Emergency plan review method, system, electronic device and storage medium

Abstract

The present application provides an emergency plan review method, system, electronic device and storage medium, which parse the emergency plan to obtain the plan content; perform document segmentation on the emergency plan according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each entry and multiple titles divided according to the logical hierarchy of the emergency plan, and perform text cleaning and formatting to obtain target chapter content, target paragraph content, target entry content and target title; extract multiple keywords related to the emergency field from each target chapter content, each target paragraph content and each target entry content, and assign corresponding semantic categories to each chapter, each paragraph and each entry; add field information to each chapter, each paragraph and each entry, and construct a nested structure for each chapter, each paragraph and each entry with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree data model.

Description

Technical Field

[0001] The present application relates to the technical field of emergency response, and more specifically, to an emergency plan review method, system, electronic device, and storage medium. Background Art

[0002] With the rapid development of natural language processing technology, natural language processing technology has broad application prospects in text analysis of emergency plans.

[0003] Currently, the key to reviewing emergency plans is selecting a natural language processing model appropriate for the plan, establishing a dedicated text library, and improving model performance through training and parameter adjustment. However, natural language processing models often perform poorly in specific emergency response scenarios due to the high level of expertise required by deep learning models. Furthermore, the field of emergency response has its own unique terminology and expressions. This can lead to misunderstandings due to a lack of relevant expertise, resulting in poor accuracy, adaptability, professionalism, and intelligence in the review of emergency plans. Summary of the Invention

[0004] In view of this, the present application provides an emergency plan review method, system, electronic device and storage medium, with the purpose of improving the accuracy, adaptability, professionalism and intelligence of the review of emergency plans.

[0005] The first aspect of the present application provides an emergency plan review method, the method comprising:

[0006] Obtain the emergency plan submitted by the user, and parse the emergency plan to obtain the corresponding plan content;

[0007] Segmenting the emergency plan document according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item, and a plurality of titles divided according to the logical hierarchy of the emergency plan;

[0008] Performing text cleaning and formatting on the chapter content, the paragraph content, the item content, and the title to obtain target chapter content, target paragraph content, target item content, and target title;

[0009] Extracting multiple keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assigning corresponding semantic categories to each of the chapters, each of the paragraphs, and each of the entries;

[0010] Adding corresponding field information to each of the chapters, each of the paragraphs, and each of the items, and constructing a nested structure for each of the chapters, paragraphs, and items with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model;

[0011] Among them, the field information of the chapter includes the target title, target chapter content, semantic category and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content and keywords of the paragraph; the field information of the entry includes the target title, target entry content and keywords of the entry.

[0012] Optionally, obtaining the emergency plan submitted by the user and parsing the emergency plan to obtain corresponding plan content includes:

[0013] Obtaining the emergency plan submitted by the user and determining the file type of the emergency plan;

[0014] The emergency plan is parsed using the file parsing method that matches the file type to obtain the plan content of the emergency plan.

[0015] Optionally, extracting multiple keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assigning corresponding semantic categories to each chapter, each paragraph, and each entry, includes:

[0016] Performing word segmentation processing on the target content using a preset word segmentation rule; wherein the target content is a target chapter content, a target paragraph content, or a target item content;

[0017] Utilize the TF-IDF algorithm to extract multiple keywords related to the emergency field from the target content after word segmentation;

[0018] The target content is processed using a pre-trained semantic classification model to assign a corresponding semantic category to each target content.

[0019] Optionally, the method further includes:

[0020] Performing a logical matching conflict detection on the emergency plan according to a pre-built logical rule library to obtain a logical detection result of the emergency plan;

[0021] Calculating the semantic embedding vector of each paragraph in the emergency plan, and calculating the semantic similarity between every two paragraphs based on the semantic embedding vectors of each paragraph;

[0022] Constructing a semantic relationship graph between the paragraphs according to the semantic similarity between the paragraphs, and performing a semantic graph analysis on the emergency plan using the semantic relationship graph to obtain a corresponding semantic graph analysis result;

[0023] A document logic consistency score is performed based on the logic detection results and semantic graph analysis results of the emergency plan to obtain a logic score for the emergency plan.

[0024] Optionally, the method further includes:

[0025] Score the content and operability of the emergency plan;

[0026] Determining the weight corresponding to the logic score, the content score, and the operability score of the emergency plan from various pre-set weights;

[0027] extracting features of the logic score, features of the content score, and features of the operability score;

[0028] The weighted regression model is used to perform weighted integration based on the weights and characteristics of the logic score, the weights and characteristics of the content score, and the weights and characteristics of the operability score to obtain a comprehensive score for the emergency plan.

[0029] Optionally, the method further includes:

[0030] Obtaining chapter completeness detection results and structural standardization detection results of the emergency plan from the content score of the emergency plan;

[0031] Identify missing chapters and chapters with incorrect sequences in the emergency plan based on the chapter completeness test results and the structural normativeness test results of the emergency plan, and mark the missing chapters and chapters with incorrect sequences in the emergency plan;

[0032] Detecting paragraphs containing logical islands and semantic contradictions in the emergency plan based on the semantic graph analysis results;

[0033] Calculating the sentence embedding similarity of each sentence in the emergency plan, determining repeated paragraphs from the emergency plan based on the sentence embedding similarities of each sentence, and merging or marking each of the repeated paragraphs;

[0034] Time and resource description information is extracted from the emergency plan, and sections exceeding the response time and uncovered risk points are determined and marked in the emergency plan based on the time and resource description information.

[0035] A second aspect of the present application provides an emergency plan review system, the system comprising:

[0036] The parsing unit is used to obtain the emergency plan submitted by the user and parse the emergency plan to obtain the corresponding plan content;

[0037] A document segmentation unit is used to segment the emergency plan according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item and a plurality of titles divided according to the logical hierarchy of the emergency plan;

[0038] a text cleaning and formatting unit, configured to perform text cleaning and formatting on the chapter content, the paragraph content, the item content, and the title to obtain target chapter content, target paragraph content, target item content, and target title;

[0039] an extraction unit, configured to extract a plurality of keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assign a corresponding semantic category to each chapter, each paragraph, and each entry;

[0040] a construction unit, configured to add corresponding field information to each of the chapters, each of the paragraphs, and each of the items, and construct a nested structure for each of the chapters, paragraphs, and items with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model;

[0041] Among them, the field information of the chapter includes the target title, target chapter content, semantic category and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content and keywords of the paragraph; the field information of the entry includes the target title, target entry content and keywords of the entry.

[0042] Optionally, the parsing unit includes:

[0043] a determination unit, configured to obtain the emergency plan submitted by the user and determine the file type of the emergency plan;

[0044] The parsing subunit is used to parse the emergency plan using the file parsing method that matches the file type to obtain the content of the emergency plan.

[0045] The third aspect of the present application provides an electronic device, comprising: a processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is used to call and execute a program stored in the memory; and the memory is used to store a program, and the program is used to implement the emergency plan review method provided in the first aspect of the present application.

[0046] The fourth aspect of the present application provides a storage medium, in which computer-executable instructions are stored. The computer-executable instructions are used to execute the emergency plan review method provided in the first aspect of the present application.

[0047] The present application provides an emergency plan review method, system, electronic device and storage medium. After obtaining the emergency plan submitted by the user, first, the emergency plan is parsed to obtain the corresponding plan content; secondly, the emergency plan is segmented according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each item and multiple titles divided according to the logical hierarchy of the emergency plan, and the chapter content, the paragraph content, the item content and the title are cleaned and formatted to obtain the target chapter content, target paragraph content, target item content and target title; from the content of each target chapter, the target paragraph content, the target item content and the target title are obtained. Extract multiple keywords related to the emergency field from the content of each target paragraph and the content of each target entry, and assign corresponding semantic categories to each chapter, each paragraph and each entry; finally, add corresponding field information to each chapter, each paragraph and each entry, and construct a nested structure for each chapter, each paragraph and each entry with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model, and complete the review of the emergency plan. This can not only improve the accuracy of the review of the emergency plan, but also improve the adaptability, professionalism and intelligence of the review of the emergency plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0049] Figure 1 A flowchart of an emergency plan review method provided in an embodiment of the present application;

[0050] Figure 2 A schematic diagram of the structure of an emergency plan review system provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0054] See also Figure 1 , shows a flow chart of an emergency plan review method provided by an embodiment of the present application, the emergency plan review method specifically comprising the following steps:

[0055] S101: Acquire the emergency plan submitted by the user, and parse the emergency plan to obtain the corresponding plan content.

[0056] During the specific execution of step S101, when a user-submitted emergency plan is detected, the emergency plan can be obtained and parsed according to its file type to extract the corresponding plan content from the emergency plan. The plan content includes each chapter, each paragraph, each item, and multiple titles in the emergency plan.

[0057] In an embodiment of the present application, a file parsing method corresponding to each file type can be pre-set so that after obtaining the emergency plan submitted by the user, the file type of the emergency plan can be determined, and the emergency plan can be parsed using the file parsing method matching the file type to obtain the plan content of the emergency plan.

[0058] It should be noted that the file type can be a PDF file type or a Word file type, which is not limited in this embodiment of the present application.

[0059] In some embodiments, if the file type of the emergency plan is a PDF file type, then the process of parsing the emergency plan using a file parsing method that matches the file type can be: analyzing the emergency plan to determine the complexity of the emergency plan; judging whether the complexity of the emergency plan is greater than a preset complexity threshold; if the complexity of the emergency plan is greater than the preset complexity threshold, it can be considered that the complexity of the emergency plan is relatively high, that is, the emergency plan can be determined to be a PDF document with complex typesetting. At this time, the position of the text block in the emergency plan (determined based on the corresponding coordinates) can be used to maintain the paragraph order of the emergency plan, and a parsing method based on pdfplumber can be used to directly read the text layer content in the emergency plan, thereby extracting the plan content containing each chapter, each paragraph, each entry, multiple titles and page numbers of the emergency plan; if the complexity of the emergency plan is not greater than the preset complexity threshold, it can be considered that the complexity of the emergency plan is relatively low. At this time, a parsing method based on pdfplumber can be used directly to directly read the text layer content in the emergency plan, thereby extracting the corresponding plan content.

[0060] It should be noted that when the file type of the emergency plan is a PDF file type, if there is a table in the emergency plan, the row and column relationship of the table can be extracted, and the table structure of the table can be rebuilt based on the extracted row and column relationship; that is, in addition to the chapters, paragraphs, entries, titles and page numbers of the emergency plan, the plan content can also include the corresponding table structure.

[0061] In other embodiments, if the file type of the emergency plan is a Word file type, the process of parsing the emergency plan using a file parsing method that matches the file type can be: parsing the emergency plan based on python-docx, extracting each paragraph, multiple entries, multiple titles, page numbers, and tables in the emergency plan, and identifying the style hierarchy in the emergency plan (such as Heading1, Heading2), parsing the chapter titles in the emergency plan and marking the corresponding chapter levels to extract each chapter in the emergency plan; finally, reading the table content of the extracted table row by row and column by column to obtain the plan content including each chapter, each paragraph, multiple entries, multiple titles, page numbers, and tables in the emergency plan.

[0062] S102: Segment the emergency plan document according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item and a plurality of titles divided according to the logical hierarchy of the emergency plan.

[0063] During the specific execution of step S102, after obtaining the emergency plan content, the emergency plan can be document segmented according to the chapters, paragraphs, items and titles in the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item and multiple titles divided according to the logical hierarchy of the emergency plan.

[0064] In an embodiment of the present application, the logical hierarchy of the emergency plan can be extracted first, and the extracted logical hierarchy can be stored; wherein the logical product is a tree structure, and the logical hierarchy includes the chapter hierarchy and the hierarchical relationship between the title and the text.

[0065] It should be noted that by extracting the logical hierarchy of the emergency plan's chapters, paragraphs, items and titles and the main text within the emergency plan, the original logical framework of the emergency plan can be retained.

[0066] In an embodiment of the present application, when the file type of the emergency plan content is a PDF file type, the chapter content of each chapter in the emergency plan (for example, "Chapter 1" or "1.1", etc.), the paragraph content of each paragraph, the entry content of each entry, and multiple titles can be identified according to information such as the font size, bold features, and position of the chapters in the plan content.

[0067] In the case that the file type of the emergency plan content is a Word file type, the title hierarchy in the emergency plan can be identified through the corresponding paragraph style (such as the Heading level), and the segmentation points of multiple titles in the emergency plan are determined according to the obtained title hierarchy by optimizing the segmentation rules, and the title segmentation points are used to extract the title corresponding to each title from the emergency plan;

[0068] It should be noted that segmentation rules can be dynamically adjusted based on document styles to accommodate enterprise-defined document formats (file types). Segmentation rules can be designed based on common emergency plan template formats and specific matching rules (e.g., "risk assessment" or "resource allocation").

[0069] It should also be noted that if the emergency plan includes table content or table structure, the table information corresponding to the table in the emergency plan can be separated from the emergency plan based on the table content or table structure. Specifically, regular expressions can be used to match the table item number and list format (such as "one, two, three" or "1.1, 1.2") based on the table content or table structure.

[0070] S103: Clean and format the chapter content, paragraph content, item content, and title to obtain target chapter content, target paragraph content, target item content, and target title.

[0071] In the specific process of executing step S103, after the emergency plan is segmented to obtain the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each entry and multiple titles in the emergency plan, the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each entry and multiple titles are noise cleaned, terminology retained and formatted to obtain the target chapter content of each chapter, the target paragraph content of each paragraph, the target entry content of each entry and multiple target titles.

[0072] It should be noted that retaining terminology while performing noise cleaning and formatting can not only remove irrelevant information and improve data quality, but also avoid mistaken cleaning of content containing important information.

[0073] In actual application, you can delete irrelevant symbols in chapter content, paragraph content, entry content, titles and tables (such as "...", "[ ]"), blank lines, headers and footers, etc.) and standardize punctuation marks (such as full-width and half-width) to complete the corresponding noise cleaning operation; correct common formatting anomalies in the paragraph content of each paragraph (such as mis-segmentation problems), restore the structure of the table and the logical order between each paragraph to complete the corresponding formatting processing; in the process of noise cleaning and formatting, you can use the pre-set emergency field vocabulary to retain the chapter content, paragraph content, entry content, titles and tables in the tree structure. Keywords related to the emergency field (such as "risk assessment", "response time", etc.) can avoid mistaken cleaning of content containing important information. Among them, the emergency field vocabulary includes multiple keywords related to the emergency field.

[0074] S104: extracting multiple keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assigning corresponding semantic categories to each chapter, each paragraph, and each entry.

[0075] In the specific process of executing step S104, first, the target content can be segmented; second, multiple keywords related to the emergency field are extracted from each segmented word; and finally, a corresponding semantic category is assigned to the target content. The target content is the target chapter content, target paragraph content, or target item content.

[0076] Optionally, the target content is segmented using pre-set segmentation rules; multiple keywords related to the emergency field are extracted from the target content after segmentation using the TF-IDF algorithm; and the target content is processed using a pre-trained semantic classification model to assign corresponding semantic categories to each target content.

[0077] It should be noted that corresponding word segmentation rules can be pre-set, wherein the word segmentation rules can be a rule-based word segmentation method, for example, Jieba combined with a custom vocabulary, which is not limited in this embodiment of the present application.

[0078] It should also be noted that this application can dynamically adjust pre-set word segmentation rules based on keyword frequency and context. Keyword extraction algorithms can also be configured for different types of emergency plans (e.g., highlighting "emergency response" for disaster plans and "risk assessment" for safety plans).

[0079] In actual application, Jieba can be used in combination with a custom vocabulary to perform word segmentation on the target content; the TF-IDF value of each word in the content obtained after word segmentation using the TF-IDF algorithm is used to measure the importance of each word in the corresponding emergency plan, and words with TF-IDF values ​​greater than the preset TF-IDF value are used as keywords, that is, high-weight words are output as preliminary keywords; for each keyword, the semantic embedding vector of the keyword is calculated through the BERT model, and the cosine similarity between the keyword and each word is calculated based on the semantic embedding vector of the keyword and the semantic embedding vector of each word in the emergency field vocabulary, and the preliminary keyword with a cosine similarity greater than the preset cosine similarity threshold is used as the final keyword.

[0080] It should also be noted that the constructed emergency domain vocabulary can be dynamically updated in real time, and a weight priority strategy for specific words can be added (such as "resource allocation" takes precedence over "resources"), so as to cluster through semantic similarity (cosine similarity) and classify similar keywords (keywords with similar semantic categories) into the same group.

[0081] In summary, combining the TF-IDF algorithm and the BERT model can effectively balance the extraction of long-tail words and keywords, thereby effectively extracting multiple keywords related to the emergency field from the target tree structure.

[0082] In an embodiment of the present application, a Transformer model (such as BERT) can be trained using pre-constructed emergency domain corpus to obtain a semantic classification model that supports semantic classification in complex contexts; by using the pre-trained semantic classification model to process the target content, the semantic category corresponding to the target content is provided to provide contextual annotations for subsequent analysis.

[0083] It should be noted that the semantic categories may be “risk description”, “response measures”, “resource classification”, “emergency response”, “risk assessment”, etc., which are not limited in the embodiments of this application.

[0084] Specifically, the BERT model is used to load pre-trained weights, a classification head is added to perform semantic classification tasks, and emergency domain corpus is input into the BERT model. The BERT model prediction is trained based on the input emergency domain corpus to obtain a semantic classification model.

[0085] It should be noted that for target content that may involve multiple semantic categories (such as "risk assessment and resource allocation"), a multi-label classification model can also be trained to obtain a semantic classification model, thereby improving the shortcomings of the traditional single-label model, supporting multi-label semantic annotation of the target content, and solving the problem of semantic diversity of complex content; among them, the training method of the multi-label classification model is the same as the training method of the BERT model, and is not limited to this in the embodiments of this application.

[0086] In actual application, first, analyze whether the target content involves multiple semantic categories; if so, use the semantic category model trained by the multi-label classification model to predict the target content and obtain multiple semantic categories of the target content; if not, input the target content into the semantic classification model trained by the BERT model to predict the input target content and obtain the semantic category of the target content; finally, perform context consistency check on the content of each target chapter, each target paragraph and each target entry. If the check passes, assign corresponding semantic categories to each target chapter, each target paragraph and each target entry.

[0087] In summary, by using emergency domain corpus to fine-tune the BERT model to improve the context adaptation ability of the BERT model, a semantic classification model that can accurately identify semantic categories is obtained. At the same time, a multi-scenario classification model can be constructed (for example, the classification rules for disaster plans and medical plans are different). When it comes to target content with multiple semantic categories, a semantic classification model obtained by training a multi-label classification model is used to predict semantic categories. Finally, contextual information is introduced to improve the accuracy of semantic classification through cross-chapter, cross-paragraph, and cross-entry feature fusion.

[0088] S105: Add corresponding field information to each chapter, each paragraph, and each entry, and construct a nested structure for each chapter, each paragraph, and each entry with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model.

[0089] Among them, the field information of the chapter includes the target title, target chapter content, semantic category and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content and keywords of the paragraph; the field information of the entry includes the target title, target entry content and keywords of the entry.

[0090] In the specific process of executing step S105, for each chapter in the emergency plan, the target title corresponding to each chapter can be determined from the various target titles, and corresponding target chapter content, target title, keywords and semantic categories can be added to each chapter in the emergency plan; wherein, the keywords added here are keywords extracted from the target chapter content, and the added semantic categories are semantic categories assigned to the target chapter content.

[0091] For each paragraph in the emergency plan, the target title corresponding to each paragraph can be determined from the target titles, and corresponding target paragraph content, target title, keywords and semantic categories can be added to each paragraph in the emergency plan; among which, the keywords added here are keywords extracted from the target paragraph content, and the added semantic categories are the semantic categories assigned to the target paragraph content.

[0092] For each entry in the emergency plan, the target title corresponding to each entry can be determined from each target title, and corresponding target entry content, target title, keywords and semantic categories can be added to each entry in the emergency plan; among which, the keywords added here are keywords extracted from the target entry content, and the added semantic categories are the semantic categories assigned to the target entry content.

[0093] Finally, the pre-stored logical hierarchy of the emergency plan is obtained, and the chapters, paragraphs and entries with corresponding field information added are constructed into a corresponding nested structure according to the chapter hierarchy in the logical hierarchy to obtain the corresponding tree-like data model.

[0094] In the embodiment of the present application, after obtaining the corresponding tree data model, the tree data model can be stored in a NoSQL database (such as MongoDB) in JSON format. At the same time, a corresponding RESTful interface can be provided to support dynamic query and call of the segmented data of the stored tree data model.

[0095] In summary, adding corresponding semantic categories and keywords to chapters, paragraphs and entries makes the resulting tree-like data model more suitable for semantic analysis. In addition, by supporting dynamic query and call through the interface, it can meet the needs of multi-module collaboration.

[0096] The present application provides an emergency plan review method. After obtaining the emergency plan submitted by the user, first, the emergency plan is parsed to obtain the corresponding plan content; secondly, the emergency plan is segmented according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each entry and multiple titles divided according to the logical hierarchy of the emergency plan, and the chapter content, paragraph content, entry content and title are cleaned and formatted to obtain the target chapter content, target paragraph content, target entry content and target title; multiple keywords related to the emergency field are extracted from each target chapter content, each target paragraph content and each target entry content, and corresponding semantic categories are assigned to each chapter, each paragraph and each entry; finally, corresponding field information is added to each chapter, each paragraph and each entry, and a nested structure is constructed for each chapter, each paragraph and each entry with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model to complete the review of the emergency plan. This can not only improve the accuracy of the review of the emergency plan, but also improve the adaptability, professionalism and intelligence of the review of the emergency plan.

[0097] On the basis of the emergency plan review method provided in the above embodiment of the present application, the emergency plan review method provided in the embodiment of the present application may further include the following contents:

[0098] Furthermore, in an embodiment of the present application, a logical matching conflict detection can be performed on the emergency plan based on a pre-constructed logical rule library to obtain a logical detection result of the emergency plan; the semantic embedding vector of each paragraph in the emergency plan is calculated, and the semantic similarity between each two paragraphs is calculated based on the semantic embedding vector of each paragraph; a semantic relationship graph between each paragraph is constructed based on the semantic similarity between each paragraph, and the semantic relationship graph is used to perform semantic graph analysis on the emergency plan to obtain a corresponding semantic graph analysis result; a document logical consistency score is performed based on the logical detection result and the semantic graph analysis result of the emergency plan to obtain a logical score of the emergency plan.

[0099] It should be noted that by performing document logic consistency testing on emergency plans, semantic contradictions or content missing problems in emergency plans can be identified, for example, the lack of corresponding response measures for risk points.

[0100] In actual application, the logical consistency test of emergency response plans can be completed based on the explicit logic of rules and the implicit logic based on semantics, that is, based on rule engine + BERT semantic graph analysis. Specifically, a corresponding logical rule library can be pre-built, where the logical rule library can include multiple logical rules. The logical rules can require that each "risk assessment" paragraph corresponds to at least one "response measures" paragraph. A rule engine (such as Drools) is used to perform logical matching and conflict detection on the emergency response plan based on the pre-built logical rule library to obtain the logical test results of the emergency response plan.

[0101] The BERT model is used to calculate the semantic embedding vector of each paragraph in the emergency plan. The semantic similarity between each two paragraphs is then calculated based on their semantic embedding vectors. Based on the semantic similarity between each paragraph, a corresponding semantic relationship graph is constructed (e.g., the connection between "response measures" and "risk description"). The semantic relationship graph is then tested for logical islands (e.g., "risk description" has no corresponding response) and semantic contradictions (e.g., the coexistence of "resolved" and "unresponded" paragraphs), resulting in a semantic graph analysis of the emergency plan. Finally, the proportion of successful logical matches in the logic detection results is calculated using explicit rules. The average value of each semantic similarity is calculated using implicit rules. Distribution anomalies (e.g., excessively high or low) in the emergency plan are detected based on the calculated semantic average and the semantic graph analysis results. The resulting logical score for the emergency plan includes the proportion of successful logical matches and the detection of distribution anomalies.

[0102] It should be noted that this application combines display rules with semantic matching, which can solve the problem that implicit logic in emergency plans is difficult to detect. It also supports logic detection at the sentence, paragraph and chapter levels, taking into account both details and the overall picture. In other words, combining the interpretability of display rules with semantic matching can fully cover the logic detection needs of emergency plans; and, semantic fusion analysis of logic across paragraphs can also improve the logic detection capabilities of long documents (emergency plans), and further mining semantic relationships with graph neural networks can improve the accuracy of logic detection.

[0103] Furthermore, in an embodiment of the present application, in addition to scoring the document logic consistency of the emergency plan, the emergency plan can also be further scored for content and operability; wherein, the document logic consistency scoring refers to detecting the document logic consistency and rationality of the emergency plan and identifying possible contradictions; the content scoring refers to quantitative analysis of the integrity and standardization of the emergency plan; the operability scoring refers to evaluating the feasibility and actual value of the emergency plan.

[0104] In some embodiments, a corresponding standard chapter template can be pre-built, where the standard chapter template includes all core chapters, such as "Risk Assessment," "Response Steps," and "Resource Allocation." The process of scoring the content of an emergency plan can specifically include: first, using regular expressions or style features (such as title format) to match the chapters in the emergency plan against the standard chapter template, that is, to detect whether the emergency plan contains all core chapters, thereby obtaining the chapter completeness test results of the emergency plan; second, using keyword extraction technology (such as TF-IDF + domain vocabulary (emergency plan-specific vocabulary)), the keyword coverage of each chapter of the emergency plan is counted, and the comprehensiveness of important content in the emergency plan (such as key risk points and emergency resource allocation) is detected to obtain the content coverage of the emergency plan; finally, the reasonableness of the chapter order in the emergency plan is verified, and whether there are logical jumps between the chapters in the emergency plan (such as "Response Measures" appearing before "Risk Assessment") is detected to obtain the structural standardization check results of the emergency plan. The content score of the emergency plan can include the chapter completeness test results, content coverage, and structural standardization check results of the emergency plan.

[0105] It should be noted that the pre-built standard chapter templates can be dynamically adjusted according to different types of emergency plans (such as earthquakes, fires, and public health events); the emergency plan-specific vocabulary and context analysis can be used to enhance the ability to recognize professional terms in emergency plans.

[0106] In some embodiments, the emergency plan can be scored for operability based on a machine learning regression model. Specifically, time-related descriptions (such as descriptions related to response time and resource arrival time) are extracted from the emergency plan. A pre-built time rule library is used to verify whether the extracted descriptions meet industry standards to obtain the response timeliness of the emergency plan. Corresponding resource information (such as personnel, materials, equipment, etc.) is extracted from the emergency plan. The degree of matching between risk points and resource allocation is compared based on the extracted resource information and the knowledge graph to generate resource coverage. The actual operability of the emergency plan (such as "clear instructions" vs. "general description") is determined based on the extracted descriptions using a text classification model (such as Transformer). The operability score of the emergency plan includes the response timeliness, resource coverage, and actual operability of the emergency plan.

[0107] It should be noted that the use of time series analysis technology can realize the automatic verification of the time logic in the emergency plan; combining the knowledge graph to generate resource coverage can provide the interpretability of the operability score.

[0108] Furthermore, in an embodiment of the present application, after obtaining the logical score, content score and operability score of the emergency plan, the weight corresponding to the logical score of the emergency plan, the weight of the content score and the weight of the operability score can be determined from the pre-set weights; the characteristics of the logical score, the characteristics of the content score and the characteristics of the operability score can be extracted; and the weighted integration can be performed according to the weight and characteristics of the logical score, the weight and characteristics of the content score, and the weight and characteristics of the operability score through a weighted regression model to obtain a comprehensive score for the emergency plan.

[0109] In some embodiments, the weights corresponding to the logic score, content score, and operability score in each application scenario can be dynamically set according to the actual application scenario. For example, if the resource allocation plan focuses on the operability score and content score, that is, the weight of the operability score and content score of the emergency plan for the resource allocation application scenario is greater than the weight of the logic score.

[0110] In actual application, after obtaining the logic score, content score and operability score of the emergency plan, we can first determine the application scenario corresponding to the emergency plan, and according to the determined application scenario, determine the weight corresponding to the logic score of the emergency plan, the weight of the content score and the weight of the operability score from the pre-set weights; extract the features of the logic score, the features of the content score and the features of the operability score; input the weights and features of the logic score, and the weights and features of the content score into a weighted regression model (such as LightGBM), so that the weighted regression model performs weighted integration according to the input weights and features, and outputs a comprehensive score of the emergency plan.

[0111] It should be noted that after obtaining the comprehensive score of the emergency plan, a dimensional score bar chart, a total score pie chart, etc. can be generated based on the comprehensive score of the emergency plan, and the generated dimensional score bar chart and total score pie chart can be displayed.

[0112] In summary, the weights of each score can be dynamically adjusted according to the actual application scenario, and the contribution of each feature to the score can be analyzed in combination with SHAP values ​​(Shapley Additive Explanations), thereby enhancing the transparency of the score and making the resulting score more convincing.

[0113] Furthermore, in an embodiment of the present application, the chapter completeness detection results and the structural normalization detection results of the emergency plan can be obtained from the content score of the emergency plan; based on the chapter completeness detection results and the structural normalization detection results of the emergency plan, the missing chapters and the chapters with incorrect order in the emergency plan are identified, and the missing chapters and the chapters with incorrect order in the emergency plan are marked; based on the semantic graph analysis results, the paragraphs with logical islands and semantic contradictions in the emergency plan are detected; the sentence embedding similarity of each sentence in the emergency plan is calculated, and based on the sentence embedding similarity of each sentence, repeated paragraphs are determined from the emergency plan, and each repeated paragraph is merged or marked; time and resource description information is extracted from the emergency plan, and based on the time and resource description information, paragraphs that exceed the response time and uncovered risk points are determined and marked from the emergency plan.

[0114] It should be noted that the sentence embedding similarity of each sentence in the emergency plan is calculated through the BERT model, so that repeated paragraphs with similar content can be identified from the emergency plan based on the sentence embedding similarity of each sentence; and clustering algorithms (such as K-Means) are used to merge or mark repeated paragraphs with similar content.

[0115] It should also be noted that NER (Named Entity Recognition) technology can be used to extract time and resource description information in emergency plans; compare the actual response time in the time and resource description information with industry standards, and mark the paragraphs that exceed the response time; check whether the "resource description" and "risk assessment" in the time and resource description information match, and mark the uncovered risk points from the emergency plan based on the matching results.

[0116] In summary, by combining semantic analysis with explicit rules, new logical rules can be automatically generated, expanding applicability. Based on semantic similarity and contextual relationships, implicit logical issues can be located, enabling fine-grained problem tagging. Finally, scoring results can be combined with user interaction data to continuously optimize the accuracy of problem detection.

[0117] Furthermore, in an embodiment of the present application, for fixed-pattern problems (missing chapters), corresponding feedback templates are pre-set, and the priority of the feedback templates can be marked according to the importance of the problem (such as missing chapters > redundant content). For example, for the missing "Risk Assessment" chapter, the generated feedback template can be: "The document is missing the 'Risk Assessment' chapter. It is recommended to supplement the identification and analysis of potential risks, focusing on the risk type, probability of occurrence and possible impact." For non-fixed-pattern problems (such as paragraphs with logical islands and semantic contradictions), deep learning generation method (NLG) can be used to generate feedback suggestions for the problem.

[0118] Therefore, for problematic chapters or paragraphs in the emergency plan, if the problem in that chapter or paragraph is a fixed-pattern problem, the corresponding feedback template can be determined from the pre-set feedback templates. If the problem is not fixed-pattern, the problem and context of the chapter or paragraph can be obtained and input into the Transformer model (such as GPT-3 or T5). The Transformer model will then generate and output feedback suggestions for the chapter or paragraph based on the input problem and context. For example, for problems not fixed-pattern (logical conflict problems), the output feedback suggestion could be "It is recommended that 'risk disposal completed' and 'risk disposal in progress' be revised to consistent descriptions, and specific progress should be provided."

[0119] It should be noted that if there are tables or flow charts in the emergency plan, image generation technology can be used to generate visual charts of process optimization suggestions based on the tables or flow charts.

[0120] It should also be noted that a flow chart of Huayou can be generated based on the problem points. For example, after the "response time exceeds the standard" problem, an "optimized timetable" can be generated.

[0121] In summary, by dynamically generating responsive feedback suggestions based on the context of problematic chapters or paragraphs, the feedback can be made more targeted and practical. This application can support multiple feedback forms such as text, charts, and flowcharts to enhance user experience. This application can also optimize the model based on users' use of feedback suggestions (such as adoption rate).

[0122] Furthermore, in an embodiment of the present application, after obtaining the problems and feedback suggestions / feedback templates of the chapter or paragraph with problems in the emergency plan, the problems and feedback suggestions / feedback templates of the chapter or paragraph can be displayed in a graphical manner to help users quickly understand and locate the problems.

[0123] In some embodiments, problematic chapters / paragraphs / sentences can be directly marked in the emergency plan, and colors can be used to distinguish the types of problems. At the same time, the context of the problematic chapters / paragraphs / sentences can be highlighted; for example, chapters or paragraphs with logical contradictions can be marked in red; chapters / paragraphs / sentences with incomplete content can be marked in yellow; and chapters / paragraphs / sentences with timeliness issues can be marked in blue. This is not limited in the embodiments of the present application.

[0124] In other embodiments, the number of chapters with problems and the types of problems in the emergency plan are counted, and a corresponding problem distribution statistical graph (such as a bar graph) is generated based on the counted number of chapters and types of problems.

[0125] In other embodiments, a user-friendly interface is built based on Plotly or Streamlit. When it is detected that a user clicks on a chapter / paragraph / sentence with problems in the emergency plan, the problem and feedback suggestions / feedback template of the chapter / paragraph / sentence can be displayed to the user; at the same time, a comparison view of the problem and the suggestions can also be displayed to help the user quickly understand the direction of improvement.

[0126] In summary, this application can highlight and annotate at three levels: chapter, paragraph, and sentence, supporting fine-grained problem viewing; displaying the contextual association of problems through semantic graphs to improve the interpretability of problems; users can directly mark, confirm, or provide feedback on problems in the interface, forming a closed-loop optimization mechanism.

[0127] It can be seen that this application dynamically generates new problem detection rules by combining semantic analysis and logical rules, breaking through the limitations of traditional static rules; using deep learning models to dynamically generate improvement suggestions that are highly relevant to the context, enhancing the practicality of feedback; integrating text highlighting, statistical charts and semantic relationship diagrams to form a multi-dimensional feedback presentation form; using user interaction data to feed back the problem detection and feedback generation module to achieve continuous optimization of the system.

[0128] Furthermore, in an embodiment of the present application, a variety of emergency plans can be collected through user interaction and the addition of new documents, and the collected diverse emergency plans can expand the emergency field corpus and ensure data quality.

[0129] Specifically, a variety of emergency plan and user interaction data can be collected based on automatic data collection and active user feedback through log analysis, and the collected emergency plan and user interaction data can be processed in combination with rule cleaning and semantic annotation. The processing results can not only expand the emergency field corpus, but also effectively ensure data quality.

[0130] During actual application, the emergency plans uploaded by users and the feedback information generated by the review system are automatically recorded, and the user's operational data on the feedback (user interaction data) (such as "the problem has been solved" or "the suggestion is not applicable") are collected; regular expressions are used to filter invalid characters, blank paragraphs and noise data in the collected emergency plans, feedback information and operational data; semi-automatic annotation tools are introduced, and pre-trained models are used to generate preliminary annotation results for emergency plans, feedback information and operational data; public emergency plans can also be collected through domain crawlers, and enhancement technologies (such as text rewriting and synonym replacement) can be used to expand the emergency domain corpus using the collected emergency plans and preliminary annotation results for emergency plans, feedback information and operational data.

[0131] It should be noted that by recording user interaction data to form a closed-loop optimization mechanism, it is ensured that the emergency field corpus is continuously updated to meet user needs; combined with external crawling and automatic expansion technology, an emergency plan corpus covering multiple fields is constructed.

[0132] Furthermore, in an embodiment of the present application, the semantic classification model can be fine-tuned in real time to adapt to the ever-changing emergency plan structure and language expression; specifically, the transfer learning method is used to perform small-scale and efficient optimization of the semantic classification model.

[0133] Specifically, emergency domain samples with high diversity and strong representativeness (such as semantic classification samples and logical detection samples) are screened out from the emergency domain corpus, and the corresponding training data sets are constructed using the screened emergency domain samples. The semantic classification model is loaded, and the first few layers of the semantic classification model are frozen. The high-level parameters are optimized to adapt to the new task. The semantic classification model is fine-tuned using the constructed training data set using small batch efficient fine-tuning technology (such as LoRA and Adapter architecture). Finally, the incremental learning technology is combined to avoid the model forgetting historical data through small batch updates of the newly added data (the constructed training data set).

[0134] This demonstrates that this application improves the semantic classification model's ability to understand specific semantics in the emergency domain through a layer-by-layer optimization strategy. Using LoRA or Adapter technology reduces the computational cost of model updates, making it suitable for online optimization. It also ensures that the model retains its memory of old data while absorbing new data.

[0135] Furthermore, in the embodiment of the present application, the logic rule library can be dynamically updated through new data and user feedback to improve the adaptability and coverage of logic detection.

[0136] Specifically, frequent pattern mining algorithms (such as Apriori) are used to analyze common logical relationships in emergency plans uploaded by users, and to extract high-frequency logical relationships in the uploaded emergency plans (such as the mapping relationship between "risk points" and "response measures"); semantic similarity calculation (calculating the semantic embedding vector of the logical relationship based on the BERT model) is used to verify whether the logical relationship is applicable to emergency plans in various document formats, and high-quality logical relationships are screened through automated testing and user confirmation; corresponding logical rules are generated based on the screened logical relationships, and the generated logical rules are dynamically added to the logical rule library, while invalid or faulty logical rules in the logical rule library are eliminated, thereby completing the update of the logical rule library.

[0137] It can be seen that this application automatically generates new logical rules through frequent pattern mining, reducing manual intervention; combined with semantic analysis technology, it improves the generalization ability of logical rules, realizes real-time updating of the logical rule library, and adapts to the ever-changing user needs.

[0138] Furthermore, options such as "Suggestion not applicable" or "Problem solved" are provided in a user-friendly interface, user evaluations are collected, and the user's modifications to the generated feedback suggestions / feedback templates are recorded as samples for model optimization (Transformer (such as GPT-4) model); interaction logs are used to analyze whether users adopt feedback suggestions / feedback templates (such as modifying the marked problem paragraph); the adoption rates of different question types and suggestion types are counted, and the statistical adoption rates are used to optimize the model; the data adopted by users are marked as high-confidence samples for optimization of the model and logical rule base; user feedback is regularly introduced into the incremental learning process to achieve dynamic updating of the model and logical rules.

[0139] It can be seen that this application realizes adaptive learning based on the real-time optimization model of user behavior and the logic rule library; combines user behavior data, quantifies the quality of feedback samples and gives priority to the application of highly credible data; combines active feedback and passive behavior data to form a comprehensive optimization basis.

[0140] Based on the emergency plan review method provided in the above embodiment of the present application, the embodiment of the present application also provides an emergency plan review system, such as Figure 2 As shown, the emergency plan review system includes:

[0141] The parsing unit 21 is used to obtain the emergency plan submitted by the user and parse the emergency plan to obtain the corresponding plan content;

[0142] The document segmentation unit 22 is used to segment the emergency plan document according to the plan content, and obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item and multiple titles divided according to the logical hierarchy of the emergency plan;

[0143] A text cleaning and formatting unit 23 is used to clean and format the chapter content, paragraph content, item content and title to obtain target chapter content, target paragraph content, target item content and target title;

[0144] An extraction unit 24 is configured to extract a plurality of keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assign a corresponding semantic category to each chapter, each paragraph, and each entry;

[0145] The construction unit 25 is used to add corresponding field information to each chapter, each paragraph, and each item, and to construct a nested structure for each chapter, each paragraph, and each item with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model;

[0146] Among them, the field information of the chapter includes the target title, target chapter content, semantic category and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content and keywords of the paragraph; the field information of the entry includes the target title, target entry content and keywords of the entry.

[0147] The specific principles and execution processes of each unit in the emergency plan review system disclosed in the above embodiment of the present application are the same as the emergency plan review method disclosed in the above embodiment of the present application. Please refer to the corresponding parts of the emergency plan review method disclosed in the above embodiment of the present application, and no further details will be given here.

[0148] The present application provides an emergency plan review system. After obtaining the emergency plan submitted by the user, first, the emergency plan is parsed to obtain the corresponding plan content; secondly, the emergency plan is segmented according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the entry content of each entry and multiple titles divided according to the logical hierarchy of the emergency plan, and the chapter content, paragraph content, entry content and title are cleaned and formatted to obtain the target chapter content, target paragraph content, target entry content and target title; multiple keywords related to the emergency field are extracted from each target chapter content, each target paragraph content and each target entry content, and corresponding semantic categories are assigned to each chapter, each paragraph and each entry; finally, corresponding field information is added to each chapter, each paragraph and each entry, and a nested structure is constructed for each chapter, each paragraph and each entry with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model to complete the review of the emergency plan. This can not only improve the accuracy of the review of the emergency plan, but also improve the adaptability, professionalism and intelligence of the review of the emergency plan.

[0149] Optional, parsing unit, including:

[0150] A determination unit, used to obtain the emergency plan submitted by the user and determine the file type of the emergency plan;

[0151] The parsing subunit is used to parse the emergency plan using a file parsing method that matches the file type to obtain the content of the emergency plan.

[0152] Optionally, an extraction unit comprising:

[0153] A word segmentation processing unit is used to perform word segmentation processing on the target content using a preset word segmentation rule; wherein the target content is a target chapter content, a target paragraph content, or a target entry content;

[0154] An extraction subunit, used to extract multiple keywords related to the emergency field from the target content after word segmentation using the TF-IDF algorithm;

[0155] The semantic category assignment unit is used to process the target content using a pre-trained semantic classification model to assign a corresponding semantic category to each target content.

[0156] Optionally, the emergency plan review system provided in the embodiment of the present application further includes:

[0157] A logic detection unit is used to perform conflict detection of logic matching on the emergency plan according to a pre-built logic rule library to obtain a logic detection result of the emergency plan;

[0158] A semantic similarity calculation unit is used to calculate the semantic embedding vector of each paragraph in the emergency plan, and calculate the semantic similarity between every two paragraphs based on the semantic embedding vectors of each paragraph;

[0159] A semantic graph analysis unit is used to construct a semantic relationship graph between each paragraph based on the semantic similarity between each paragraph, and use the semantic relationship graph to perform semantic graph analysis on the emergency plan to obtain corresponding semantic graph analysis results;

[0160] The logic scoring unit is used to score the logical consistency of the document based on the logic detection results and semantic graph analysis results of the emergency plan to obtain the logic score of the emergency plan.

[0161] Optionally, the emergency plan review system provided in the embodiment of the present application further includes:

[0162] Content and operability scoring unit, used to score the content and operability of emergency plans;

[0163] A weight determination unit, configured to determine the weight corresponding to the logic score of the emergency plan, the weight of the content score, and the weight of the operability score from various pre-set weights;

[0164] A feature extraction unit, used to extract features of logic scoring, content scoring, and operability scoring;

[0165] The comprehensive scoring unit is used to obtain a comprehensive score of the emergency plan by weighted integration based on the weights and characteristics of the logical score, the weights and characteristics of the content score, and the weights and characteristics of the operability score through a weighted regression model.

[0166] Optionally, the emergency plan review system provided in the embodiment of the present application further includes:

[0167] A test result acquisition unit, used to acquire the chapter completeness test result and the structural standardization test result of the emergency plan from the content score of the emergency plan;

[0168] a chapter marking unit, for identifying missing chapters and chapters with incorrect sequences in the emergency plan based on the chapter completeness test results and the structural normativeness test results of the emergency plan, and marking the missing chapters and chapters with incorrect sequences in the emergency plan;

[0169] The paragraph detection unit is used to detect paragraphs with logical islands and semantic contradictions in the emergency plan based on the results of semantic graph analysis;

[0170] A calculation unit, configured to calculate the sentence embedding similarity of each sentence in the emergency plan, determine repeated paragraphs from the emergency plan based on the sentence embedding similarity of each sentence, and merge or mark each repeated paragraph;

[0171] The marking unit is used to extract time and resource description information from the emergency plan, and determine and mark the sections exceeding the response time and the uncovered risk points from the emergency plan based on the time and resource description information.

[0172] The present application embodiment provides an electronic device, such as Figure 3 As shown, the electronic device includes a processor 301 and a memory 302, the memory 302 is used to store the program code and data of the emergency plan review method, and the processor 301 is used to call the program instructions in the memory to execute the steps shown in the emergency plan review method in the above embodiment.

[0173] An embodiment of the present application provides a storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the emergency plan review method shown in the above embodiment.

[0174] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without making any creative efforts.

[0175] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0176] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0177] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for reviewing an emergency plan, characterized in that: The method comprises: Obtain the emergency plan submitted by the user, and parse the emergency plan to obtain the corresponding plan content; Segmenting the emergency plan document according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item, and a plurality of titles divided according to the logical hierarchy of the emergency plan; Performing text cleaning and formatting on the chapter content, the paragraph content, the item content, and the title to obtain target chapter content, target paragraph content, target item content, and target title; Extracting multiple keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assigning corresponding semantic categories to each of the chapters, each of the paragraphs, and each of the entries; Add corresponding field information to each of the chapters, each of the paragraphs, and each of the entries, and construct a nested structure for each of the chapters, paragraphs, and entries with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model; wherein the field information of the chapter includes the target title, target chapter content, semantic category, and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content, and keywords of the paragraph; and the field information of the entry includes the target title, target entry content, and keywords of the entry; Performing a logical matching conflict detection on the emergency plan according to a pre-built logical rule library to obtain a logical detection result of the emergency plan; Calculating the semantic embedding vector of each paragraph in the emergency plan, and calculating the semantic similarity between every two paragraphs based on the semantic embedding vectors of each paragraph; Constructing a semantic relationship graph between the paragraphs according to the semantic similarity between the paragraphs, and performing a semantic graph analysis on the emergency plan using the semantic relationship graph to obtain a corresponding semantic graph analysis result; Performing a document logic consistency score based on the logic detection results and semantic graph analysis results of the emergency plan to obtain a logic score for the emergency plan; Score the content and operability of the emergency plan; Determining the weight corresponding to the logic score, the content score, and the operability score of the emergency plan from various pre-set weights; extracting features of the logic score, features of the content score, and features of the operability score; The weighted regression model is used to perform weighted integration based on the weights and characteristics of the logic score, the weights and characteristics of the content score, and the weights and characteristics of the operability score to obtain a comprehensive score for the emergency plan.

2. The method according to claim 1, characterized in that The emergency plan submitted by the user is obtained and parsed to obtain the corresponding plan content, including: Obtaining the emergency plan submitted by the user and determining the file type of the emergency plan; The emergency plan is parsed using the file parsing method that matches the file type to obtain the plan content of the emergency plan.

3. The method according to claim 1, characterized in that The step of extracting multiple keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assigning a corresponding semantic category to each chapter, each paragraph, and each entry, includes: Performing word segmentation processing on the target content using a preset word segmentation rule; wherein the target content is a target chapter content, a target paragraph content, or a target item content; Utilize the TF-IDF algorithm to extract multiple keywords related to the emergency field from the target content after word segmentation; The target content is processed using a pre-trained semantic classification model to assign a corresponding semantic category to each target content.

4. The method according to claim 1, wherein The method further comprises: Obtaining chapter completeness detection results and structural standardization detection results of the emergency plan from the content score of the emergency plan; Identify missing chapters and chapters with incorrect sequences in the emergency plan based on the chapter completeness test results and the structural normativeness test results of the emergency plan, and mark the missing chapters and chapters with incorrect sequences in the emergency plan; Detecting paragraphs containing logical islands and semantic contradictions in the emergency plan based on the semantic graph analysis results; Calculating the sentence embedding similarity of each sentence in the emergency plan, determining repeated paragraphs from the emergency plan based on the sentence embedding similarities of each sentence, and merging or marking each of the repeated paragraphs; Time and resource description information is extracted from the emergency plan, and sections exceeding the response time and uncovered risk points are determined and marked in the emergency plan based on the time and resource description information.

5. An emergency plan review system, characterized in that: The system comprises: The parsing unit is used to obtain the emergency plan submitted by the user and parse the emergency plan to obtain the corresponding plan content; A document segmentation unit is used to segment the emergency plan according to the plan content to obtain the chapter content of each chapter, the paragraph content of each paragraph, the item content of each item and a plurality of titles divided according to the logical hierarchy of the emergency plan; a text cleaning and formatting unit, configured to perform text cleaning and formatting on the chapter content, the paragraph content, the item content, and the title to obtain target chapter content, target paragraph content, target item content, and target title; an extraction unit, configured to extract a plurality of keywords related to the emergency field from the contents of each target chapter, each target paragraph, and each target entry, and assign a corresponding semantic category to each chapter, each paragraph, and each entry; a construction unit, configured to add corresponding field information to each of the chapters, each of the paragraphs, and each of the items, and construct a nested structure for each of the chapters, paragraphs, and items with added field information according to the logical hierarchy of the emergency plan to obtain a corresponding tree-like data model; The field information of the chapter includes the target title, target chapter content, semantic category and keywords of the chapter; the field information of the paragraph includes the target title, target paragraph content and keywords of the paragraph; the field information of the entry includes the target title, target entry content and keywords of the entry; A logic detection unit, configured to perform a logic matching conflict detection on the emergency plan according to a pre-built logic rule library, and obtain a logic detection result of the emergency plan; A semantic similarity calculation unit, configured to calculate a semantic embedding vector for each paragraph in the emergency plan, and to calculate a semantic similarity between every two paragraphs based on the semantic embedding vectors of the paragraphs; a semantic graph analysis unit, configured to construct a semantic relationship graph between the paragraphs according to the semantic similarity between the paragraphs, and perform semantic graph analysis on the emergency plan using the semantic relationship graph to obtain a corresponding semantic graph analysis result; A logic scoring unit, configured to score the logical consistency of the document based on the logic detection result and the semantic graph analysis result of the emergency plan, and obtain a logic score for the emergency plan; A content and operability scoring unit, configured to score the content and operability of the emergency plan; A weight determination unit, configured to determine, from among various pre-set weights, a weight corresponding to a logic score, a weight for a content score, and a weight for an operability score of the emergency plan; a feature extraction unit, configured to extract features of the logic score, features of the content score, and features of the operability score; The comprehensive scoring unit is used to perform weighted integration according to the weights and characteristics of the logical score, the weights and characteristics of the content score, and the weights and characteristics of the operability score through a weighted regression model to obtain a comprehensive score of the emergency plan.

6. The system according to claim 5, characterized in that The parsing unit includes: a determination unit, configured to obtain the emergency plan submitted by the user and determine the file type of the emergency plan; The parsing subunit is used to parse the emergency plan using the file parsing method that matches the file type to obtain the content of the emergency plan.

7. An electronic device, characterized in that: include: A processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is configured to call and execute a program stored in the memory; The memory is used to store a program, and the program is used to implement the emergency plan review method as described in any one of claims 1-4.

8. A storage medium, characterized in that: The storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the emergency plan review method according to any one of claims 1 to 4.

Images