A Method for Constructing an Emergency Accident Case-Based Decision Support System Based on a Large Model

Abstract

This invention provides a method and system for constructing an emergency accident case-based decision support system based on a large model, comprising: Step 1: reading the target business domain, analyzing the target business domain based on the target large model, and storing the analysis results in a vector database; Step 2: retrieving historical accident cases, matching the analysis results in the vector database with the historical accident cases according to the target large model, and importing the matching results into a graph database; Step 3: obtaining several question-answer pairs and the relationships between question-answer pairs in the target business domain, and simultaneously saving the relationships between question-answer pairs as an evaluation dataset to the vector database; Step 4: processing the several question-answer pairs and the evaluation dataset based on the target large model, the vector database, and the graph database, and constructing the emergency accident case-based decision support system based on the processing results. This improves the efficiency of constructing the accident case database.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for constructing an emergency accident case-based decision support system based on a large model. Background Technology

[0002] Currently, emergency management encompasses several key operational areas: monitoring and early warning, command and rescue, and decision support. Emergency incident case analysis typically refers to the process of using historical emergency incident data to support emergency management, risk assessment, and prevention efforts. By analyzing and summarizing historical incident cases, valuable lessons can be learned, helping to develop more effective contingency plans and measures.

[0003] However, the construction of traditional knowledge graph-based accident case databases cannot be separated from the abstract ontology modeling work that requires a deep understanding of business needs and professional knowledge. This work is complex, tedious, and inefficient.

[0004] Therefore, in order to overcome the above-mentioned technical problems, the present invention provides a method for constructing an emergency accident case-assisted decision-making system based on a large model. Summary of the Invention

[0005] This invention provides a method for constructing an emergency accident case-based decision support system based on a large model. It analyzes a target business domain using a target large model, accurately and effectively constructs a vector database, and matches the analysis results with historical accident cases to effectively determine the specific relationships between entities, relations, and entity pairs within the target business domain. Based on these relationships, a graph database is constructed to facilitate the subsequent construction of the emergency accident case-based decision support system. Furthermore, by analyzing several question-and-answer pairs and the relationships between them using the obtained vector and graph databases, the emergency accident case-based decision support system is effectively constructed, improving the efficiency of accident case database construction. Periodically introduced relevant accident case investigation reports are used to extract question-and-answer pairs as an evaluation benchmark database. Based on similarity analysis, the question-and-answer results of the constructed target knowledge base are evaluated, and the relevant entity content in the accident case database is revised according to the results, achieving automatic calibration and updating of the accident case database and ensuring the comprehensiveness and reliability of the emergency accident case-based decision support system.

[0006] A method for constructing an emergency accident case-based decision support system based on a large model, characterized by comprising:

[0007] Step 1: Read the target business domain, analyze the target business domain based on the target big model, and store the analysis results in the vector database;

[0008] Step 2: Retrieve historical accident cases, match the analysis results in the vector database with the historical accident cases according to the target large model, and import the matching results into the graph database;

[0009] Step 3: Obtain several question-answer pairs and the relationships between them in the target business domain. At the same time, use the relationships between the question-answer pairs as an evaluation dataset and save them to the vector database simultaneously.

[0010] Step 4: Based on the target large model, process several question-answer pairs and evaluation datasets using vector database and graph database, and construct an emergency accident case auxiliary decision-making system based on the processing results.

[0011] Preferably, a method for constructing an emergency accident case-based decision support system based on a large model includes, in step 1, reading the target business domain and analyzing the target business domain based on the target large model, including:

[0012] Read the target business domain and determine the text information of the target business domain;

[0013] Based on the target big model, the text information of the target business domain is analyzed and processed to extract the target entities in the target business domain and the initial relationships between the target entities;

[0014] The target entities are classified based on the target big model, and conceptual entities are created based on the classification results. The conceptual entities are then analyzed based on the target big model to determine the conceptual relationships between them.

[0015] Store the target entities, the initial relationships between target entities, the conceptual entities, and the conceptual relationships between conceptual entities in the vector database.

[0016] Preferably, a method for constructing an emergency accident case-assisted decision-making system based on a large model retains in the vector database: the mapping relationship between conceptual entities and target entities, and the mapping relationship between conceptual relationships and initial relationships.

[0017] Preferably, in a method for constructing an emergency accident case-based decision support system based on a large model, step 2 involves retrieving historical accident cases, matching the analysis results in the vector database with the historical accidents according to the target large model, and importing the matching results into a graph database, including:

[0018] S201: Retrieve multiple historical accident cases based on a preset method and obtain the conceptual entities and conceptual relationships from the vector database;

[0019] S202: Based on the target large model, historical accident cases are matched according to conceptual entities and conceptual relationships to generate corresponding conceptual entity-conceptual relationship-conceptual entity pairs;

[0020] S203: Revise the concept entity-concept relationship-concept entity pair and import it into the graph database.

[0021] Preferably, a method for constructing an emergency accident case-based decision support system based on a large model involves revising the concept entity-concept relationship-concept entity pairs and importing them into a graph database, including:

[0022] Retrieve specific entities under the concept entity in the vector database, match the specific entities under the concept entity according to the classification, and determine the category similarity between the specific entities under the concept entity and the concept entities in the concept entity-concept relationship-concept entity pair;

[0023] If the category similarity between a concrete entity under a conceptual entity and the conceptual entities at both ends of the conceptual relationship in the conceptual entity-conceptual relationship-conceptual entity pair is greater than a preset threshold, then the conceptual entity is replaced with a concrete entity, and the conceptual entity-conceptual relationship-conceptual entity pair is revised based on the replacement result to obtain a concrete entity-conceptual relationship-conceptual entity pair.

[0024] Based on the replacement results, all concrete entities-concrete relations-concrete entity pairs are sequentially imported into the graph database.

[0025] Preferably, in a method for constructing an emergency accident case-based decision support system based on a large model, step 3 involves obtaining several question-answer pairs and the relationships between them in the target business domain, and simultaneously saving them to a vector database, including:

[0026] Based on the target large model and the preset prompts, several question-answer pairs are generated, and the relationships between the question-answer pairs are also generated.

[0027] The relationships between question-and-answer pairs are used as the evaluation dataset;

[0028] Several question-answer pairs and evaluation datasets are synchronously saved to a vector database.

[0029] Preferably, a method for constructing an emergency accident case-based decision support system, before classifying target entities based on the large model, further includes:

[0030] Similarity analysis is performed on target entities based on a large model to obtain the similarity between target entities;

[0031] Obtain the baseline similarity, compare the similarity between target entities with the baseline similarity, and determine the duplicate target entities whose similarity between target entities is greater than the baseline similarity based on the comparison results;

[0032] The system performs deduplication on duplicate target entities and classifies them based on the target entity model according to the deduplication results.

[0033] Preferably, in a method for constructing an emergency accident case-based decision support system based on a large model, step 4 involves processing several question-answer pairs and evaluation datasets based on a vector database and a graph database using the large model, and constructing the emergency accident case-based decision support system based on the processing results, including:

[0034] Obtain several questions, perform semantic understanding on the questions based on the target large model, and extract specific entities i from the questions;

[0035] With the support of the vector database, the specific entity i in the problem is matched with the specific entities in the graph database to obtain the set k of the top ten specific entities with the highest similarity.

[0036] Retrieve all relations of the top ten specific entity set k, and generate a relation set R, where the relation set R contains multiple relations r;

[0037] A second similarity matching is performed based on the relationships between question-answer pairs and the relation set R, extracting the relation r with the highest similarity greater than a preset threshold. max ;

[0038] Based on the specific entity i and the relation r max Match specific entity I in the graph database;

[0039] Based on the target large model, semantic understanding is performed on the answers in the assessment dataset to obtain the entity set S corresponding to the answers in multiple assessment datasets, where the entity set S contains multiple entities s;

[0040] The entity set S corresponding to the answers in multiple assessment datasets is matched with the specific entity I using the third similarity method, and it is determined whether there is a similarity higher than the preset similarity threshold.

[0041] If there exists an entity s with the highest similarity score that is higher than the preset similarity threshold. max When that happens, the entity s corresponding to the answer in the evaluation dataset will be... max Replace entity I with the answer for the current specific entity i to form specific entity ir. max -Entities max and the specific entity ir max -Entities max Import to graph database;

[0042] Otherwise, the specific entity ir max - Import entity I into the graph database;

[0043] A decision support system for emergency accident cases was built based on the imported results.

[0044] Preferably, a method for constructing an emergency accident case-based decision support system based on a large model further includes:

[0045] Retrieve accident case investigation reports from the pre-set emergency management database;

[0046] Obtain the runtime environment of the target large model, and analyze the accident case investigation report in the runtime environment of the target large model according to the preset technology to construct a training text dataset;

[0047] The target large model is trained based on the training text dataset, and the optimal adjustment parameters of the model are determined based on the model loss function;

[0048] Based on the optimal adjustment parameters of the model, the parameters of the target large model are adjusted to obtain the optimal target large model.

[0049] Preferably, a method for constructing an emergency accident case-based decision support system based on a large model involves analyzing and processing textual information in the target business domain based on the target large model, extracting target entities in the target business domain and the initial relationships between target entities, including:

[0050] The first text information of the target business domain is input into the target large model for format rewriting to obtain the second text information;

[0051] Based on the target large model, the second text information is identified, and multiple key information segments in the second text information are obtained;

[0052] Multiple key information segments are marked to obtain multiple key information tags, and the correlation between each key information tag and the tag of the other key information is calculated;

[0053] Obtain the relevance threshold range, and compare the relevance between the tag of each key information and the tags of the other key information with the relevance threshold range;

[0054] Each key information tag is sequentially associated with the tags of the other key information tags, and the tags whose relevance falls within the relevance threshold range are then associated with the tags.

[0055] Determine the relationships between each key piece of information based on the correlation results;

[0056] Based on the target big model, each key piece of information is identified to obtain the central word vector of the key information. At the same time, the industry vector dictionary of the target business domain is retrieved.

[0057] The central word vector is matched in the industry vector dictionary. If the central word vector exists in the industry vector dictionary, the central word vector is used as the target entity in the target business domain.

[0058] Otherwise, the central vocabulary vector will not be used as the target entity in the target business domain;

[0059] Based on the matching results, multiple target entities in the target business domain are obtained. At the same time, the relationships between each key piece of information are mapped in the corresponding target entities to obtain the initial relationship of each target entity.

[0060] Compared with existing technologies, the beneficial effects of this invention are as follows: By analyzing the target business domain through a large target model, an accurate and effective vector database is constructed. The analysis results are then matched with historical accident cases to effectively determine the specific relationships between entities, relations, and entity pairs within the target business domain. A graph database is constructed based on these relationships, facilitating the subsequent construction of an emergency accident case-based decision support system. Furthermore, by analyzing several question-and-answer pairs and the relationships between them using the obtained vector and graph databases, the emergency accident case-based decision support system is effectively constructed, improving the efficiency of accident case database construction. Additionally, by periodically introducing relevant accident case investigation reports, question-and-answer pairs are extracted as an evaluation benchmark database. Based on similarity analysis, the question-and-answer results for constructing the target knowledge base are evaluated. The relevant entity content in the accident case database is revised based on the results, achieving automatic calibration and updating of the accident case database, thus ensuring the comprehensiveness and reliability of the emergency accident case-based decision support system.

[0061] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0062] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0063] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0064] Figure 1 This is a flowchart illustrating a method for constructing an emergency accident case-assisted decision-making system based on a large model, as described in an embodiment of the present invention.

[0065] Figure 2 This is a flowchart of step 1 in a method for constructing an emergency accident case-assisted decision-making system based on a large model, as described in an embodiment of the present invention.

[0066] Figure 3This is a flowchart of step 2 in a method for constructing an emergency accident case-assisted decision-making system based on a large model, as described in an embodiment of the present invention. Detailed Implementation

[0067] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0068] Example 1:

[0069] This embodiment provides a method for constructing an emergency accident case-based decision support system based on a large model, such as... Figure 1 As shown, it includes:

[0070] Step 1: Read the target business domain, analyze the target business domain based on the target big model, and store the analysis results in the vector database;

[0071] Step 2: Retrieve historical accident cases, match the analysis results in the vector database with the historical accident cases according to the target large model, and import the matching results into the graph database;

[0072] Step 3: Obtain several question-answer pairs and the relationships between them in the target business domain. At the same time, use the relationships between the question-answer pairs as an evaluation dataset and save them to the vector database simultaneously.

[0073] Step 4: Based on the target large model, process several question-answer pairs and evaluation datasets using vector database and graph database, and construct an emergency accident case auxiliary decision-making system based on the processing results.

[0074] In this embodiment, the target business domain may include: monitoring and early warning domain, command and rescue domain, decision support domain, etc.

[0075] In this embodiment, analyzing the target business domain may include: analyzing text information corresponding to the target business domain, and then obtaining the target entities, initial relationships between target entities, conceptual entities, and conceptual relationships between conceptual entities (i.e., the analysis results).

[0076] In this embodiment, retrieving historical accident cases and matching the analysis results in the vector database with historical accidents according to the target large model can include: obtaining a certain number of historical accident cases through third-party public opinion information services or by using information collection technology, and automatically extracting and matching them according to the established conceptual entities and conceptual relationships using the prompting engineering technology of the large model.

[0077] In this embodiment, prompting engineering is used to generate several question-answer pairs in the local large model, covering business areas such as monitoring and early warning, command and rescue, and decision support. The relationship between these question-answer pairs is given as an evaluation dataset and is synchronously stored in the vector database. The question-answer pair is formed between questions and answers in the target domain, and the relationship between the question-answer pairs represents the relationship between the questions and answers.

[0078] In this embodiment, the vector database is a database used to store entity-relationship-entity pairs and concept entity-relationship-concept entity pairs in the target business domain, as well as the initial relationships between them. Here, a concept entity can be, for example, a car, and an entity can be a car corresponding to various brands, such as Mercedes-Benz.

[0079] In this embodiment, the graph database is a database used to store concept graphs generated based on entity-relationship-entity pairs and concept entity-relationship-concept entity pairs.

[0080] In this embodiment, the evaluation dataset can be the specific relationship between question-answer pairs, where the question-answer pairs are the input content that needs to be used for decision support in emergency incident cases.

[0081] The working principle and beneficial effects of the above technical solution are as follows: By analyzing the target business domain through a large target model, an accurate and effective vector database is constructed. The analysis results are then matched with historical accident cases to effectively determine the specific relationships between entities, relations, and entity pairs within the target business domain. Based on these determined relationships, a graph database is constructed to facilitate the subsequent construction of an emergency accident case-based decision support system. Secondly, by analyzing several question-and-answer pairs and the relationships between them using the obtained vector and graph databases, the emergency accident case-based decision support system is effectively constructed, improving the efficiency of accident case database construction. Furthermore, by periodically introducing relevant accident case investigation reports, question-and-answer pairs are extracted as an evaluation benchmark database. Based on similarity analysis, the question-and-answer results for constructing the target knowledge base are evaluated. The relevant entity content in the accident case database is revised based on the results, achieving automatic calibration and updating of the accident case database, thus ensuring the comprehensiveness and reliability of the emergency accident case-based decision support system.

[0082] Example 2:

[0083] Based on Example 1, this example provides a method for constructing an emergency accident case-based decision support system, such as... Figure 2 As shown, step 1, which involves reading the target business domain and analyzing it based on the target large model, includes:

[0084] S101: Read the target business domain and determine the text information of the target business domain;

[0085] S102: Analyze and process the text information of the target business domain based on the target big model, and extract the target entities in the target business domain and the initial relationships between the target entities;

[0086] S103: Classify the target entities based on the target big model, create conceptual entities based on the classification results, analyze the conceptual entities based on the target big model, and determine the conceptual relationships between the conceptual entities;

[0087] S104: Store the target entity, the initial relationship between the target entities, the conceptual entity, and the conceptual relationship between the conceptual entities into the vector database.

[0088] In this embodiment, a business domain is selected, relevant laws, regulations, national standards and industry standards related to the domain are collected, a local large model running environment is built using open source large model, and prompting engineering technology is used to automatically analyze relevant text according to the conventional subject-verb-object structure, extract relevant entities and their initial relationships, and store these entities and initial relationships in a vector database.

[0089] Based on the extracted entities and relationships, we continue to use the prompting engineering techniques of the large model to carry out similarity and deduplication analysis, as well as classification analysis. For entities that are automatically classified into the same category, we use the classification name of the large model as conceptual entities, retain the mapping relationship between conceptual entities and specific entities, and store these conceptual entities and entities in the vector database.

[0090] Based on the specific entities under the conceptual entity classification, the relationships between them and other conceptual entities are further classified using the hinting engineering technique of the large model. The names of the classified relationships are used as the conceptual relationships between the aforementioned conceptual entities. The mapping relationship between conceptual relationships and specific relationships is retained, and these conceptual relationships and relationships are stored in the vector database.

[0091] In this embodiment, the vector database retains: the mapping relationship between conceptual entities and target entities, and the mapping relationship between conceptual relationships and initial relationships.

[0092] In this embodiment, before classifying the target entity based on the large model, the following steps are also included:

[0093] Similarity analysis is performed on target entities based on a large model to obtain the similarity between target entities;

[0094] Obtain a baseline similarity (pre-defined to measure whether target entities are consistent), compare the similarity between target entities with the baseline similarity, and determine the duplicate target entities whose similarity between target entities is greater than the baseline similarity based on the comparison results;

[0095] The system performs deduplication on duplicate target entities and classifies them based on the target entity model according to the deduplication results.

[0096] In this embodiment, the initial relationship is used to characterize the association between target entities.

[0097] The beneficial effects of the above technical solution are: by analyzing and processing textual information in the target business domain, it effectively determines the target entities and their initial relationships, as well as the conceptual entities and their conceptual relationships, and effectively stores them in a vector database, thereby providing data support and assurance for building an emergency accident case-based decision support system.

[0098] Example 3:

[0099] Based on Example 1, this example provides a method for constructing an emergency accident case-based decision support system, such as... Figure 3 As shown, in step 2, historical accident cases are retrieved, and the analysis results in the vector database are matched with the historical accidents according to the target large model. The matching results are then imported into the graph database, including:

[0100] S201: Retrieve multiple historical accident cases based on a preset method and obtain the conceptual entities and conceptual relationships from the vector database;

[0101] S202: Based on the target large model, historical accident cases are matched according to conceptual entities and conceptual relationships to generate corresponding conceptual entity-conceptual relationship-conceptual entity pairs;

[0102] S203: Revise the concept entity-concept relationship-concept entity pair and import it into the graph database.

[0103] In this embodiment, revising the concept entity-concept relationship-concept entity pair and importing it into the graph database includes: retrieving the specific entities under the concept entity in the vector database; matching the specific entities under the concept entity according to classification to determine the category similarity between the specific entities under the concept entity and the concept entities in the concept entity-concept relationship-concept entity pair; if the category similarity between a specific entity under a concept entity and the concept entities at both ends of the concept relationship in the concept entity-concept relationship-concept entity pair is greater than a preset threshold, then the concept entity is replaced with a specific entity, and the revision of the concept entity-concept relationship-concept entity pair is completed based on the replacement result to obtain a specific entity-specific relationship-specific entity pair; and all specific entity-specific relationship-specific entity pairs are sequentially imported into the graph database based on the replacement result.

[0104] The working principle and beneficial effects of the above technical solution are as follows: A certain number of historical accident cases are obtained through third-party public opinion information services or by self-collection technology. Using large-scale modeling and prompting engineering techniques, the system automatically extracts and matches existing conceptual entities and relationships to generate corresponding conceptual entity-conceptual relationship-conceptual entity pairs. These newly generated pairs are then revised by retrieving specific entities from the vector database under the conceptual entity and comparing their category similarity with the conceptual entities at both ends of the relationship. If the category similarity between the two specific entities and the conceptual entity is greater than 90% (i.e., a preset threshold), the original specific relationship between the two specific entities replaces the relationship based on the conceptual entity. All generated specific entity-specific relationship-specific entity pairs are then sequentially imported into a selected graph database. This helps ensure the accuracy and completeness of the graph database, facilitates the concretization of conceptual relationships, and thus promotes the effective application of the graph database.

[0105] Example 4:

[0106] Based on Example 1, this example provides a method for constructing an emergency accident case-based decision support system. Step 3 involves obtaining several question-answer pairs and the relationships between them within the target business domain, and simultaneously saving them to a vector database, including:

[0107] Based on the target large model and the preset prompts, several question-answer pairs are generated, and the relationships between the question-answer pairs are also generated.

[0108] The relationships between question-and-answer pairs are used as the evaluation dataset;

[0109] Several question-answer pairs and evaluation datasets are synchronously saved to a vector database.

[0110] In this embodiment, the preset prompting process is known in advance and is a strategy used to determine the relationship between entities and conceptual entities.

[0111] In this embodiment, prompting engineering is used to generate several question-answer pairs that cover business domains such as monitoring and early warning, command and rescue, and decision support (i.e. target business domains) from the local large model, and the relationship between these question-answer pairs is given as an evaluation dataset, which is synchronously stored in the vector database.

[0112] The beneficial effects of the above technical solution are: by generating several question-answer pairs and the relationships between them based on preset prompts through the target large model, and storing the question-answer pairs and their corresponding relationships in a vector database, the system effectively ensures the auxiliary decision-making for emergency accident cases and improves the efficiency of auxiliary decision-making for emergency accident cases.

[0113] Example 5:

[0114] Based on Example 2, this example provides a method for constructing an emergency accident case-assisted decision-making system based on a large model. Before classifying the target entities based on the large model, the method further includes:

[0115] Similarity analysis is performed on target entities based on a large model to obtain the similarity between target entities;

[0116] Obtain the baseline similarity, compare the similarity between target entities with the baseline similarity, and determine the duplicate target entities whose similarity between target entities is greater than the baseline similarity based on the comparison results;

[0117] The system performs deduplication on duplicate target entities and classifies them based on the target entity model according to the deduplication results.

[0118] In this embodiment, the baseline similarity is pre-set and used to measure whether the similarity between target entities meets the similarity criteria. It can be adjusted.

[0119] In this embodiment, the duplicate target entity can be the same target entity that exists between target entities with a similarity greater than the benchmark similarity.

[0120] The beneficial effects of the above technical solution are: it ensures the effectiveness of target entity classification and improves the efficiency and accuracy of target entity classification.

[0121] Example 6:

[0122] Based on Example 1, this example provides a method for constructing an emergency accident case-based decision support system. In step 4, based on the large model, several question-answer pairs and evaluation datasets are processed using a vector database and a graph database, and an emergency accident case-based decision support system is constructed based on the processing results, including:

[0123] Obtain several questions, perform semantic understanding on the questions based on the target large model, and extract specific entities i from the questions;

[0124] With the support of the vector database, the specific entity i in the problem is matched with the specific entities in the graph database to obtain the set k of the top ten specific entities with the highest similarity.

[0125] Retrieve all relations of the top ten specific entity set k, and generate a relation set R, where the relation set R contains multiple relations r;

[0126] A second similarity matching is performed based on the relationships between question-answer pairs and the relation set R, extracting the relation r with the highest similarity greater than a preset threshold. max ;

[0127] Based on the specific entity i and the relation r max Match specific entity I in the graph database;

[0128] Based on the target large model, semantic understanding is performed on the answers in the assessment dataset to obtain the entity set S corresponding to the answers in multiple assessment datasets, where the entity set S contains multiple entities s;

[0129] The entity set S corresponding to the answers in multiple assessment datasets is matched with the specific entity I using the third similarity method, and it is determined whether there is a similarity higher than the preset similarity threshold.

[0130] If there exists an entity s with the highest similarity score that is higher than the preset similarity threshold. max When that happens, the entity s corresponding to the answer in the evaluation dataset will be... max Replace entity I with the answer for the current specific entity i to form specific entity ir. max -Entities max and the specific entity ir max -Entities max Import to graph database;

[0131] Otherwise, the specific entity ir max - Import entity I into the graph database;

[0132] A decision support system for emergency accident cases was built based on the imported results.

[0133] The working principle and beneficial effects of the above technical solution are as follows: A local large-scale model is used to sequentially perform semantic understanding on the problem to extract entities, and then perform similarity matching with entities in the graph database with the support of a vector database to obtain the top 10 entities and all relations of these entities, denoted as relation set R. Next, the relations obtained in step 6 are used to perform similarity matching with the relations in R, selecting the relation with the highest similarity (above 90%, i.e., the preset similarity threshold), and obtaining the entity I at the other end of the graph database for this relation. Then, the local large-scale model is used to perform semantic understanding on the answers in the evaluation dataset to extract entities, and I is matched with these entities for similarity matching. For those with the highest similarity (above 90%), the entity extracted from the evaluation dataset answers is directly used to replace entity I in the graph database; otherwise, the new entity-relation-entity pair is imported into the graph database. This effectively ensures the accuracy and effectiveness of the construction of the emergency accident case auxiliary decision-making system, providing a reliable and effective guarantee for the operation of the emergency accident case auxiliary decision-making system.

[0134] Example 7:

[0135] Based on Example 1, this example provides a method for constructing an emergency accident case-based decision support system, which also includes:

[0136] Retrieve accident case investigation reports from the pre-set emergency management database;

[0137] Obtain the runtime environment of the target large model, and analyze the accident case investigation report in the runtime environment of the target large model according to the preset technology to construct a training text dataset;

[0138] The target large model is trained based on the training text dataset, and the optimal adjustment parameters of the model are determined based on the model loss function;

[0139] Based on the optimal adjustment parameters of the model, the parameters of the target large model are adjusted to obtain the optimal target large model.

[0140] In this embodiment, accident case investigation reports from emergency departments are obtained through public information platforms of local governments. The aforementioned local large model operating environment is used to construct a training text dataset based on SFT or LoRA technology, and the large model is adjusted.

[0141] In this embodiment, the model loss function can be pre-set, such as the cross-entropy loss function. The parameters of the target large model are adjusted through the model loss function, thereby achieving the adjustment of the target large model.

[0142] The beneficial effects of the above technical solution are: by retrieving accident case investigation reports from the preset emergency department management database and analyzing them to obtain a training text dataset, the target large model is trained using the training text data, and the model loss function is used to enable it to have domain understanding capabilities for emergency accident cases.

[0143] Example 8:

[0144] Building upon Example 2, this example provides a method for constructing an emergency accident case-based decision support system based on a large model. It analyzes and processes textual information from the target business domain based on the target large model, extracting target entities and initial relationships between them, including:

[0145] The first text information of the target business domain is input into the target large model for format rewriting to obtain the second text information;

[0146] Based on the target large model, the second text information is identified, and multiple key information segments in the second text information are obtained;

[0147] Multiple key information segments are marked to obtain multiple key information tags, and the correlation between each key information tag and the tag of the other key information is calculated;

[0148] Obtain the relevance threshold range, and compare the relevance between the tag of each key information and the tags of the other key information with the relevance threshold range;

[0149] Each key information tag is sequentially associated with the tags of the other key information tags, and the tags whose relevance falls within the relevance threshold range are then associated with the tags.

[0150] Determine the relationships between each key piece of information based on the correlation results;

[0151] Based on the target big model, each key piece of information is identified to obtain the central word vector of the key information. At the same time, the industry vector dictionary of the target business domain is retrieved.

[0152] The central word vector is matched in the industry vector dictionary. If the central word vector exists in the industry vector dictionary, the central word vector is used as the target entity in the target business domain.

[0153] Otherwise, the central vocabulary vector will not be used as the target entity in the target business domain;

[0154] Based on the matching results, multiple target entities in the target business domain are obtained. At the same time, the relationships between each key piece of information are mapped in the corresponding target entities to obtain the initial relationship of each target entity.

[0155] In this embodiment, format rewriting can be performed by rewriting the first text information into a format that the large model can analyze and recognize, thereby obtaining the second text information.

[0156] In this embodiment, the first text information may be specific text content corresponding to monitoring, early warning, and command and rescue in the target business domain.

[0157] In this embodiment, the second text information may be data with the same content but different formats obtained by converting the first text information in the target business domain into a target large model.

[0158] In this embodiment, the key information segment can be the core data segment corresponding to the second text information determined after the second text information is identified by the target large model.

[0159] In this embodiment, the label can be a marker symbol that distinguishes different key information.

[0160] In this embodiment, the correlation degree is used to measure the degree of subordinate relationship between different key information. The larger the value, the stronger the correlation between key information.

[0161] In this embodiment, the correlation threshold range is pre-set and used as the basis for measuring whether different key information meets the correlation requirement; it can be adjusted.

[0162] In this embodiment, the central vocabulary vector can be the main data content of each key piece of information.

[0163] In this embodiment, the industry vector dictionary is pre-defined and used to represent all data text content existing in the target business domain, that is, the specific data information corresponding to different entities.

[0164] In this embodiment, the association mapping can be used to limit the time association relationship of different target entities based on the association relationship between key information corresponding to data content in the target business domain, wherein the key information corresponds to the target entity.

[0165] The working principle and beneficial effects of the above technical solution are as follows: by processing the text information of the target business domain, the target entity can be effectively acquired, improving the effectiveness and accuracy of target entity extraction. At the same time, by mapping the correlation between each key information in the corresponding target entity, the initial relationship of each target entity can be obtained more accurately, effectively ensuring the comprehensiveness and effectiveness of the construction of the emergency accident case auxiliary decision-making system.

[0166] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing an emergency accident case-based decision support system based on a large model, characterized in that, include: Step 1: Read the target business domain, analyze the target business domain based on the target big model, and store the analysis results in the vector database; Step 2: Retrieve historical accident cases, match the analysis results in the vector database with the historical accident cases according to the target large model, and import the matching results into the graph database; Step 3: Obtain several question-answer pairs and the relationships between them in the target business domain. At the same time, use the relationships between the question-answer pairs as an evaluation dataset and save them to the vector database simultaneously. Step 4: Based on the target large model, process several question-answer pairs and evaluation datasets using vector database and graph database, and construct an emergency accident case auxiliary decision-making system based on the processing results; Step 1 involves reading the target business domain and analyzing it based on the target large model, including: Read the target business domain and determine the text information of the target business domain; Based on the target big model, the text information of the target business domain is analyzed and processed to extract the target entities in the target business domain and the initial relationships between the target entities; The target entities are classified based on the target big model, and conceptual entities are created based on the classification results. The conceptual entities are then analyzed based on the target big model to determine the conceptual relationships between them. Store the target entities, the initial relationships between target entities, the conceptual entities, and the conceptual relationships between conceptual entities in the vector database; Based on the target big model, textual information in the target business domain is analyzed and processed to extract target entities and initial relationships between them, including: The first text information of the target business domain is input into the target large model for format rewriting to obtain the second text information; Based on the target large model, the second text information is identified, and multiple key information segments in the second text information are obtained; Multiple key information segments are marked to obtain multiple key information tags, and the correlation between each key information tag and the tag of the other key information is calculated; Obtain the relevance threshold range, and compare the relevance between the tag of each key information and the tags of the other key information with the relevance threshold range; Each key information tag is sequentially associated with the tags of the other key information tags, and the tags whose relevance falls within the relevance threshold range are then associated with the tags. Determine the relationships between each key piece of information based on the correlation results; Based on the target big model, each key piece of information is identified to obtain the central word vector of the key information. At the same time, the industry vector dictionary of the target business domain is retrieved. The central word vector is matched in the industry vector dictionary. If the central word vector exists in the industry vector dictionary, the central word vector is used as the target entity in the target business domain. Otherwise, the central vocabulary vector will not be used as the target entity in the target business domain; Based on the matching results, multiple target entities in the target business domain are obtained. At the same time, the association relationship between each key information is mapped in the corresponding target entity to obtain the initial relationship of each target entity. In step 4, based on the large model, several question-answer pairs and evaluation datasets are processed using vector and graph databases. Based on the processing results, an emergency incident case-based decision support system is constructed, including: Obtain several questions, perform semantic understanding on the questions based on the target large model, and extract specific entities i from the questions; With the support of the vector database, the specific entity i in the problem is matched with the specific entities in the graph database to obtain the set k of the top ten specific entities with the highest similarity. Retrieve all relations of the top ten specific entity set k, and generate a relation set R, where the relation set R contains multiple relations r; A second similarity matching is performed based on the relationships between question-answer pairs and the relation set R, extracting the relation r with the highest similarity greater than a preset threshold. max ; Based on the specific entity i and the relation r max Match specific entity I in the graph database; Based on the target large model, semantic understanding is performed on the answers in the assessment dataset to obtain the entity set S corresponding to the answers in multiple assessment datasets, where the entity set S contains multiple entities s; The entity set S corresponding to the answers in multiple assessment datasets is matched with the specific entity I using the third similarity method, and it is determined whether there is a similarity higher than the preset similarity threshold. If there exists an entity s with the highest similarity score that is higher than the preset similarity threshold. max When that happens, the entity s corresponding to the answer in the evaluation dataset will be... max Replace entity I with the answer for the current specific entity i to form specific entity ir. max -Entities max and the specific entity ir max -Entities max Import to graph database; Otherwise, the specific entity ir max - Import entity I into the graph database; A decision support system for emergency accident cases was built based on the imported results.

2. The method for constructing an emergency accident case-based decision support system according to claim 1, characterized in that, The vector database retains: the mapping relationship between conceptual entities and target entities, and the mapping relationship between conceptual relationships and initial relationships.

3. The method for constructing an emergency accident case-based decision support system according to claim 1, characterized in that, In step 2, historical accident cases are retrieved, and the analysis results in the vector database are matched with the historical accidents according to the target large model. The matching results are then imported into the graph database, including: S201: Retrieve multiple historical accident cases based on a preset method and obtain the conceptual entities and conceptual relationships from the vector database; S202: Based on the target large model, historical accident cases are matched according to conceptual entities and conceptual relationships to generate corresponding conceptual entity-conceptual relationship-conceptual entity pairs; S203: Revise the concept entity-concept relationship-concept entity pair and import it into the graph database.

4. The method for constructing an emergency accident case-based decision support system according to claim 3, characterized in that, Revise the concept entity-concept relationship-concept entity pairs and import them into the graph database, including: Retrieve specific entities under the concept entity in the vector database, match the specific entities under the concept entity according to the classification, and determine the category similarity between the specific entities under the concept entity and the concept entities in the concept entity-concept relationship-concept entity pair; If the category similarity between a concrete entity under a conceptual entity and the conceptual entities at both ends of the conceptual relationship in the conceptual entity-conceptual relationship-conceptual entity pair is greater than a preset threshold, then the conceptual entity is replaced with a concrete entity, and the conceptual entity-conceptual relationship-conceptual entity pair is revised based on the replacement result to obtain a concrete entity-conceptual relationship-conceptual entity pair. Based on the replacement results, all concrete entities-concrete relations-concrete entity pairs are sequentially imported into the graph database.

5. The method for constructing an emergency accident case-based decision support system according to claim 1, characterized in that, In step 3, several question-answer pairs and the relationships between them within the target business domain are obtained and synchronously saved to the vector database, including: Based on the target large model and the preset prompts, several question-answer pairs are generated, and the relationships between the question-answer pairs are also generated. The relationships between question-and-answer pairs are used as the evaluation dataset; Several question-answer pairs and evaluation datasets are synchronously saved to a vector database.

6. The method for constructing an emergency accident case-based decision support system according to claim 1, characterized in that, Before classifying the target entities based on the large model, the following steps are also included: Similarity analysis is performed on target entities based on a large model to obtain the similarity between target entities; Obtain the baseline similarity, compare the similarity between target entities with the baseline similarity, and determine the duplicate target entities whose similarity between target entities is greater than the baseline similarity based on the comparison results; The system performs deduplication on duplicate target entities and classifies them based on the target entity model according to the deduplication results.

7. The method for constructing an emergency accident case-based decision support system according to claim 1, characterized in that, Also includes: Retrieve accident case investigation reports from the pre-set emergency management database; Obtain the runtime environment of the target large model, and analyze the accident case investigation report in the runtime environment of the target large model according to the preset technology to construct a training text dataset; The target large model is trained based on the training text dataset, and the optimal adjustment parameters of the model are determined based on the model loss function; Based on the optimal adjustment parameters of the model, the parameters of the target large model are adjusted to obtain the optimal target large model.

Images