectCEAR) divided by the total number of CEARs extracted from the text (Abstracted Number of CEAR).

[0033] In terms of coverage evaluation, the present invention uses random sampling of raw data to manually construct knowledge triples and compares them with the results automatically generated in the knowledge graph. Similarly, the evaluation of coverage (CEARCoverage) is also divided into the above aspects to ensure that the quality of the constructed TyphoonKG is fully reflected from different levels, as shown in Formula 4.

[0034] (4) Wherein, the coverage is calculated as: the total number of CEARs in TyphoonKG (Number of CEAR in theTyphoonKG) divided by the total number of CEARs in the original text (Total Number of CEAR).

[0035] The quantitative analysis of the invention results is shown in Table 2. The LLaMA / ChatGLM model after instruction fine-tuning shows excellent performance on the TyphoonKG construction task, which is superior to other methods and reaches the optimal level. Specifically, after instruction fine-tuning, both LLaMA and ChatGLM achieved the best value, and the performance was improved by about 30% compared with similar models without fine-tuning; compared with the performance of GPT-3.5 in three-sample reasoning, the accuracy of LLaMA and ChatGLM models in the evolution process text increased by 2.38%, 4.05%, 2.23% and 4.99% respectively, and the accuracy in the disaster effect text increased by 2.41%, 4.49%, 2.49% and 1.47% respectively.

[0036] Table 2: Effect verification table of typhoon disaster knowledge graph constructed by the present invention Embodiment 2: In the second aspect, as Figure 4 As shown, in order to achieve the above-mentioned purpose, the present invention discloses a typhoon disaster knowledge graph generation system based on a large language model, comprising: The instruction generation module 11 is used to obtain typhoon disaster data, and to carry out the cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and the typhoon disaster effect; based on the cognition and expression results of typhoon disaster knowledge, generate knowledge extraction instructions, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster situation; The graph generation module 12 is used to input the knowledge extraction instruction into the pre-established large language model LLMs for training to obtain TyphoonKGent; input the typhoon disaster data into TyphoonKGent, and output the structured knowledge graph extracted by TyphoonKGent.

[0037] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0038] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by the processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0039] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0040] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A method for generating a typhoon disaster knowledge graph based on a large language model, characterized in that: The method comprises the following steps: Acquire typhoon disaster data, and carry out cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and typhoon disaster effects; generate knowledge extraction instructions based on the cognition and expression results of typhoon disaster knowledge, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster conditions; The knowledge extraction instructions are input into the pre-established large language model LLMs for training to obtain TyphoonKGent; the typhoon disaster data is input into TyphoonKGent, and the structured knowledge graph extracted by TyphoonKGent is output.

2. The method for generating a typhoon disaster knowledge graph based on a large language model according to claim 1, characterized in that: The typhoon disaster data describes the cognitive and expression process of typhoon disaster knowledge through information aggregation of entity layer, stage layer, state layer and attribute layer.

3. The method for generating a typhoon disaster knowledge graph based on a large language model according to claim 2, characterized in that: The entity layer is used to express the components of typhoon disasters in the real world; The stage layer is used to express the specific changes, actions or records of the typhoon object in time and space, and is divided into four stages: formation stage, development stage, continuation stage and decay stage; The state layer is used to decompose the typhoon object in the entity layer into different states. The decomposition is achieved by analyzing the key attributes of the object, including the intensity, speed and landing place of the typhoon; the change of the object is represented by a set of states, and the development process of the event itself is expressed by the different states of the object; The attribute layer is used to define the timestamp, geographic location, specific attributes and behavior mode of the typhoon object in the entity layer.

4. The method for generating a typhoon disaster knowledge graph based on a large language model according to claim 1, characterized in that: The generating of the knowledge extraction instruction includes: one or more of: spatiotemporal knowledge completion, coreference resolution, entity alignment and life cycle determination.

5. The method for generating a typhoon disaster knowledge graph based on a large language model according to claim 4, characterized in that: The spatiotemporal knowledge completion is used to parse LLMs and complete the ambiguous spatiotemporal information in the text; The coreference resolution is used to automatically identify and parse pronouns and demonstrative expressions in texts for LLMs; The entity alignment is used to automatically identify and merge multiple representations of the same entity from different sources in LLMs; The life cycle determination is used to automatically generate corresponding time points and state changes for LLMs according to different stages of typhoon disaster evolution.

6. The method for generating a typhoon disaster knowledge graph based on a large language model according to claim 1, characterized in that: The process of inputting the knowledge extraction instruction into the pre-established large language model LLMs for training includes: Low-rank adapted weight matrices are used for low-rank approximation training of large language models (LLMs).

7. A typhoon disaster knowledge graph generation system based on a large language model, which adopts a typhoon disaster knowledge graph generation method based on a large language model as claimed in any one of claims 1 to 6, characterized in that: include: An instruction generation module is used to obtain typhoon disaster data, and to carry out cognition and expression of typhoon disaster knowledge based on the typhoon disaster data, wherein the cognition and expression of typhoon disaster knowledge includes the typhoon evolution process and typhoon disaster effects; based on the cognition and expression results of typhoon disaster knowledge, a knowledge extraction instruction is generated, wherein the typhoon disaster data includes wind speed, wind direction, central air pressure, motion trajectory, wave height, rainfall and disaster situation; The graph generation module is used to input knowledge extraction instructions into the pre-established large language model LLMs for training to obtain TyphoonKGent; typhoon disaster data is input into TyphoonKGent, and the structured knowledge graph extracted by TyphoonKGent is output.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, a typhoon disaster knowledge graph generation method based on a large language model as described in any one of claims 1 to 6 is adopted.

9. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by the processor, a typhoon disaster knowledge graph generation method based on a large language model as described in any one of claims 1 to 6 is adopted.

Images