Cross-organization emergency response process model extraction method based on large language model

Abstract

The invention discloses a cross-organization emergency response process model extraction method based on a large language model, belongs to the field of process mining and natural language processing, and constructs a large language model general prompt framework containing emergency response process element definition and judgment criteria. Based on a general prompt framework, a prompt strategy fusing role playing and context learning is adopted to extract explicit entities such as emergency organizations and response tasks and relationships of the explicit entities. Inferring and extracting implicit entities and implicit relation elements on the basis of explicit elements by adopting a thinking chain fused prompt strategy; and finally, making a mapping rule to convert the extracted complete process elements into a logically coherent cross-organization emergency response process model, and presenting a result in a graphical manner. According to the method, domain data training is not needed, accurate flow element extraction can be achieved under the condition of few samples, the problem of implicit flow element missing is solved, and technical support is provided for emergency decision making.

Description

Technical Field

[0001] This invention belongs to the fields of process mining and natural language processing, specifically relating to a method for extracting cross-organizational emergency response process models based on a large language model. Background Technology

[0002] An emergency response plan is a pre-established plan or scheme to respond quickly, orderly, and effectively to emergencies, minimizing casualties and economic losses. It provides clear guidance and standards for emergency response actions. The emergency response section details the execution process of four levels of emergency response and the interaction and collaboration processes between organizations. It is a typical process description text, called a "response process text." The emergency response process text describes the response process at two abstract levels: organization and task. The organization-level response process is coarse-grained, primarily aimed at emergency decision-makers, focusing on the collaboration and interaction processes between organizations during the emergency response. The task-level response process is fine-grained, aimed at emergency implementers, focusing on the sequential dependencies between response tasks within the organization. Process model extraction automatically extracts process information from the process description text and describes the extraction results in a formalized manner, involving technologies such as natural language processing and process mining. Emergency response cross-organizational process model extraction differs from traditional process model extraction. It requires comprehensive consideration of the organizational interaction and task execution characteristics of the emergency response process text, and in-depth mining of semantic information to construct a complete and high-quality process model. The extracted emergency response process model can be used to assist domain experts in completing modeling work, and facilitate emergency personnel at different levels to understand the emergency response execution process from different perspectives.

[0003] Process model extraction typically includes two methods: rule-based matching and deep neural network-based methods. Rule-based process model extraction uses natural language processing (NLP) tools to obtain analysis results such as syntactic trees, dependency relationships, and semantic role labeling. Extraction rules are then formulated based on these results to extract process elements and build the model. This method heavily relies on NLP tools, and due to the performance limitations of these tools, it performs better on English text than Chinese text. Furthermore, it is difficult to design extraction rules that satisfy all sentence structures; this method is only suitable for process description texts with complete sentence structures and fixed sentence patterns. Process elements in text are described as continuous word sequences; therefore, process element extraction is treated as a sequence labeling problem, leading to the development of deep neural network-based process model extraction methods. Deep neural network-based methods require training a probabilistic model on a large amount of labeled data. If the labeled data is abundant and of high quality, this method outperforms rule-based methods in terms of extraction accuracy and efficiency. Due to the lack of high-quality labeled data in the field of emergency management, extracting emergency response process models based on few-sample and zero-sample data is a current research focus and challenge. Summary of the Invention

[0004] To address the problems of existing technologies, such as reliance on large amounts of labeled data, difficulty in handling long text logic, and inability to effectively mine implicit process elements, this invention proposes a cross-organizational emergency response process model extraction method based on a large language model. This method adopts a hierarchical prompting framework to simulate the cognitive process of domain experts. It can work under zero-sample or few-sample conditions without targeted training and can extract implicit process elements through controlled logical reasoning, thereby significantly improving the accuracy and completeness of the constructed process model.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for extracting cross-organizational emergency response process models based on large language models includes the following steps: Step 1: Construct a general prompting framework to guide the large language model in extracting process elements; Step 2: Based on the general prompting framework, extract the explicit process elements clearly stated in the emergency response plan text; Step 3: Based on the explicit process elements, logical reasoning is performed by guiding the large language model to extract the implicit process elements hidden in the text. The implicit process elements include implicit entities and implicit relationships. Step 4: Based on all extracted process elements, construct a structured cross-organizational emergency response process model.

[0006] Furthermore, the general prompting framework includes three modules: task instructions, few-sample examples, and input text; The task instructions are used to guide the large language model into a specific domain state through role-playing instructions, inject domain knowledge into the large language model through process element definition, and finally stipulate that the output must be in a specific structured format; the few-shot examples are used to demonstrate the standardized input and output format for a specific extraction task; the input text is the original emergency response process text to be processed.

[0007] Furthermore, the process elements include entity elements and relational elements; The entity elements of the emergency response process are defined as the set of emergency organizations O and the set of response tasks T. The emergency organization is defined as the specific executor or receiver of the emergency response task, that is, a noun phrase that acts as the subject of the action in the text. The response task is defined as the specific action performed in the emergency response process, that is, a verb-object phrase consisting of a predicate verb and an object. Define the relational elements between entity elements as triples. ,in and Represents entity elements and relation type This includes: the relationship between the executing subject and the interactive object between organizations and tasks; the information transmission between organizations; the relationship between task synchronization and task selection; and the relationship between the order, parallelism, and selection of tasks.

[0008] Furthermore, step 2 specifically includes: Step 2.1: Based on the general prompting framework described in Step 1, construct an explicit entity element extraction prompt template. This template integrates role-playing, process element definition, and context learning strategies in task instructions and examples, driving the large language model to identify and extract two types of entities, emergency organizations and response tasks, from the input emergency response process text. Step 2.2: Based on the general prompting framework described in Step 1, construct an explicit relation element extraction prompt template, drive the large language model to identify and extract the explicit relations between the explicit entities according to the eight predefined relation types, and output them in the form of a triple list.

[0009] Furthermore, step 3 specifically includes: Step 3.1: Based on the general prompting framework, construct an implicit entity element extraction prompting template that integrates the thinking chain mechanism for the extraction results of the explicit process elements, and use the template to drive the large language model, combined with the context and domain knowledge, to infer the missing implicit entity elements. Step 3.2: Based on the general prompting framework, for the explicit relation elements and the set of explicit and implicit entity elements, construct an implicit relation element extraction prompting template that integrates the thinking chain mechanism, and use this template to drive the large language model, combine the context and domain knowledge to infer the missing implicit relation elements, and perform matching and logical verification according to the predefined relation types to ensure the type legality and logical rationality of the extracted implicit relations.

[0010] Further, step 4 includes: Step 4.1: Define mapping rules to map the extracted complete process elements to the semantic components of the BPMN model. The mapping rules include: (1) Map emergency organizations to swim lanes; (2) Map the response task to an activity, and locate the activity in the swimlane corresponding to its executing organization based on the relationship between the extracted organization and the executing entity of the task. (3) Map the relationships between tasks to control flow; (4) Map inter-organizational relationships to message flows; Step 4.2: Apply the mapping rules to transform the extracted process elements into a graphical cross-organizational emergency response process model and present it.

[0011] The beneficial technical effects of this invention are as follows: (1) The large language model is used to extract process elements by leveraging its powerful zero-shot and few-shot learning capabilities. This eliminates the need to rely on large-scale, high-quality labeled data, fundamentally solving the problem of data scarcity in the field of emergency response plans. Through a hierarchical prompting framework, the large language model is gradually guided to complete the identification of explicit process elements and the reasoning and identification of complex implicit process elements. The reasoning process is constrained, which effectively reduces the "illusion" phenomenon of the large language model and ensures the accuracy and reliability of the process element extraction results.

[0012] (2) Some process elements are implied in the semantics of the response process text and cannot be directly extracted through literal description. This results in the lack of completeness and coherence of the extracted process model, requiring further rule formulation to construct a complete process model. Introducing the thinking chain reasoning mechanism into the large language model enables the extraction of implicit organization, implicit tasks, and implicit relationships, solving the problem that traditional methods struggle to handle implicit logic and ensuring the logical completeness and coherence of the extracted process model.

[0013] (3) The entire extraction method automatically transforms unstructured emergency response plan texts into structured cross-organizational emergency response process models, comprehensively depicting the deep semantic features of emergency response plan texts and mining the emergency response execution process from both organizational and task perspectives. The extracted cross-organizational emergency response process models can not only assist domain experts in completing modeling work, but also facilitate emergency personnel at different levels to understand the emergency response execution process from different perspectives, providing technical support for improving emergency response efficiency. Attached Figure Description

[0014] Figure 1 This is a framework diagram of the cross-organizational emergency response process model extraction method based on a large language model in this invention.

[0015] Figure 2 This is a schematic diagram of a prompt template used to extract explicit entity elements in this invention.

[0016] Figure 3 This is a schematic diagram of a prompt template used to extract explicit relational elements in this invention.

[0017] Figure 4 This is a schematic diagram of a prompt template used to extract implicit entity elements in this invention.

[0018] Figure 5 This is a schematic diagram of a prompt template used to extract implicit relationship elements in this invention.

[0019] Figure 6 This is an example of an emergency response plan for an accident at a chemical group.

[0020] Figure 7This is an example diagram of a cross-organizational emergency response process model. Detailed Implementation

[0021] The specific embodiments of the present invention will be further described below with reference to specific examples: This invention proposes a method for extracting cross-organizational emergency response process models based on a large language model. It aims to simulate the cognitive process of domain experts, gradually achieving the extraction of explicit process elements and complex implicit process elements. This completes the end-to-end automatic transformation from emergency response plan text to a cross-organizational emergency response process model. The extraction framework is as follows: Figure 1 As shown. Specifically, it includes the following steps: Step 1: Construct a general prompting framework to guide the large language model in extracting process elements; This step first clarifies the extraction objective, namely, defining the entity and relational elements in the emergency response process, and then constructs a general prompting framework for the large language model based on the extraction objective. Emergency response process elements include entity and relational elements; the specific definitions of these process elements are as follows: The entity elements of the emergency response process are defined as the set of emergency organizations O and the set of response tasks T, where the set of emergency organizations... An emergency response organization is defined as the specific executor or receiver of emergency response tasks; that is, a noun phrase that acts as the subject of the action in the text; a set of response tasks. A response task is defined as a specific action performed during an emergency response, that is, a verb-object phrase consisting of a predicate verb and an object. Define the relational elements between entity elements as triples. ,in and Represents entity elements and relation type Includes three types: The relationship between the executing entity and the interacting object between the organization and the task: In this relationship, the entity Recorded as ,entity Recorded as ,relation .in, Indicates organization It is a task The implementing entity; Indicates organization It is a task Interactive objects are typically used to describe the receiving objects in a message receiving task.

[0022] Information transfer, task synchronization, and task selection relationships between organizations: In this relationship, entities Recorded as ,entity Recorded as ,relation .in, Indicates organization Towards Report messages or issue orders; Indicates organization and Together, they complete the same response task; Indicates organization and Only one of them needs to complete a certain response task; The sequential, parallel, and selection relationships between tasks: In this relationship, entities Recorded as And entity Recorded as ,relation .in, Indicates task The task can only be executed after completion. ; Indicates task and There is no strict sequential execution order, but all tasks must be completed before subsequent tasks can be executed; Indicates task and Only tasks that meet the execution conditions can be executed.

[0023] Based on the clearly defined extraction objectives, and to adapt to the specific needs of different extraction tasks such as explicit entities, explicit relations, implicit entities, and implicit relations, while ensuring the standardization and consistency of the input and output formats of large language models, this invention designs a unified and universal prompting framework. Subsequent extraction prompting templates for different tasks are all built upon this framework. This universal prompting framework includes three modules: task instructions, few-shot examples, and input text. Task Instructions: This section is the core of the prompts, clearly guiding the large language model to complete specific extraction tasks. First, role-playing instructions (e.g., "You are an emergency management expert, your task is...") guide the large language model into a specific domain context. Second, by defining process elements, precise definitions and criteria are provided for "emergency organization" and "response task," injecting domain knowledge into the large language model. Finally, the output must be in a specific structured format; ensuring consistency of the generated results and ease of subsequent processing.

[0024] Few-shot examples: Based on the specific extraction task, several "input text-output result" samples are randomly selected from the labeled dataset as few-shot examples to demonstrate the standard input and output format for a specific extraction task; The input text is the original emergency response plan text to be processed.

[0025] Step 2: Based on the general prompting framework, extract the explicit process elements clearly stated in the emergency response plan text; The goal of this step is to identify and extract explicitly stated process elements from the input emergency response plan text, laying the foundation for building a complete process model.

[0026] Step 2.1: Based on the general prompting framework described in Step 1, construct as follows: Figure 2 The explicit entity element extraction prompt template shown integrates role-playing, process element definition and context learning strategies in the task instructions and examples, driving the large language model to identify and extract two types of entities, emergency organization and response task, from the input emergency response process text. Upon receiving the prompt, the large language model will analyze the input sentence based on the instructions and examples, and output the emergency organization entities and response task entities it has identified in a preset structured format.

[0027] Step 2.2: After extracting the display entity elements, it is necessary to further identify the relationships between them. The extraction of display relationship elements uses the same prompting framework and strategy as the extraction of display entity elements, and is constructed based on the general prompting framework described in Step 1, as follows: Figure 3 The explicit relation element extraction prompt template shown explicitly requires the model to identify the above eight specific relation types and output them as a list of triples. The context example section shows examples of extracting relation triples from typical sentences, driving the large language model to identify and extract the explicit relations between the explicit entities based on the eight predefined relation types.

[0028] Step 3: Based on the explicit process elements, guide the large language model to perform logical reasoning and extract the implicit process elements hidden in the text. The implicit process elements include implicit entities and implicit relationships. Emergency response plan texts often omit certain implicit process elements due to linguistic omissions, including implicit entity elements and implicit relational elements. Implicit entity elements are key entities not directly stated in the text but inferred from contextual logic and domain knowledge. Implicit relational elements are semantically connected elements that are not explicitly stated but are essential for maintaining the logical coherence and integrity of the emergency response process. The task of extracting implicit process elements is inherently subjective and uncertain; without constraints, it can easily lead to divergent reasoning in large models. Therefore, this invention, based on the explicit extraction results, strictly limits the inference scope to two core dimensions: "information interaction" and "process continuity." Specifically, this includes: Step 3.1: Based on the general prompting framework, construct a system for extracting explicit process elements, as shown below. Figure 4 The implicit entity element extraction prompt template shown is based on the fusion of the thinking chain mechanism. This template is used to drive a large language model, combining contextual information and domain knowledge to infer the missing implicit entity elements. At the same time, in order to improve the accuracy and interpretability of the implicit element extraction results, the prompt template introduces strict constraints.

[0029] Step 3.2: Based on the general hint framework, construct the following for explicit relational elements and sets of explicit and implicit entity elements: Figure 5 The implicit relation element extraction prompt template shown is used to integrate the thinking chain mechanism. This template drives the large language model, which combines context and domain knowledge to infer the missing implicit relation elements. It also performs matching and logical verification based on eight predefined relation types to ensure the legality and logical rationality of the extracted implicit relation types. Finally, it outputs a list of implicit relation triples.

[0030] Step 4: Based on all extracted process elements, construct a structured cross-organizational emergency response process model.

[0031] The goal of this step is to transform the extracted complete process elements into an intuitive and standardized graphical process model.

[0032] Step 4.1: Define mapping rules to map the extracted complete process elements to the semantic components of the BPMN model. The mapping rules include: (1) Organizational mapping: Map each emergency organization entity element to a swimlane in the BPMN model to construct the organizational structure and responsibility boundaries of the cross-organizational emergency response process model; (2) Task mapping: Map each response task entity element to an activity in the BPMN model. Based on the extracted "subject" relationship between the organization and the task, locate the activity in the swimlane corresponding to its executing organization. (3) Control flow mapping: The “sequential”, “parallel” and “selective” relationships between tasks are transformed into “sequential flow”, “parallel gateway” and “exclusive gateway” in the BPMN model, respectively, to describe the execution path and logical conditions of the process; (4) Message flow mapping: Maps the relationships between organizations as “message flows” that cross different swimlane boundaries, intuitively demonstrating cross-departmental interaction and collaboration; Step 4.2: Apply the mapping rules to transform the extracted process elements into a graphical cross-organizational emergency response process model and present it.

[0033] To verify the effectiveness and practicality of the cross-organizational emergency response process model extraction method based on a large language model proposed in this invention, this embodiment selects a text fragment of the emergency response process from an accident site emergency plan of a chemical group as an analysis case, such as... Figure 6 As shown, strictly following the four steps described above, the system demonstrates the transformation process from raw unstructured text to a structured workflow model. Based on step 2.1, the extracted display organization entities are shown in Table 1, and the extracted display task entities are shown in Table 2.

[0034] Table 1 shows the results of the organizational entity extraction. ; Table 2 shows the results of task entity extraction. ; Based on the results of explicit entity extraction, and in accordance with step 2.2, the extracted explicit relational elements are shown in Table 3.

[0035] Table 3 shows the results of relation element extraction. ; Based on the explicit element extraction, implicit process elements are extracted according to step 3. According to step 3.1, the extracted implicit entity elements are shown in Table 4, using "*" to distinguish between explicit and implicit entity elements. According to step 3.2, implicit relationship elements are extracted based on explicit relationships and the principles of "information interaction" and "process continuity," and the extraction results are shown in Table 5.

[0036] Table 4. Results of Implicit Entity Feature Extraction ; Table 5. Results of Implicit Relation Extraction ; Based on the extracted explicit and implicit process elements, and according to the rules described in step 4.1, generate as follows: Figure 7 The cross-organizational emergency response process model is shown.

[0037] To demonstrate the feasibility and superiority of the proposed method in extracting cross-organizational emergency response process models, a comparative experiment was conducted with existing mainstream prompting engineering strategies. The experiment was performed using the DeepSeek-V3.2 large model. The proposed method uses precision (P), recall (R), and F1 score as evaluation metrics. Considering the flexibility of large language models in generating text, a soft-matching metric mechanism was introduced in the evaluation process to more accurately assess the extraction performance. Unlike traditional exact matching, soft matching allows for reasonable differences in literal expression between predicted and labeled entities. The determination of soft matching is based on edit distance similarity, calculated as follows: ; in, and These represent the entity text predicted by the model and the text annotated by humans, respectively. This represents the Lewinstein edit distance between the two. and These represent the length of the string. In the experimental setup of this invention, when... (threshold) (Set to 0.65) or exist When there is an inclusion relationship, it is determined to be a correct extraction.

[0038] In the process element extraction experiment, the commonly used few-shot learning (FS), chain-of-thought (CoT), and role-playing (RP) strategies were selected as benchmarks and compared with the hierarchical composite prompting strategy (Ours) proposed in this invention.

[0039] Table 6 shows the extraction results of different prompting strategies on the two core explicit entities: "emergency organization" and "response task". Experimental results show that the composite prompting strategy proposed in this invention significantly outperforms other benchmark methods in overall performance. Specifically, in terms of organizational entity extraction, the strategy proposed in this invention achieved an F1 score of 0.8485, an improvement of approximately 3.3% compared to the second-best performing few-shot learning strategy (0.8148). Although the thought chain strategy achieved the highest recall of 0.9468, its precision was only 0.6260, indicating that the thought chain strategy easily leads to over-association in the model, identifying a large number of irrelevant entities. In contrast, the strategy proposed in this invention maintains a relatively high precision (0.8030) while maintaining a high recall (0.8994), effectively balancing the requirements of "representation" and "precision". Regarding task entity extraction... The strategy proposed in this invention also performed well, with an F1 score of 0.8411, which is significantly higher than the few-shot learning strategy (0.7895), the thought chain strategy (0.7733), and the role-playing strategy (0.6774). This indicates that the hierarchical prompting strategy can help the model better define the semantic boundaries of the "task" and reduce the omission and misjudgment of action descriptions in long texts.

[0040] Table 6. Explicit feature extraction effects under different prompting strategies ; Table 7 shows the extraction performance of three types of explicit relations. Relation extraction requires a deep understanding of the semantic connections within the context, making it more difficult than entity extraction. Experimental results show that the strategy proposed in this invention achieves the highest F1 score for all relation types. In organization-to-organization relation extraction, the proposed strategy achieves an F1 score of 0.6520, far exceeding few-shot learning (0.5786) and other strategies. In organization-to-task relation extraction, although the few-shot learning strategy achieves a recall of 1.0000, its precision is extremely low (0.4479), indicating that this method tends to brute-force match all organizations with tasks, generating a large number of erroneous relations. The strategy proposed in this invention, through the constraint of compound prompts, achieves an F1 score of 0.6796, demonstrating its accuracy in determining the key logic of "who performs what." In task and task relationship extraction, the strategy proposed in this invention achieved an F1 score of 0.6563, which is nearly 9 percentage points higher than the second-best few-shot learning strategy and significantly better than the thought chain strategy (0.4222) and role-playing strategy (0.4044), demonstrating the effectiveness of this method in handling complex process logic such as sequential and parallel processes.

[0041] Table 7. Effects of Explicit Relationship Element Extraction under Different Hint Strategies ; Table 8 shows the results of implicit process element extraction. The strategy proposed in this invention achieves an F1 score of 0.5770, while the best-performing baseline method, the few-shot learning strategy, only achieves 0.4515. Role-playing and thought chain strategies both fall below 0.40, indicating that general prompting methods struggle to guide the model to perform deep reasoning and completion. The composite strategy, integrating thought chain and context learning, can stimulate the model's reasoning ability and effectively identify implicit entities. In implicit relation extraction, the strategy proposed in this invention also maintains a leading position with an F1 score of 0.5560, significantly higher than other methods (0.3500-0.4684). The role-playing strategy performed the worst in this task (Recall only 0.2942), indicating that simple role-playing is insufficient to support complex logical inference.

[0042] In summary, for basic explicit entity recognition, complex explicit relationship determination, and highly challenging implicit logic inference, the hierarchical composite prompting strategy (Ours) proposed in this invention demonstrates significant advantages in Precision, Recall, and F1 score. Experimental results show that the method proposed in this invention can not only extract explicit process information but also effectively complete implicit logic breakpoints in the process, providing technical support for constructing a complete cross-organizational emergency response process model.

[0043] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for extracting cross-organizational emergency response process models based on a large language model, characterized in that, Includes the following steps: Step 1: Construct a general prompting framework to guide the large language model in extracting process elements; Step 2: Based on the general prompting framework, extract the explicit process elements clearly stated in the emergency response plan text; Step 3: Based on the explicit process elements, logical reasoning is performed by guiding the large language model to extract the implicit process elements hidden in the text. The implicit process elements include implicit entities and implicit relationships. Step 4: Based on all extracted process elements, construct a structured cross-organizational emergency response process model.

2. The method for extracting cross-organizational emergency response process models based on a large language model according to claim 1, characterized in that, The general prompting framework includes three modules: task instructions, few-sample examples, and input text. The task instructions are used to guide the large language model into a specific domain state through role-playing instructions, inject domain knowledge into the large language model through process element definition, and finally stipulate that the output must be in a specific structured format; the few-shot examples are used to demonstrate the standardized input and output format for a specific extraction task; the input text is the original emergency response process text to be processed.

3. The method for extracting a cross-organizational emergency response process model based on a large language model according to claim 2, characterized in that, The process elements include entity elements and relational elements; The entity elements of the emergency response process are defined as the set of emergency organizations O and the set of response tasks T. The emergency organization is defined as the specific executor or receiver of the emergency response task, that is, a noun phrase that acts as the subject of the action in the text. The response task is defined as the specific action performed in the emergency response process, that is, a verb-object phrase consisting of a predicate verb and an object. Define the relational elements between entity elements as triples. ,in and Represents entity elements and relation type This includes: the relationship between the executing subject and the interactive object between organizations and tasks; the information transmission between organizations; the relationship between task synchronization and task selection; and the relationship between the order, parallelism, and selection of tasks.

4. The method for extracting a cross-organizational emergency response process model based on a large language model according to claim 3, characterized in that, Step 2 specifically includes: Step 2.1: Based on the general prompting framework described in Step 1, construct an explicit entity element extraction prompt template. This template integrates role-playing, process element definition, and context learning strategies in task instructions and examples, driving the large language model to identify and extract two types of entities, emergency organizations and response tasks, from the input emergency response process text. Step 2.2: Based on the general prompting framework described in Step 1, construct an explicit relation element extraction prompt template, drive the large language model to identify and extract the explicit relations between the explicit entities according to the eight predefined relation types, and output them in the form of a triple list.

5. The method for extracting a cross-organizational emergency response process model based on a large language model according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Based on the general prompting framework, construct an implicit entity element extraction prompting template that integrates the thinking chain mechanism for the extraction results of the explicit process elements, and use the template to drive the large language model, combined with the context and domain knowledge, to infer the missing implicit entity elements. Step 3.2: Based on the general prompting framework, for the explicit relational elements and the set of explicit and implicit entity elements, construct an implicit relational element extraction prompting template that integrates the thinking chain mechanism, and use this template to drive the large language model, combine the context and domain knowledge to infer the missing implicit relational elements, and perform matching and logical verification according to the predefined relational types to ensure the type legality and logical rationality of the extracted implicit relations.

6. The method for extracting a cross-organizational emergency response process model based on a large language model according to claim 1, characterized in that, Step 4 includes: Step 4.1: Define mapping rules to map the extracted complete process elements to the semantic components of the BPMN model. The mapping rules include: (1) Map emergency organizations to swim lanes; (2) Map the response task to an activity, and locate the activity in the swimlane corresponding to its executing organization based on the relationship between the extracted organization and the executing entity of the task. (3) Map the relationships between tasks to control flow; (4) Map inter-organizational relationships to message flows; Step 4.2: Apply the mapping rules to transform the extracted process elements into a graphical cross-organizational emergency response process model and present it.

Images