Semantic consistency-based aeronautical equipment environment test knowledge graph construction method and system

Abstract

The invention provides an aeronautical equipment environment test knowledge graph construction method and system based on semantic consistency, and relates to the technical field of aeronautical equipment test environment establishment and testing.The method comprises the steps that S1, an aeronautical equipment environment test standard text is obtained, and a standard knowledge system library is constructed; s2, extracting an aviation equipment environment test standard entity, an entity relationship and an entity attribute according to the large language model; s3, detecting and combining aviation equipment environment test standard entities with the same semantics according to the semantic vector similarity; and S4, according to the graph relationship and the attribute similarity, realizing aviation equipment environment test standard entity consistency detection. And S5, fusing results of the steps S3 and S4 to determine an aviation equipment environment test standard entity, and constructing an aviation equipment environment test standard knowledge graph. The method comprises the following steps: constructing a large language model to extract extraction results of standard entities, relationships and attributes of the aviation equipment environment test; and generating a high-quality and high-consistency aviation equipment environment test standard knowledge graph based on semantic consistency enhancement.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aviation equipment test environment establishment and testing technology, and particularly relates to an aviation equipment environment test knowledge graph construction method and system based on semantic consistency. BACKGROUND

[0002] As a core technology of structured knowledge representation, knowledge graph has shown great value in the fields of intelligent search, decision support and semantic understanding. It realizes the effective organization and reasoning of complex knowledge by converting the concepts, entities and their associations in the real world into a graph structure form of triples for entity-relation-entity or entity-property-value. In the field of aviation equipment environment test standards, the construction of knowledge graph can significantly improve the accessibility and application efficiency of standard information, such as supporting fast retrieval of standard clauses, compliance check of technical indicators and correlation analysis of cross-domain knowledge. However, aviation equipment environment test standards are usually published in PDF format, which is a non-structured or semi-structured document form that brings serious challenges to automated processing. Existing methods mostly rely on optical character recognition (OCR) technology to directly extract text content, but the OCR process is easily affected by document layout, formula symbols or table structure, introducing recognition errors, format confusion and even semantic distortion, which greatly reduces the accuracy of subsequent knowledge extraction. In addition, aviation equipment environment test standard texts contain a large number of professional terms, normative expressions and cross-chapter references, and traditional natural language processing (NLP) tools are difficult to effectively analyze their deep semantics, so a preprocessing mechanism that can balance text quality repair and content structuring is urgently needed.

[0003] Currently, the knowledge graph construction method faces many bottlenecks in the application of vertical fields: first, the information extraction method based on rules or traditional machine learning highly depends on manual annotation data and domain feature engineering, while in the professional field of aviation equipment environmental test standards, high-quality training data is scarce and the annotation cost is high, which leads to insufficient model generalization ability; second, the existing entity, relationship and attribute extraction of aviation equipment environmental test projects, test equipment and test parameters mostly uses independent modules, lacks a unified framework, and is easy to cause inconsistent extraction results, entity redundancy or relationship omission, for example, the entity recognition of aviation equipment environmental test projects, test equipment and test parameters may ignore synonym variants, the relationship extraction is difficult to capture implicit semantic association, and the attribute extraction cannot accurately process complex information in long text description. Although the large language model (LLM) provides a new way for zero-shot or few-shot information extraction with its powerful context understanding and generation ability, its direct application in aviation equipment environmental test standard knowledge graph construction still has significant defects: on the one hand, LLM is highly sensitive to input text quality, and OCR errors will accumulate and amplify model bias; on the other hand, aviation equipment environmental test standards involve a highly structured knowledge system, such as classification system, entity type and relationship setting, which requires the design of prompt words and knowledge guidance strategies, while existing methods lack a dynamic construction and adaptation mechanism for the domain knowledge system. In addition, the post-checking and fusion of the extraction results are often ignored, leading to problems such as entity misalignment, relationship breakage or attribute conflict in the knowledge graph. Therefore, developing an aviation equipment environmental test standard knowledge graph extraction and construction method that integrates text preprocessing, knowledge system construction, LLM-driven extraction and result optimization has become a key requirement to break through the bottlenecks of existing technologies. SUMMARY

[0004] In order to solve the above-mentioned deficiencies of the prior art, the purpose of the present application is to provide a semantic consistency-based aviation equipment environmental test knowledge graph construction method, which dynamically formulates an aviation equipment environmental test standard knowledge system library containing entity types, relationships and attribute settings for different fields of aviation equipment environmental test standards; constructs aviation equipment environmental test standard entity extraction methods, entity relationship extraction methods and entity attribute extraction methods using a large language model to obtain extraction results of aviation equipment environmental test standard entities, relationships and attributes; and generates a high-quality and highly consistent aviation equipment environmental test standard knowledge graph based on semantic consistency enhancement verification.

[0005] Specifically, in one aspect, the present application provides a semantic consistency-based aviation equipment environmental test knowledge graph construction method, which comprises the following steps: S1: Obtain the input text of the aviation equipment environmental test standard from the optical character recognition aviation equipment environmental test standard document; preset the entity type, entity relationship and entity attribute of the aviation equipment environmental test standard; the aviation equipment environmental test standard entity includes: test item, test equipment and test parameter; and a knowledge system library of the aviation equipment environmental test standard is constructed; S2: An aviation equipment environmental test standard entity extraction method is constructed using a large language model to extract the aviation equipment environmental test standard entity, entity relationship and entity attribute; and the aviation equipment environmental test standard entity extraction result is supplemented according to the entity relationship and entity attribute extraction result; S3: Aviation equipment environmental test standard entity semantic vectorization is performed, the aviation equipment environmental test standard entity semantic similarity is calculated, and entity classification and screening are performed; the aviation equipment environmental test standard entity pairs are detected using the semantic vector similarity, and the aviation equipment environmental test standard entities with the same semantics are merged; S4: Aviation equipment environmental test standard entity consistency detection is realized according to the graph relationship and attribute similarity; the relationship topological structure and attribute characteristics of the aviation equipment environmental test standard entity in the knowledge graph are used to determine the association relationship of the test equipment used by the test item, the test parameters possessed by the test item and the physical attribute characteristics contained in the test parameters, the aviation equipment environmental test standard entities with similar relationships are detected and merged, and the knowledge graph structure is consistent; S5: The aviation equipment environmental test standard entity clustering similarity is calculated; the aviation equipment environmental test standard entity semantic vector similarity result in step S3 and the aviation equipment environmental test standard entity graph relationship similarity result in step S4 are fused; the final aviation equipment environmental test standard entity list is determined, and the aviation equipment environmental test standard knowledge graph is obtained.

[0006] On the other hand, the present application provides a knowledge graph construction system of an aviation equipment environmental test knowledge graph construction method based on semantic consistency, which comprises: an entity recognition module, a relationship recognition module, an attribute recognition module and a data processing module; The entity recognition module extracts the prompt word template according to the aviation equipment environmental test standard entity, extracts the aviation equipment environmental test standard entity from the input text using a large language model, extracts all aviation equipment environmental test standard entities that meet the conditions, generates an aviation equipment environmental test standard entity list, and realizes the automatic recognition of key entities in the aviation equipment environmental test standard text; The relationship recognition module forms the relationship edge in the knowledge graph by associating the aviation equipment environmental test standard entities, and enhances the structured representation of knowledge; the semantic relationship between the aviation equipment environmental test standard entities is extracted from the aviation equipment environmental test standard text and the recognized aviation equipment environmental test standard entities; The attribute recognition module includes detailed information of the aviation equipment environment test standard entity, perfects the feature description of the aviation equipment environment test standard entity in the knowledge graph, thereby supporting more fine-grained knowledge query and analysis, and is responsible for extracting aviation equipment environment test standard entity attribute information from aviation equipment environment test standard texts and recognized aviation equipment environment test standard entities; The data processing module is responsible for integrating, cleaning and optimizing the semantic consistency of the aviation equipment environment test standard entity, relationship and attribute data output by the foregoing modules.

[0007] Compared with the prior art, the beneficial effects of the present application are as follows: (1) The present application realizes the aviation equipment environment test standard knowledge graph construction method based on a large language model, and aims at the problem of lack of dynamic adaptation of the aviation equipment environment test standard field knowledge system. Through the large language model, a knowledge system library about the entity types, relationships and attribute settings of aviation equipment environment test projects, test equipment and test parameters is dynamically constructed, the targeted guidance of aviation field equipment test is realized, the accuracy and generalization ability of knowledge extraction are improved, and the cost of manual annotation and the dependence on feature engineering are reduced.

[0008] (2) The present application aims at the inconsistency problem caused by the independence of the entity, relationship and attribute extraction modules of the aviation equipment environment test project, test equipment and test parameter. Based on the unified large language model framework, the modules are collaboratively extracted, high-precision information extraction is realized through customized prompt word templates, the semantic consistency and integrity between the entities, relationships and attributes of the aviation equipment environment test project, test equipment and test parameter are ensured, and the redundancy and omission of data are reduced.

[0009] (3) The present application aims at the redundancy and alignment problem caused by synonymous entities. Through the traversal matching and large language model semantic disambiguation in the post-processing step, the entities of the aviation equipment environment test project, test equipment and test parameter that are not matched are verified and fused, the alignment quality and overall consistency of the knowledge graph are effectively improved, and more efficient knowledge reasoning and application are supported. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 The present application is a semantic-consistent aviation equipment environment test knowledge graph construction method. Figure 2 The present application is a semantic-consistent aviation equipment environment test knowledge graph construction method. Figure 3 The present application is a semantic-consistent aviation equipment environment test knowledge graph construction method. Figure 4 The present application is a semantic-consistent aviation equipment environment test knowledge graph construction method. Figure 5 A large language model driven entity extraction mechanism diagram of the present application; Figure 6 A comparison chart of experimental results of entity extraction and relationship extraction of the present application and GraphRAG, LightRAG. DETAILED DESCRIPTION

[0011] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings.

[0012] The present application proposes a method for constructing an aviation equipment environmental test knowledge graph based on semantic consistency, as shown in Figure 1 The aviation equipment environmental test standard text is obtained, and a standard knowledge system library is constructed; the aviation equipment environmental test standard entity, entity relationship and entity attribute are extracted according to the large language model; the aviation equipment environmental test standard entities with the same semantics are merged according to the semantic vector similarity detection; the aviation equipment environmental test standard entity consistency detection is realized according to the graph relationship and attribute similarity; the aviation equipment environmental test standard entity is determined by fusion, and the aviation equipment environmental test standard knowledge graph is constructed; which includes the following steps: Step S1: Obtain the aviation equipment environmental test standard file and perform pretreatment; construct the aviation equipment environmental test standard knowledge system library.

[0013] Step S11: Obtain the aviation equipment environmental test standard file in portable document format (PDF), perform optical character recognition, and obtain the original recognized aviation equipment environmental test standard text ; use the preset prompt word to guide the large language model to correct and unify the format of the aviation equipment environmental test standard text , and obtain the pretreated aviation equipment environmental test standard input text . In the embodiment of the present application, the aviation equipment environmental test standard file in PDF format has tables, line graphs, flowcharts and other charts; for such charts, a chart information extraction step is performed to convert them into text that can be directly input into the large language model; specifically including the following steps: Step S111: divide the aviation equipment environmental test standard file to obtain a standard picture block; in the embodiment of the application, the PyMuPDF library function is used to convert the pages of the aviation equipment environmental test standard file into images, and the optical character recognition (OCR) technology is used for recognition. In the method, the PaddleOCR model is used to obtain all the standard picture blocks of the aviation equipment environmental test standard file, including the boundary box and the text content, and each text area is recorded as a standard text block. The part in the entire aviation equipment environmental test standard file that does not belong to any text area is extracted and filtered as a standard picture block. The PyMuPDF library function in the embodiment of the application is a Python library that is specially used for processing PDF, XPS and other documents. The core function used by it is to convert the PDF page into a raster image. The PaddleOCR model is an open source, multi-language optical character recognition (OCR) tool library that can accurately recognize the text in the image. The input image can output the text content, position and confidence, support multiple languages, tables and layout analysis, and have high precision and high performance.

[0014] Step S112: classify the standard picture blocks of the aviation equipment environmental test standard file; based on the standard picture blocks obtained in step S111, the standard picture blocks are classified; in the aviation equipment environmental test standard file, the commonly occurring chart types are tables, flowcharts and line charts, and the standard picture blocks are divided into the following four categories: tables, flowcharts, line charts and others. In the embodiment, a lightweight standard picture classification model is trained based on the deep convolutional neural network model ResNet50D model, and the classification result accuracy is not less than 95% in the test. Based on the classification result, different processing is performed on the table type standard picture blocks, the flowchart type standard picture blocks and the line chart type standard picture blocks. The core function of the deep convolutional neural network model ResNet-50D is to perform high-precision image feature extraction and classification; through the optimized residual module structure of the model, the complex features in the image can be effectively captured, and the model can be used as a high-performance backbone network in computer vision tasks such as image recognition, target detection and semantic segmentation, and has excellent representation ability and generalization performance.

[0015] Step S113: for the table type standard picture blocks in the aviation equipment environmental test standard file, the table type standard picture blocks are converted into a gray-scale image and binarized, the PaddleOCR model is used to recognize the table type standard picture block content to obtain the text, the text is classified by row, and then the classified aviation equipment environmental test standard text is input to the large language model.

[0016] Step S114: extract the flowchart type standard picture block information in the aviation equipment environmental test standard file, including the following steps: Step S1141: preprocessing and contour recognition of the flowchart standard picture block in the aviation equipment environmental test standard document; converting it into a gray-scale picture and binarizing it, and performing contour recognition based on the contour tracking algorithm Suzuki-Abe to obtain a contour set of the flowchart standard picture block . The Suzuki-Abe contour tracking algorithm is used to extract the contour boundary of a target object from a binary image; by inputting a binary image, a sequence of external and internal contours organized in a hierarchical structure is output, which can accurately describe the topological relationship of the object shape.

[0017] Step S1142: node detection of the flowchart standard picture block in the aviation equipment environmental test standard document; calculating the contour boundary box of each flowchart standard picture block ; using the Douglas-Peucker algorithm to realize contour polygon approximation to obtain the approximate vertex number of the flowchart standard picture block ; if the approximate vertex number and the angle is close to 90°, it is determined to be a rectangle, and the node of the rectangle represents a process or operation, which is used to represent a specific execution step or task; if it is a good fit for an ellipse or a circle, it is determined to be an oval starting or ending node, which is used to represent the beginning and end of the flow; if it is a rhombus, it has four groups of edges but the rotation angle is close to 45°, it is determined to be a judgment decision node, which usually contains a yes or no question, and guides the flow to different branches according to the result. Combining the two, the output of this step is: a node candidate set of the flowchart standard picture block ; is the boundary of the th flowchart standard picture block node; is the shape of the flowchart standard picture block. The Douglas-Peucker algorithm is a geometric simplification algorithm for vector curve data compression; by inputting a curve composed of ordered vertices, a simplified curve composed of key vertices is output, which maximally retains the original geometric shape within a specified tolerance.

[0018] Step S1143: edge detection and direction determination of the flowchart standard picture block in the aviation equipment environmental test standard document; based on the contour set of the binarized flowchart standard picture block in step S1141 and the node candidate set of the flowchart standard picture block in S1142 , the thin line structure is extracted based on the Zhang-Suen algorithm, the line region is compressed into a standard skeleton graph with a single-pixel width , the connected structure is retained, then the endpoint degree = 1 and the intersection degree ≥ 3 of the standard skeleton graph are identified according to the pixel neighborhood degree calculation, and the pixel path set of each line is extracted by the depth-first search algorithm DFS ; wherein, the Zhang-Suen thinning algorithm is a parallel thinning algorithm for binary images, which inputs a binary image and outputs its single-pixel-wide skeleton representation, and can effectively maintain the topological structure and connectivity of the original shape. The depth-first search algorithm DFS is an algorithm for traversing or searching tree and graph structures. The algorithm starts from a specified starting node and systematically explores each branch until the end, and its output is the depth-first traversal sequence of the node or the specific target node obtained by searching. The Douglas-Peucker algorithm is used near both ends of each path to identify the arrow shape, and the direction of the connecting line is determined by the direction vector of the triangular tip; the end points of the flowchart-like standard picture block line segment are mapped to the adjacent flowchart-like standard picture block node boundaries , if the distance between the end of the flowchart-like standard picture block line segment and the flowchart-like standard picture block node boundary is less than or equal to the minimum Euclidean distance of the flowchart-like standard picture block, the shortest Euclidean distance of the flowchart-like standard picture block is obtained as follows: ; wherein, is the shortest Euclidean distance of the flowchart-like standard picture block; is the flowchart-like standard picture block; is the standard node element in the set ; is the minimum value function; is the flowchart-like standard picture block node boundary; is the flowchart-like standard picture block number; is the flowchart-like standard picture block node boundary set.

[0019] According to the shortest Euclidean distance of the flowchart-like standard picture block, the path end points are matched to the corresponding nodes to realize the semantic association of the connecting line and the node. If there is an arrow in the connecting line, the direction is determined by the direction of the arrow; if the arrow is located at one end of the connecting line and points to another node, a directed edge is established, which is recorded as the preliminary edge set of the flowchart-like standard picture block with geometric path and arrow direction / con dence.

[0020] Step S1144: Determine the recognition text and node type of the flowchart in the aviation equipment environmental test standard document; for the node candidate set of the flowchart-like standard picture block obtained in step S1142 , containing the flowchart-like standard picture block node boundary and the flowchart-like standard picture block shape Crop the text region for each node of the flowchart-like standard picture block , and send it into the PaddleOCR model to get the recognition result as: the input text of the aviation equipment environmental test standard With the confidence , in the form of ; and finally get the node set of the text confidence of the flowchart-like standard picture block , where represents the geometric position transformation information of the text region of the flowchart-like standard picture block node in the image; if the shape of the flowchart-like standard picture block obtained in step S1142 is an ellipse or the text contains "start / finish", then , that is, the start or end node; if the shape of the flowchart-like standard picture block is a rhombus or the text contains a question / exclamation word "whether / whether satisfied / OK?", then , that is, the judgment or decision node; otherwise, it is considered as , that is, the process or operation node, and this step obtains the node set of the flowchart-like standard picture block .

[0021] Step S1145: Reconstruct the graph structure of the flowchart-like standard picture block in the aviation equipment environmental test standard file; according to the node set of the flowchart-like standard picture block obtained in step S1144 and the preliminary edge set of the flowchart-like standard picture block obtained in step S1143 , map the path of each geometric edge to the edge of the flowchart: if the path endpoints are mapped to the text confidence nodes of the flowchart-like standard picture block and , then the edge is established; if the path contains multiple inflection points, the polyline path point sequence is retained for visualization; for the judgment node to node connection judgment or decision branch, try to extract the near-text label of the flowchart-like standard picture block on the edge: do small text detection near the connection to find the edge label such as "yes / no", and the result is the near-text label of the flowchart-like standard picture block ; assemble the edge set of the flowchart-like standard picture block , each edge with a label and a geometric path; each edge of the flowchart-like standard picture block is usually represented as a four-tuple as: ; ; where is the edge of the th flowchart-like standard picture block, which constitutes the edge set of the flowchart-like standard picture block ; ​Text confidence node ID of the flowchart type standard picture block at the starting point; Text confidence node ID of the flowchart type standard picture block at the ending point, determined by node matching; Geometric path of the line, a sequence of pixel coordinates; Near-text label of the flowchart type standard picture block on the edge, such as "yes / no" and other OCR results.

[0022] Step S1146: Convert the graph structure of the flowchart type standard picture block in the aviation equipment environmental test standard file into natural language; according to the node set of the flowchart type standard picture block obtained in step S1145 and the edge set of the flowchart type standard picture block obtained in step S1145 , construct a directed graph of the flowchart type standard picture block , record the node set of the flowchart type standard picture block in json format and the edge set , input the preset template into the large language model, and the large language model will convert the directed graph into a text paragraph.

[0023] Step S115: Extract the information of the line chart type standard picture block in the aviation equipment environmental test standard file. It needs to be binarized first, and then morphological opening operation is applied to remove noise points and connect disconnected lines. Morphological opening operation is a nonlinear filtering operation for image processing; by inputting a binary image and a structure element, a processed image is output, which can effectively eliminate small noise, separate adhered objects and smooth the target contour, while keeping the main body geometry unchanged.

[0024] Step S1151: Detect the coordinate axes of the line chart type standard picture block in the aviation equipment environmental test standard file; use the probability Hough transform HoughLinesP to detect the line segments in the line chart type standard picture block, and perform DBSCAN clustering on the detected line segments according to the angle to identify the approximately horizontal and vertical line segment groups, and the longest horizontal line segment as the x-axis and the longest vertical line segment as the y-axis. The probability Hough transform HoughLinesP is used to detect straight line segments from an image; by inputting a binary image, a set composed of line segment start and end point coordinates is output, which can effectively locate the linear geometric structure in the image. DBSCAN is a density-based spatial clustering algorithm; by inputting a set of data points and neighborhood parameters, it can discover clusters of arbitrary shape and identify noise points, and its output is the cluster label or noise label to which each data point belongs.

[0025] Step S1152: Identify the scale label of the line graph type standard picture block in the aviation equipment environment test standard document; use the PaddleOCR model for text recognition; first determine the label area based on the position of the coordinate axis, and use the projection analysis method to locate the single character area; configure the optimization parameters for the PaddleOCR model, and set to only recognize numbers and common symbols to improve the recognition accuracy. The projection analysis method is a method for optical character recognition (OCR) text layout analysis; by counting the pixel projection distribution of the image in the horizontal and vertical directions, the text line is effectively located, the characters are segmented, and the layout structure is identified, laying a foundation for subsequent recognition.

[0026] Step S1153: Extract the data points of the line graph type standard picture block in the aviation equipment environment test standard document; use the contour tracking algorithm Suzuki-Abe to extract the line contour, and then apply the Zhang-Suen skeleton thinning algorithm to obtain a single-pixel-width line. Along the thinned line, sample points at fixed intervals, and determine the line turning points by nearest neighbor search.

[0027] Step S1154: Coordinate mapping of the line graph type standard picture block in the aviation equipment environment test standard document; establish the mapping relationship between the image pixel coordinate system and the actual data value; identify the scale label value and its position in the image, and use the linear interpolation formula to convert the pixel coordinates of the data points to actual values: actual value = minimum value + (pixel coordinate - pixel minimum value) x (maximum value - minimum value) / (pixel maximum value - pixel minimum value).

[0028] Step S1155: Data structure of the line graph type standard picture block in the aviation equipment environment test standard document; organize the extracted data into json format, design a structured prompt word template, embed the extracted data into it, and describe the data as natural language by a large language model.

[0029] Step S116: In the aviation equipment environment test standard document, for other types of images not covered by the foregoing process, directly input them into the large language model, and the large language model performs information analysis and content induction, and generates a summary natural language summary. In this step, the input aviation equipment environment test standard text Due to factors such as scanning quality, layout, and character similarity, the OCR recognition result often has various errors, such as confusion between the number "0" and the letter "O", paragraph title position disorder, redundant page numbers, and non-uniformity of the short horizontal line symbol in the standard number, etc. By comparing the aviation equipment environment test standard text The input large language model is used for error correction, correction of character recognition errors, adjustment of paragraph structure, deletion of meaningless fragmented text, and unification of symbols in standard numbers, for example, the hyphen "-" commonly used in standard documents is often recognized as "one" or "—" during recognition. Through the processing of the large language model, it is unified and standardized as a short horizontal line "-", to ensure the unity and uniqueness of the entity recognition results. Through the combination of rule guidance and semantic understanding, while maximizing the preservation of original text information, the structural standardization and character accuracy of the text are significantly improved.

[0030] Step S12: presetting an aviation equipment environmental test standard entity type list; the entity types include but are not limited to test items, test equipment, test parameters, test methods, standards, items, and industry fields; based on the pre-processed aviation equipment environmental test standard text as input, the aviation equipment environmental test standard text is classified by step S11, aviation equipment environmental test standard entity labels are extracted, and the aviation equipment environmental test standard entity types are determined.

[0031] Step S121: classifying the content of the aviation equipment environmental test standard; based on the pre-processed aviation equipment environmental test standard input text in step S11, the aviation equipment environmental test standard is classified according to the main content of the aviation equipment environmental test standard, and the aviation equipment environmental test standard text based on content classification is obtained; in the subsequent knowledge system construction process, the aviation equipment environmental test standard knowledge system constructed by the method is usually divided into two parts, one part is the common knowledge system of the aviation equipment environmental test standard, containing chapter structure, release subject and other information, as shown in Figure 3 is the common knowledge system ontology constructed based on the aviation equipment environmental test standard; the other part is the knowledge system of the aviation equipment environmental field for content.

[0032] Step S122: extracting aviation equipment environmental test standard entity labels; according to the classified aviation equipment environmental test standard in step S121, select several texts in the same standard, input the large language model, and combine the preset prompt words to construct the aviation equipment environmental test standard entity label and entity category table under this content classification, as shown in Table 1.

[0033] Table 1: Part of the entity label table Term Entity tag Term Entity tag Salt spray test chamber Test equipment type Vibration table Test equipment type Temperature Test parameter type Humidity Test parameter type High temperature test Test item type Low temperature test Test item type Step S123: Determine the entity types of the aviation equipment environmental testing standards; based on the entity tags of the aviation equipment environmental testing standards obtained in step S122, and combined with the input text of the aviation equipment environmental testing standards in step S11 and the preset prompt words input large language model, obtain the list of entity types in the aviation equipment environmental testing standards. List of entity types in the embodiments The format is as follows: The main body is the name of the subject appearing in the text, usually the drafting body, etc.; Test parameters: the physical quantities that need to be controlled or measured in the test; Test sample: the product or component being tested.

[0034] Step S13: Pre-set the entity relationships of the aviation equipment environmental testing standards; the types of entity relationships include, but are not limited to, including, containing, adopting, having characteristics, composed of, applicable to, and published; based on the list of aviation equipment environmental testing standard entity types obtained in step S12, and in conjunction with the entity type list, input the aviation equipment environmental testing standard text. Construct appropriate prompt words, input them into a large language model, and obtain the entity relationships of aviation equipment environmental testing standards. This relationship setting should include a relationship description, the entity setting domain of the relationship (i.e., the subject entity of the relationship), and the entity value domain of the relationship (i.e., the object entity of the relationship).

[0035] Table 2 Relationship Setting Data table Relationship name Relationship setting Subject type Entity type Contain Describe the hierarchy and composition structure between standard documents or chapters, or which specific operation steps a process or procedure contains. Standard, chapter, item, test method Standard, chapter, item, test method Have characteristics Describe the significant characteristics of a certain test item, sample, or equipment Test item, test sample, test equipment Item, test parameter Step S14: Pre-set the entity attributes of the aviation equipment environmental testing standard; entity attributes include physical quantities used to quantify and describe test conditions and results, including but not limited to equipment size, severity level, temperature, time, humidity, pressure, solution pH value, etc.; based on the list of aviation equipment environmental testing standard entity types obtained in step S12, construct prompt words, and input the entity type list and aviation equipment environmental testing standard text. Combined, they are input into a large language model to obtain the entity attributes of aviation equipment environmental testing standards. The entity attributes of the aviation equipment environmental testing standard in the embodiments. The format is as follows: Test / Method: representing various test methods, procedures or experiments described in the document; Parameter: representing various technical indicators, conditions or configurations involved in the experiment; Identification attributes: such as name, standard number, code, number, etc.

[0036] Step S15: List of entity types for environmental testing standards for aviation equipment obtained from steps S12-S14 Substantive Relationship of Environmental Testing Standards for Aviation Equipment Physical attributes of environmental testing standards for aviation equipment , the entity relationship attribute connection graph is constructed, the aviation equipment environmental test standard knowledge system library is constructed, as shown in Figure 4 , the entity relationship attribute connection graph related to the aviation equipment environmental test. Since there are many entity relationships and attributes in the actual extraction, Figure 4 only some key relationships and attributes are displayed as references.

[0037] Step S2: Using a large language model to construct an aviation equipment environmental test standard entity extraction method, extract aviation equipment environmental test standard entities, aviation equipment environmental test standard entity relationships, and aviation equipment environmental test standard entity attributes.

[0038] Step S21: Extracting aviation equipment environmental test standard entities using a large language model, as shown in Figure 5 ; first, input the prompt word template containing task description, entity type list, example, and output format into the large language model LLM together with the aviation equipment environmental test standard preprocessed text X, and finally output the entity type list in the aviation equipment environmental test standard containing entity names and entity types . Using a large language model to extract aviation equipment environmental test standard entities, output the entity type list in the aviation equipment environmental test standard containing entity names and entity types ; the entity type is predefined to include test items, test equipment, test parameters, test methods, standards, entries, and industry fields, etc. to describe aviation equipment environmental test entity types.

[0039] Step S211: Constructing an aviation equipment environmental test standard entity extraction prompt word template; the prompt words for extracting aviation equipment environmental test standard entities are mainly composed of the following parts: large language model task description, related field entity type list, input aviation equipment environmental test standard text , extraction example, and output format. The task description clearly needs to refer to the example, according to the entity type list in the aviation equipment environmental test standard in step S12 , extract all entities in the input national standard text, and return the results according to the output format. The entity type list in the aviation equipment environmental test standard is obtained in step S12, and the input aviation equipment environmental test standard text is obtained in step S11. The core content of this step is the construction of the prompt word, and the construction method of the entity extraction prompt word is set to , where the aviation equipment environmental test standard entity type list is the aviation equipment environmental test standard entity type list obtained in step S12, and the entity type prompt word represents the aviation equipment environmental test standard entity type list The combination with the template is as follows: Task: Refer to "Examples", extract entities in "Input Text" according to "Entity Type List", and return the result according to "Output Format".

[0040] Entity Type is: <input variable ET>.

[0041] Examples are: <Example 1>, <Example 2>...

[0042] Output format: like "XXX||YYY", where XXX is the entity name and YYY is the entity type. Do not output content outside the specified format.

[0043] Step S212: Use a large language model to extract aviation equipment environmental test standard entities in the aviation equipment environmental test standard text ; After the construction of the prompt word is completed, combine the entity type prompt word with the aviation equipment environmental test standard input text , and input it into the large language model. Record this process as , and record the output result as , which is: ; Among them, is the aviation equipment environmental test standard entity type list; is the large language model function; is the entity type prompt word; is the aviation equipment environmental test standard entity type list; is the aviation equipment environmental test standard input text.

[0044] When using a large language model to extract aviation equipment environmental test standard entities, the following conditions must be met: ; Among them, is the entity type extraction parameter, which refers to a set of variables or results that make the target function reach the maximum value; is the set E that reaches the maximum value in all possible sets E, which makes the function P(E|IT,ET) reach the maximum value; ; is the set of aviation equipment environmental test standard entity lists.

[0045] According to the given aviation equipment environmental test standard input text and the preset aviation equipment environmental test standard entity type list , the most probable aviation equipment environmental test standard entity sequence is obtained, where, representing the sequence of the highest probability of generating the aviation equipment environmental test standard entity under the known text and type constraints ; due to the autoregressive characteristics of the large language model, this joint probability can be decomposed into the product of the probability of generating each aviation equipment environmental test standard entity in the sequence generation process, specifically: ; wherein, is the likelihood of the list of aviation equipment environmental test standard entities ; is the name of the aviation equipment environmental test standard entity is the type of the aviation equipment environmental test standard entity is the number of entities in the sequence of aviation equipment environmental test standard entities ; is the list of all aviation equipment environmental test standard entities that have been generated before the generation of the i-th aviation equipment environmental test standard entity .

[0046] This formula describes the joint probability of generating the entire list of aviation equipment environmental test standard entities under the given conditions of the input text of the aviation equipment environmental test standard and the list of aviation equipment environmental test standard entity types . The large language model will select the sequence of aviation equipment environmental test standard entities with the highest probability under the given content as the result for output.

[0047] Step S213: parsing the result output by the large language model, specifically: ; wherein, is the sequence of aviation equipment environmental test standard entities is the parsing function of entity recognition, used to ensure the uniformity of the output result is the i-th aviation equipment environmental test standard entity.

[0048] Since the generation result of the large language model has a certain randomness, there is a certain probability that the prompt text such as "the following is the entity recognition result based on the reference text" will appear in the recognition result, and what is needed is only the list of aviation equipment environmental test standard entities in the form of binary tuples ; therefore, the recognition result needs to be parsed. In the parsing function , regular expressions are used for matching; the matching result obtained is a binary tuple in the form of "XXX||YYY", and the aviation equipment environmental test standard entity For: ; wherein, is the name of the n th entity of the aviation equipment environmental test standard; is the type of the n th entity of the aviation equipment environmental test standard.

[0049] Step S22: An aviation equipment environmental test standard entity relationship extraction method based on a large language model is constructed to extract aviation equipment environmental test standard entity relationships. The types of entity relationships include but are not limited to include, adopt, have characteristics, consist of, apply to, and publish. When extracting the attributes of test parameters and test items, the focus is on extracting the associated physical quantity values and their units, mainly including the values and units of dimensions, the values and units of temperatures, the values and units of humidities, the values and units of pressures, and the values and units of durations. A large language model is used to extract aviation equipment environmental test standard entity relationships; the relationship types are predefined as include, adopt, have characteristics, consist of, apply to, and publish, and the like, which describe the relationship types of entities in aviation equipment environmental tests.

[0050] Step S221: An aviation equipment environmental test standard entity relationship extraction prompt word template is constructed; the prompt words of the aviation equipment environmental test standard entity relationship part mainly consist of the following parts: a large language model task, entity extraction results, related field entity relationships, input aviation equipment environmental test standard text, extraction examples, and output format. In the task module, it is specified that the large language model needs to refer to examples, according to the entity extraction results and the relationship setting list, all entity relationships in the input national standard text are extracted, and the results are returned according to the output format. The entity extraction results are the aviation equipment environmental test standard entity sequence E obtained in step S21, the related field relationship setting list is the aviation equipment environmental test standard entity relationship RD obtained in step S13, the aviation equipment environmental test standard input text is the aviation equipment environmental test standard text obtained in step S11, the extraction examples are the aviation equipment environmental test standard entity relationship extraction examples obtained in step S12, and the output format is the aviation equipment environmental test standard entity relationship extraction results obtained in step S22. Determined by step S11.

[0051] Step S222: A large language model is used to perform entity relationship extraction on the aviation equipment environmental test standard text ; the goal of the relationship extraction task is to determine the relationship types between the aviation equipment environmental test standard entities in the aviation equipment environmental test standard input text , for each aviation equipment environmental test standard input text segment , in combination with the aviation equipment environmental test standard entities contained therein, to determine the relationship types between them , where indicates no relationship, i.e., there is no aviation equipment environmental test standard entity relationship between the two aviation equipment environmental test standard entities​Any relationship in the above table; relationship extraction is a function mapping problem, specifically: ; Wherein, is the entity relationship of the aviation equipment environmental test standard; represents no relationship.

[0052] For a given aviation equipment environmental test standard input text segment and the aviation equipment environmental test standard entity pair , through the entity relationship extraction function , the aviation equipment environmental test standard entity pair is outputted with the relationship type that the entity pair has in the text, or is outputted, which represents no relationship.

[0053] In implementation, the aviation equipment environmental test standard entity relationship obtained in S14 is combined with the template to form the prompt word, and this process is set as the entity relationship prompt word ; The specific implementation is as follows: "Task" extracts the relationship between entities according to the provided "national standard text" and "annotated entities", and outputs the result in the output format.

[0054] "Relationship extraction requirements" core relationship: must contain at least but not limited to the following relationship types: < relationship setting list >.

[0055] "Extraction principles" requirements: accuracy: relationships must be based on explicit descriptions or strong implications in the text, and cannot be invented; full coverage: try to find all entity pairs that meet the above relationship; chain extraction: if A contains B, B contains C, then (A, contains, B) and (B, contains, C) need to be extracted at the same time.

[0056] "Output format" is: each relationship object must be strictly output in the format of "AAA||XXX||BBB", where AAA represents the relationship subject, XXX represents the relationship type, and BBB represents the relationship object.

[0057] In the aviation equipment environmental test standard entity relationship extraction, the entity relationship prompt word , the aviation equipment environmental test standard input text and the aviation equipment environmental test standard entity list obtained in S21 are combined and input into the large language model, which is recorded as , and the output result is recorded as , which is specifically: ; in, The results of entity relationship extraction for environmental testing standards of aviation equipment; This is a prompt for entity relationships.

[0058] When extracting entity relations, large language models use the following formula: ; in, List of entity relationships for environmental testing standards for aviation equipment.

[0059] For a given input text of the environmental testing standard for aviation equipment List of entities for environmental testing standards for aviation equipment and the predefined list of entity relationships This will yield a list of entity relationships with the highest probability. .in, This indicates how to generate a list of entity-relationships given the input text, entities, and constraints. The likelihood of the target. In implementation, the large language model will perform a simple comparison of each aviation equipment environmental testing standard entity pair. Calculate which relation type it belongs to. Based on the probability, select the relation type with the highest probability.

[0060] The example contains a standard input text for environmental testing of aviation equipment. Aircraft Equipment Environmental Testing Standard Input Text Segment The following is an example: 11. Information that should be provided in relevant specifications. When the relevant specifications include this test, the following information should be provided as much as possible; the severity level is temperature and the number of cycles. In this example, there are two entities related to the environmental testing standards for aviation equipment: the chapter entity "11. Information that should be provided in relevant specifications" and the item entity "a. Severity level: temperature and number of cycles," denoted as follows: and The information to be provided in the relevant specifications (section 11) includes the relationship between "a. Severity: Temperature and Cycle Count", denoted as... This is the input text segment of the aviation equipment environmental testing standard. and entity pair Specifically, "11. Information that should be given in the relevant specifications" and "a. Severity: temperature and number of cycles", extracted through a relational extraction function. , and thus "Contains". For other unrelated entity pairs, such as if the text also contains the entity "b, initial detection" (an aviation equipment environmental testing standard), let's set it as... ,So It may output Because the relationship between "the information that should be given in the 11 related specifications" and "b, initial detection" is not explicitly mentioned in this text segment.

[0061] Step S223: Analyze the result output by the large language model, and perform the following transformation: ; is a parsing function for relationship recognition, used to ensure the uniformity of the output result. In the parsing function for entity relationship recognition , a regular expression is used for matching; the matching result is a triple in the form of "AAA||XXX||BBB", that is, the extraction result. Among them, each aviation equipment environmental test standard entity relationship is a triple: ; Among them, is the aviation equipment environmental test standard entity relationship triple; is the aviation equipment environmental test standard subject entity; is the aviation equipment environmental test standard entity relationship type; is the aviation equipment environmental test standard object entity.

[0062] Step S23: Construct an aviation equipment environmental test standard entity attribute extraction method based on a large language model, and extract aviation equipment environmental test standard entity attributes. Use a large language model to extract aviation equipment environmental test standard entity attributes; in this step, the focus is on extracting physical quantity attributes related to test conditions and result quantification, including but not limited to the numerical value and unit of size, temperature, humidity, pressure, and duration.

[0063] Step S231: Construct an entity attribute extraction prompt word template; the prompt words for extracting the aviation equipment environmental test standard entity attribute part mainly consist of the following parts: large language model task, entity extraction result, related field entity attribute, input aviation equipment environmental test standard text, extraction example, output format. In the task module, it is clear that the large language model needs to refer to the example, according to the entity extraction result and the attribute setting list, extract all aviation equipment environmental test standard entity attributes in the input national standard text, and return the result according to the output format. The entity extraction result is the aviation equipment environmental test standard entity relationship triple obtained in step S22, the related field attribute setting list is the aviation equipment environmental test standard entity attribute obtained in step S14, and the input aviation equipment environmental test standard text is obtained in step S11.

[0064] Step S232: using a large language model to process the aviation equipment environmental test standard text Entity attribute extraction is performed; in attribute extraction, the aviation equipment environmental test standard entity attributes obtained in step S14 need to be combined with the templates to form entity attribute prompts, and this process is set as entity attribute prompts ; the specific implementation is as follows: The "task" is to extract the attribute values of entities according to the provided "national standard text" and "annotated entities", and output the results in the output format.

[0065] "Extraction range and entity setting" please pay attention to and extract the attributes of the following entity types: < attribute setting list AD >.

[0066] "Output format" each attribute is output separately in the format of "entity name || attribute name || attribute value".

[0067] The entity attribute prompts , the aviation equipment environmental test standard input text , and the aviation equipment environmental test standard entity list obtained in step S21 are combined and input into the large language model, and this process is recorded as , and the output result is recorded as , which is specifically: ; Among them, is the aviation equipment environmental test standard entity attribute.

[0068] Step S233: parse the result output by the large language model and perform the following transformation, which is specifically: ; Among them, is the aviation equipment environmental test standard entity attribute list; is the attribute recognition parsing function, which is used to ensure the uniformity of the output result. In the attribute recognition parsing function , regular expressions are used for matching; the matching result is a triple in the form of "AAA||XXX||YYY", that is, the extraction result, where each aviation equipment environmental test standard entity attribute and is a triple, which is specifically: ; Among them, is the attribute item; is the subject entity for entity attribute extraction; is the entity attribute type;​ corresponding value of the entity attribute.

[0069] Step S24: For the list of aviation equipment environmental test standard entity types preliminarily extracted in step S21 , according to the aviation equipment environmental test standard entity relationship and attribute extraction results, the aviation equipment environmental test standard entity extraction results are supplemented.

[0070] Step S241: Traverse the aviation equipment environmental test standard entity relationship extraction results in step S22, count all subject entities and object entities, check whether they appear in the entity extraction results, if not, record and save; extract all subject entities and object entities from the aviation equipment environmental test standard entity relationship set to form the aviation equipment environmental test standard entity set in the relationship extraction results ; Among them, is the aviation equipment environmental test standard entity set in the relationship extraction results; is the aviation equipment environmental test standard subject entity; is the aviation equipment environmental test standard object entity.

[0071] For the list of aviation equipment environmental test standard entities , check whether each aviation equipment environmental test standard entity in the aviation equipment environmental test standard entity set in the relationship extraction results appears in , get the first aviation equipment environmental test standard entity set is: ; Among them, is the first aviation equipment environmental test standard entity set, is the list of aviation equipment environmental test standard entity types containing all entities in but not in ; is the set difference operation.

[0072] Step S242: Traverse the aviation equipment environmental test standard entity attribute extraction results in step S23, count all aviation equipment environmental test standard subject entities, check whether they appear in the aviation equipment environmental test standard entity extraction results, if not, record and save; extract all aviation equipment environmental test standard subject entities ​​This forms a set of aviation equipment environmental testing standard entities in the attribute extraction results. : ; in, This refers to the set of entities representing aviation equipment environmental testing standards in the attribute extraction results.

[0073] For the list of entities that meet the environmental testing standards for aviation equipment Examine the set of aviation equipment environmental testing standard entities in the attribute extraction results. Does each aviation equipment environmental testing standard entity appear in the list? China; Second set of environmental testing standards for aviation equipment for: ; in, The second set of environmental testing standards for aviation equipment includes all those in... In but not in The physical entity of the aviation equipment environmental testing standards.

[0074] Step S243: Combine the first set of aviation equipment environmental test standard entities obtained in steps S241 and S242. With the second set of environmental testing standards for aviation equipment The corresponding national standard text and the entity recognition results of the aviation equipment environmental testing standard are input into the large language model, combined with preset prompt words; the parts of the results saved in steps S241 and S242 that do not exist in the entity recognition results are filtered out, and the aviation equipment environmental testing standard entity list is updated for these parts of the results. .

[0075] Step S3: Detect aviation equipment environmental testing standard entity pairs with different expressions but the same meaning based on semantic vector similarity; merge those aviation equipment environmental testing standard entities with different textual expressions but the same semantics to ensure the uniqueness and consistency of aviation equipment environmental testing standard entities in the knowledge graph; specifically including the following sub-steps: Step S31: Standardize the list of standard entity types for environmental testing of aviation equipment; for the list of standard entities for environmental testing of aviation equipment obtained in step S21 The first set of standard entities for environmental testing of aviation equipment obtained in step S241 The second set of standard entities for environmental testing of aviation equipment obtained in step S242 Merge and update them to obtain the set of environmental testing standards for aviation equipment. for: ; in, an entity set of the environmental test standards for aviation equipment, containing all entities extracted from the environmental test standard documents; is the environmental test standard entity for aviation equipment, such as “GB / T 2423.22-2012 / IEC 60068-2-14:2009”, n is the number of entities, and there are usually .

[0076] This stage needs to perform standardized cleaning on each environmental test standard entity for aviation equipment , and transform as follows: ; wherein, is the pre-processed environmental test standard entity text for aviation equipment, such as “gbt2423.22-2012 / iec60068-2-14:2009”, and is the cleaning mode, and in the implementation, is set to “()”“《》\s+” for removing unnecessary punctuation in the environmental test standards for aviation equipment; is the part of the string that matches the pattern ; represents removing the white space at the beginning and end of the string , and represents converting the string to lowercase form, and the standardized entity set can be obtained through standardized cleaning. .

[0077] Step S32: Perform semantic vectorization of the environmental test standard entity for aviation equipment, and calculate the semantic similarity of the environmental test standard entity for aviation equipment; for the standardized entity set obtained through cleaning, use the transformer encoder to map the text to the semantic space as follows: ; wherein, is the original semantic vector of the environmental test standard entity for aviation equipment ; is the encoder function of the Sentence-BERT model, and represents a 384-dimensional real number vector space.

[0078] L2-normalize the semantic vector of the environmental test standard entity for aviation equipment, specifically as follows: ; The normalized aviation equipment environmental test standard entity semantic vector is used to form a normalized semantic vector set ; is the L2 norm length of vector ; is the component of the i-th dimension of vector .

[0079] For the normalized aviation equipment environmental test standard entity semantic vector set , an aviation equipment environmental test standard entity semantic similarity matrix is calculated, and the calculation method is as follows: ; wherein, is the entity semantic similarity matrix of ; is an element of the i-th row and j-th column of the entity semantic similarity matrix , indicating the cosine similarity between the aviation equipment environmental test standard entities and ; is the included angle between the semantic vector and ; is the included angle cosine value, which is equivalent to the similarity . Step S33: Aviation equipment environmental test standard entity pre-classification and screening are performed; the entity semantic similarity matrix S and the aviation equipment environmental test standard entity set are obtained, and binary classification is performed using a similarity threshold

[0080] , and the classification decision function is: ; ; ; wherein, is an aviation equipment environmental test standard entity index set that needs to participate in clustering; is an index set of independent and unique aviation equipment environmental test standard entities; is a clustering determination threshold, which is set to 0.85 in the aviation equipment environmental test standard.

[0081] The obtained clustering candidate aviation equipment environmental test standard entity set is , and the independent aviation equipment environmental test standard entity set is .

[0082] ​​The semantic conflict detection of the aerial equipment environmental test standard entity is performed, and the semantic confidence of the aerial equipment environmental test standard entity is calculated , the conflict pair satisfying the condition is detected ; Among them, is a set of potential aerial equipment environmental test standard entity semantic conflicts; is a similarity judgment threshold, which is set to 0.85 in the environmental test standard.

[0083] The confidence of each aerial equipment environmental test standard entity semantic conflict pair is calculated as ; Among them, is the detection confidence of the aerial equipment environmental test standard entity semantic conflict pair .

[0084] The aerial equipment environmental test standard entity semantic conflict detection result is ; Among them, is the total number of aerial equipment environmental test standard entity semantic conflict pairs.

[0085] Step S34: aerial equipment environmental test standard entity semantic clustering analysis; the clustering candidate aerial equipment environmental test standard entity set obtained in step S33 The DBSCAN clustering algorithm is applied, specifically ; Among them, is the clustering label of the th aerial equipment environmental test standard entity, and when the value is-1, it indicates that it is a noise point, that is, an independent entity that does not need to be clustered, and DBSCAN is a density-based clustering algorithm function; is the neighborhood radius parameter in the DBSCAN algorithm, which is used to judge whether two aerial equipment environmental test standard entities are“similar enough”, which is set to 0.3 in the aerial equipment environmental test standard; is the minimum sample number parameter, which is set to 1.

[0086] The aerial equipment environmental test standard entity semantic clustering result obtained is is: ; Among them, is the aerial equipment environmental test standard entity semantic clustering result; The semantic clustering category of the aviation equipment environmental test standard entity.

[0087] The result integration and output; based on the results obtained in the above steps, the calculation of statistical indicators is carried out; the output of this step mainly includes: the semantic conflict pair of the aviation equipment environmental test standard entity ; the semantic clustering result of the aviation equipment environmental test standard entity ; the clustering candidate aviation equipment environmental test standard entity set , the independent aviation equipment environmental test standard entity set ; the total number of semantic conflict pairs of the aviation equipment environmental test standard entity is ; the semantic average similarity of the aviation equipment environmental test standard entity ; the semantic clustering quality parameter of the aviation equipment environmental test standard entity ; wherein, is the internal average similarity of the semantic clustering category of the aviation equipment environmental test standard entity .

[0088] Step S4: aviation equipment environmental test standard entity consistency detection based on graph relationship and attribute similarity; this step aims to utilize the relationship topological structure and attribute characteristics of the aviation equipment environmental test standard entity in the knowledge graph to detect and merge those aviation equipment environmental test standard entities with highly similar relationships in the graph, so that the knowledge graph structure is consistent. The relationship topological structure pays special attention to the association relationship of the test project adopting the test equipment, the test project having the characteristic test parameter, and the test parameter containing the physical quantity attribute; specifically including the following sub-steps: Step S41: extract the aviation equipment environmental test standard entity relationship and attribute set; based on the aviation equipment environmental test standard entity relationship extraction result obtained in step S22, which is in the form of “subject entity || relationship type || object entity”, and the aviation equipment environmental test standard entity attribute extraction result obtained in step S23, which is in the form of “subject entity || attribute type || attribute value”. For any two aviation equipment environmental test standard entities to be compared and , respectively collect the aviation equipment environmental test standard entity relationship set , and the aviation equipment environmental test standard entity attribute set , ; wherein, represents all relationship triples with the aviation equipment environmental test standard entity as the subject entity or object entity. Each aviation equipment environmental test standard entity relationship is represented as: relationship type, associated entity. representing all property triples of the aerial equipment environmental test standard entity ; each property is represented as: property type, property value; , Similarly.

[0089] Step S42: calculating the aerial equipment environmental test standard entity relationship similarity; using Jaccard similarity to calculate the aerial equipment environmental test standard entity relationship similarity: ; wherein, is the aerial equipment environmental test standard entity relationship similarity; is the set intersection, that is, the common relationship pair; is the set union, that is, all unique relationship pairs. If two relationship pairs have the same relationship type and associated entity, they are considered as the same relationship.

[0090] Similarly, the aerial equipment environmental test standard entity attribute similarity is calculated using Jaccard similarity: ; wherein, is the aerial equipment environmental test standard entity attribute similarity; is the set intersection operator; is the set union. If two attribute pairs have the same attribute type and attribute value, they are considered to be the same.

[0091] Combining the aerial equipment environmental test standard entity relationship similarity and attribute similarity, weighted average is performed, specifically: ; wherein, is the overall similarity, the overall similarity ranges from [0, 1], and the value closer to 1 indicates a higher possibility of being the same entity; α is the first weight coefficient; β is the second weight coefficient, satisfying α+β=1. In the aerial equipment environmental test standard, α=0.7 and β=0.3 are set.

[0092] The aerial equipment environmental test standard entity graph relationship similarity matrix is constructed as: ; wherein, is the aerial equipment environmental test standard entity graph relationship similarity matrix of ; is the element of the graph relationship similarity matrix in the row column position, representing the aerial equipment environmental test standard entity and graph relationship similarity.

[0093] Step S43: Perform the aerial equipment environmental test standard entity graph relationship conflict detection; based on the graph relationship similarity matrix, detect potential conflict aerial equipment environmental test standard entity pairs, specifically: ; wherein, is the conflict aerial equipment environmental test standard entity pair; is the graph relationship similarity threshold, which is set to 0.85 in the environmental test standard.

[0094] The aerial equipment environmental test standard entity graph relationship conflict confidence is calculated as: ; wherein, is the aerial equipment environmental test standard entity graph relationship conflict confidence; is the minimum value function.

[0095] The aerial equipment environmental test standard entity graph relationship conflict detection result is obtained as: ; wherein, is the aerial equipment environmental test standard entity graph relationship conflict detection result.

[0096] Step S44: Realize aerial equipment environmental test standard entity clustering according to the similarity threshold; based on the aerial equipment environmental test standard entity graph relationship similarity matrix, directly perform entity clustering using the similarity threshold: Initialize the aerial equipment environmental test standard entity clustering set Clusters=∅, and the unassigned entity set U={ , ,..., }; For each unassigned aerial equipment environmental test standard entity ∈U, create an aerial equipment environmental test standard entity iterative clustering ={ }; For each other unassigned aerial equipment environmental test standard entity ∈U and ≠ : If the aerial equipment environmental test standard entity graph relationship similarity matrix > , add to the aerial equipment environmental test standard entity iterative clustering After completing one round of detection, remove the aerial equipment environmental test standard entity iterative clustering from the unassigned aerial equipment environmental test standard entity set U all entities in the and iteratively cluster the AEEES entities add AEEES entity Clusters; wherein, is the graph relation clustering threshold, set to 0.75; the result of the unassigned AEEES entity graph relation clustering is: ; wherein, is the unassigned AEEES entity graph relation clustering result; is the AEEES entity graph relation clustering class.

[0097] For the clustering result, the result is divided into: independent unassigned AEEES entities and clustered unassigned AEEES entities, specifically: ; ; wherein, is the list of independent unassigned AEEES entity types; is the clustered unassigned AEEES entity.

[0098] Step S5: fuse the AEEES entity semantic vector similarity result obtained in step S3 and the AEEES entity graph relation similarity result obtained in step S4: For each AEEES entity semantic cluster ∈ and AEEES entity graph relation cluster ∈ , the AEEES entity clustering similarity is calculated as: ; wherein, is the AEEES entity clustering similarity; is the semantic similarity clustering result; is the graph relation similarity clustering result.

[0099] For the clustering of the AEEES entity clustering similarity , the two clusters are merged; for the clustering of , it is retained, and the result of this process is recorded as the AEEES entity fusion clustering set ; the final AEEES entity list is determined as: ; wherein, is the final list of aviation equipment environmental test standard entities; is the aviation equipment environmental test standard entity fusion clustering set; is the independent aviation equipment environmental test standard entity set based on semantic similarity.

[0100] Final list of aviation equipment environmental test standard entities , aviation equipment environmental test standard entity relationship extraction result , aviation equipment environmental test standard entity attribute extraction result obtained in step S23 That is, the entity, relationship and attribute extraction results extracted by the present application jointly constitute the core framework of the aviation equipment environmental test standard knowledge graph. Specifically, the entity is the node of the graph, the relationship is the edge, and the attribute is the node feature, forming a complete graph structure data, that is, the aviation equipment environmental test standard knowledge graph.

[0101] The aviation equipment environmental test knowledge graph construction method based on semantic consistency provided by the present application has wide and far-reaching practical application value in the engineering field. Its core value lies in converting massive, unstructured aviation equipment environmental test standard text into structured, computable and inferable knowledge network, thereby directly empowering core industrial links such as research and development, design, production and manufacturing, quality detection and equipment operation and maintenance.

[0102] In the research and development and manufacturing field of high-end complex equipment, such as the development process of aerospace, rail transportation and smart grid equipment, the method can construct a professional knowledge graph covering GB / T, GJB and other series of standards. Specifically, in the research and development of an aero-engine, an aviation equipment environmental test standard knowledge graph integrating many standards such as high-temperature test, vibration test and fatigue test can be constructed, which can intelligently retrieve all test items, qualification criteria and equipment requirements related to the current design parameters, greatly improving the research and development efficiency and standard compliance. For example, in the upgrading of intelligent test and detection equipment, the knowledge graph constructed by the method is embedded into the equipment control system, giving it the ability to "understand" and "execute" the standard. For example, based on the constructed GB / T2423 series standard graph, the environmental test box automatically matches and sets the corresponding test conditions such as temperature, humidity and vibration and the severity level according to the type of the product to be tested, realizes the intelligentization and standardization of the test process, and improves the objectivity of the test result and the analysis efficiency.

[0103] To verify the effectiveness and reliability of the aviation equipment environment test knowledge graph construction method based on semantic consistency proposed in the application, an evaluation experiment is designed. The following section will quantitatively analyze the results of entity extraction, relationship extraction and attribute extraction through indicators such as precision, recall and F1-score, to objectively evaluate the performance of the method in practical application.

[0104] To ensure the reliability and effectiveness of the evaluation results, the application first manually annotates the aviation equipment environment test standard entities and their relationships in the files according to the established entity recognition and relationship extraction rules, constructs a high-quality annotated data set, and uses this annotated data set as the benchmark for evaluation.

[0105] The number of true positives, TP, false negatives, FN, and false positives, FP, output by the method and the comparative method is counted. TP refers to the number of triples that exist in the manually annotated file and are also correctly annotated in the output file, which reflects the effective entity relationship accurately identified by the model; FN represents the number of triples that exist in the manually annotated file but do not exist in the output file, which reflects the missed detection of the model on the actual relationship; FP represents the number of triples that exist in the output file but do not exist in the manually annotated file, i.e. the entity relationship incorrectly identified by the model. For example: there are the following entities in the manually annotated file: "GB / T8170||standard; Preface||chapter"; the output result is: "GB / T8170||standard 3 terms and definitions||chapter". It can be seen that the entity "GB / T8170||standard" appears in both the manually annotated file and the output result file, so it is counted as a TP example, while the entity "Preface||chapter" exists in the manually annotated file but does not exist in the output file, so it is counted as a FN example, and the entity "3 terms and definitions||chapter" exists in the output file but does not exist in the manually annotated file, so it is counted as a FP example.

[0106] The following is a detailed introduction to each evaluation indicator: Precision is used to measure the accuracy of the model in named entity recognition and relationship extraction, i.e. the proportion of correctly predicted entities in all output entities. From a mathematical point of view, the calculation method of precision is the ratio of true positives TP to the sum of true positives TP and false positives FP, and its calculation formula is: ; where, Precision is the ratio of the number of true positives TP to the sum of the number of true positives TP and the number of false positives FP, and the higher the precision, the more accurate and reliable the model is in identifying specific relationships; is the number of triples that exist in the manually annotated file and are also correctly annotated in the output file; is the number of triples that exist in the output file but do not exist in the manually annotated file.

[0107] Recall focuses on evaluating the comprehensiveness of named entity recognition and relation extraction, i.e., the proportion of correct content in the output to all correct content in the standard file; its calculation method is the ratio of the number of true positives TP to the sum of the number of true positives TP and the number of false negatives FN, and the calculation formula is as follows: ; wherein, Recall is the ratio of the number of true positives TP to the sum of the number of true positives TP and the number of false negatives FN, and the higher the recall, the more actual relationships the model can identify, the higher the coverage of actual relationships, and the fewer missed cases; is the number of triples that exist in the manually annotated file but do not exist in the output file.

[0108] F1 score is the harmonic mean of precision and recall, which is a comprehensive index that combines accuracy and recall; it balances the relationship between precision and recall and avoids the limitations of single-index evaluation. The calculation method of F1 score is the product of precision and recall divided by twice the sum of precision and recall, and the calculation formula is as follows: ; wherein, is the product of precision and recall divided by twice the sum of precision and recall, and the higher the F1 score, the better the model performs in both precision and recall dimensions, and the better the overall performance; by introducing F1 score, the overall performance of different methods in entity recognition and relation extraction tasks can be more comprehensively and accurately evaluated.

[0109] In the evaluation experiment, the GraphRAG and the LightRAG used for constructing the knowledge graph are selected as the comparison objects. The GraphRAG is a structure used for enhancing the reasoning capability of a large language model, which constructs a graph structure from text and uses the graph structure for global information retrieval, but usually depends on a preset or general domain entity and relationship schema; and the LightRAG is a light-weight retrieval enhancement generation method, which focuses on obtaining relevant context through vector retrieval and other efficient methods, but has weak capability in deep analysis and structuring of complex semantic relationships. As shown in Table 3, in the specific evaluation link of the relationship extraction task, since the storage method of the GraphRAG is more complex, it is difficult to directly express the relationship extraction result in the form of a triple, therefore, in the relationship extraction aspect, only the method is compared with the LightRAG. Figure 6

[0110] Table 3 Performance evaluation of named entity recognition of three methods Implementation Precision Recall F1 score This method 92.35% 83.2% 87.54% GraphRAG 20.45% 19.07% 19.74% LightRAG 30.49% 28.81% 29.63% As shown in Table 3, in the named entity recognition task, the method has better performance than the GraphRAG and the LightRAG in terms of precision, recall and F1 score. This result fully shows that the method can accurately identify entities in the text in the named entity recognition task, and the missing and misjudgment cases are less, and the comprehensive performance is better than the GraphRAG and the LightRAG.

[0111] Table 4 Performance evaluation of relationship extraction of the method and the LightRAG Implementation Precision Recall F1 score This method 90.51% 82.48% 86.31% LightRAG 25.73% 32.23% 28.61% As shown in Table 4, in the relationship extraction task, the method also has good performance. The method is better than the LightRAG in terms of precision, recall and F1 score. This result shows that in the relationship extraction task, the method can more accurately identify the relationship between the aviation equipment environmental test standard entities, effectively reduces the misjudgment and missing judgment cases, and has better performance than the LightRAG.

[0112] The second aspect of the embodiment of the application provides a knowledge graph construction system of the aviation equipment environmental test knowledge graph construction method based on semantic consistency, which comprises an entity recognition module, a relationship recognition module, an attribute recognition module and a data processing module.

[0113] ​The entity recognition module is responsible for automatically identifying and extracting the list of aviation equipment environmental test standard entity types from the pre-processed aviation equipment environmental test standard text. Specifically, it includes: constructing an aviation equipment environmental test standard entity extraction prompt word template, which contains the task description of the large language model, the list of aviation equipment environmental test standard entity types, the input text, the extraction examples and the output format; using the large language model to extract the aviation equipment environmental test standard entity from the input text, based on the given list of aviation equipment environmental test standard entity types, extracting all aviation equipment environmental test standard entities that meet the conditions, and outputting them in the format of "entity name || entity type"; parsing the model output results, ensuring the uniformity of the output format through regular expression matching, removing irrelevant text, and generating a list of aviation equipment environmental test standard entities, each of which includes the name and type of the aviation equipment environmental test standard entity; realizing the automatic identification of key entities in the aviation equipment environmental test standard text, and providing basic data for subsequent relationship identification and attribute identification.

[0114] The relationship identification module is responsible for extracting the semantic relationship between aviation equipment environmental test standard entities from the aviation equipment environmental test standard text and the identified aviation equipment environmental test standard entities. The specific functions include: constructing a relationship extraction prompt word template, which contains the task description of the large language model, the aviation equipment environmental test standard entity extraction results, the relationship setting list in the relevant field, the input text, the extraction examples and the output format; using the large language model to extract the relationship from the input text and the entity list, the model based on the relationship setting list, judges whether there is a pre-set relationship between the aviation equipment environmental test standard entity pair, and outputs it in the format of "subject entity || relationship type || object entity"; parsing the model output results, ensuring the uniformity of the format, generating a list of aviation equipment environmental test standard entity relationships, each of which includes the subject entity, the relationship type and the object entity; this module establishes the association between the aviation equipment environmental test standard entities, forms the relationship edges in the knowledge graph, and enhances the structured representation of knowledge.

[0115] The attribute recognition module is responsible for extracting the attribute information of the aviation equipment environmental test standard entity from the aviation equipment environmental test standard text and the recognized aviation equipment environmental test standard entity. The specific functions include: constructing an attribute extraction prompt word template, which contains the task description of the large language model, the aviation equipment environmental test standard entity extraction result, the attribute setting list of the relevant field, the input text, the extraction example and the output format; using the large language model to extract the attributes of the input text and the aviation equipment environmental test standard entity list, based on the aviation equipment environmental test standard entity attribute list, extracting the corresponding attribute value for each aviation equipment environmental test standard entity, and outputting it in the format of "entity || attribute type || attribute value"; analyzing the model output result to ensure the format is uniform, generating an attribute list, each attribute including aviation equipment environmental test standard entity, attribute type and attribute value; this module enriches the detailed information of the aviation equipment environmental test standard entity, perfects the feature description of the aviation equipment environmental test standard entity in the knowledge graph, thereby supporting more fine-grained knowledge query and analysis.

[0116] The data processing module is responsible for integrating, cleaning and optimizing the semantic consistency of the aviation equipment environmental test standard entity, relationship and attribute data output by the foregoing modules; the specific functions include: traversing the relationship extraction result and the attribute extraction result, selecting the aviation equipment environmental test standard entity that does not exist in the entity extraction result as a new aviation equipment environmental test standard entity. Based on the semantic vector similarity and the graph relationship and attribute feature similarity, the aviation equipment environmental test standard entity pairs with different expressions but the same meaning are detected and are subjected to semantic clustering; this module performs post-processing on the preliminary extraction result, so that the environmental test aviation equipment environmental test standard knowledge graph has high accuracy and consistency.

[0117] The beneficial effects of the embodiments of the present application are as follows: the present application realizes the construction of the aviation equipment environment test standard knowledge graph through a large language model, and aims at the problem of lack of dynamic adaptation of the aviation equipment environment test standard field knowledge system, dynamically constructs the knowledge system library of entity types, relationships and attribute settings about aviation equipment environment test projects, test equipment and test parameters through a large language model, realizes the targeted guidance of aviation field equipment test, improves the accuracy and generalization ability of knowledge extraction, reduces the cost of manual annotation and the dependence on feature engineering; through collaborative extraction in a unified large language model framework, high-precision information extraction is realized using customized prompt word templates, ensuring the semantic consistency and integrity between entities, relationships and attributes of aviation equipment environment test projects, test equipment and test parameters, reducing data redundancy and omission; aiming at the redundancy and alignment problem caused by synonymous entities, through traversal matching and large language model semantic disambiguation, the unmatching entities of aviation equipment environment test projects, test equipment and test parameters are verified and fused, effectively improving the alignment quality and overall consistency of the knowledge graph, supporting more efficient knowledge reasoning and application.

[0118] The above-described embodiments are only used to describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by ordinary technical personnel in the field shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for constructing an aviation equipment environmental test knowledge graph based on semantic consistency, characterized in that: It comprises: S1: optical character recognition aviation equipment environmental test standard file obtains aviation equipment environmental test standard input text; The preset aviation equipment environmental test standard entity type, entity relationship and entity attribute; The aviation equipment environmental test standard entity includes: test project, test equipment and test parameter; The aviation equipment environmental test standard knowledge system library is constructed; S2: using large language model constructs aviation equipment environmental test standard entity extraction method, extracts aviation equipment environmental test standard entity, entity relationship and entity attribute; According to the entity relationship, entity attribute extraction result, supplement aviation equipment environmental test standard entity extraction result; S3: aviation equipment environmental test standard entity semantic vectorization is carried out, the aviation equipment environmental test standard entity semantic similarity is calculated, and entity classification and screening are carried out; The aviation equipment environmental test standard entity pair is detected by using semantic vector similarity; The aviation equipment environmental test standard entities with the same semantics are merged; S4: according to the graph relationship and attribute similarity, the aviation equipment environmental test standard entity consistency detection is realized; The relationship topological structure and attribute characteristics of the aviation equipment environmental test standard entity in the knowledge graph are used to determine the association relationship of the test equipment used by the test project, the test parameters of the test project and the physical attribute characteristics contained in the test parameters, and the aviation equipment environmental test standard entities with similar relationship are detected and merged, so that the knowledge graph structure is consistent; S5: the aviation equipment environmental test standard entity clustering similarity is calculated; The aviation equipment environmental test standard entity semantic vector similarity result in step S3 and the aviation equipment environmental test standard entity graph relationship similarity result in step S4 are fused; The final aviation equipment environmental test standard entity list is determined, and the aviation equipment environmental test standard knowledge graph is obtained.

2. The method of claim 1, wherein the method is based on semantic consistency of an aviation equipment environment test knowledge graph. Step S3 is specifically: S31: standardizing the aviation equipment environment test standard entity to obtain an aviation equipment environment test standard entity set , , is the first aviation equipment environment test standard entity; S32: Perform the aerial equipment environment test standard entity semantic vectorization, and calculate an aerial equipment environment test standard entity semantic similarity matrix ; S33: Proceed with the pre-classification and screening of the aerial equipment environmental test standard entity; obtain the aerial equipment environmental test standard entity semantic conflict detection result ; S34: The aviation equipment environmental test standard entity semantic clustering analysis is performed, and aviation equipment environmental test standard entity semantic clustering results are obtained , and the semantic vector similarity detection result is output.

3. The method of claim 2, wherein the method comprises: determining a semantic consistency of the test data; and determining a semantic consistency of the test data based on the semantic consistency of the test data. Step S33 is specifically: The entity pre-classification and screening of the aviation equipment environment test standard is performed; the entity semantic similarity matrix S and the aviation equipment environment test standard entity set are obtained , the similarity threshold is used to perform binary classification, and the clustering candidate aviation equipment environment test standard entity set is obtained , and the independent aviation equipment environment test standard entity set is ; The aviation equipment environment test standard entity index set that needs to participate in clustering is obtained. The index set of independent unique aviation equipment environment test standard entities is obtained; the aviation equipment environment test standard entity semantic conflict detection is performed, the aviation equipment environment test standard entity semantic confidence is calculated; the entity semantic similarity matrix S and the aviation equipment environment test standard entity set are obtained , the aviation equipment environment test standard entity semantic conflict set satisfying the condition is detected ; the confidence of each aviation equipment environment test standard entity semantic conflict pair is calculated ; The entity semantic conflict detection result of the aviation equipment environmental test standard is obtained as .

4. The method of claim 2, wherein the method comprises: The output semantic vector similarity detection result in step S34, specifically including: pairs of semantic conflicts of entities of environmental test standards for aerospace equipment , semantic clustering results of entities of environmental test standards for aerospace equipment , set of clustering candidate entities of environmental test standards for aerospace equipment , set of independent entities of environmental test standards for aerospace equipment , total number of pairs of semantic conflicts of entities of environmental test standards for aerospace equipment , average semantic similarity of entities of environmental test standards for aerospace equipment , semantic clustering quality parameter of entities of environmental test standards for aerospace equipment , wherein is the internal average similarity of semantic clustering class of entities of environmental test standards for aerospace equipment .

5. The method of claim 1, wherein the method is characterized by: Step S4 is specifically: S41: extracting the entity relationship according to the aviation equipment environmental test standard entity relationship extraction result and the aviation equipment environmental test standard entity attribute extraction result , extracting the entity relationship and attribute set of the aviation equipment environmental test standard S42: Calculate the similarity of the entity relationship of the aviation equipment environmental test standard and the attribute similarity , and the overall similarity is obtained by weighted average , the entity graph relationship similarity matrix of the aviation equipment environmental test standard is constructed ; S43: Perform the aerial equipment environmental test standard entity graph relationship conflict detection; calculate the aerial equipment environmental test standard entity graph relationship conflict confidence , Get aerial equipment environmental test standard entity graph relationship conflict detection results ; S44: Clustering the independent unassigned aviation equipment environmental test standard entities according to the similarity threshold value With the clustered unassigned aviation equipment environmental test standard entities .

6. The aviation equipment environment test knowledge graph construction method based on semantic consistency according to claim 5, characterized in that: Step S42 is specifically: The Jaccard similarity is used to calculate the aviation equipment environmental test standard entity relationship similarity and attribute similarity: ； ； wherein, is an environmental test standard entity relationship similarity for aerospace equipment; is a set intersection operator; is a set union operator; is an environmental test standard entity attribute similarity for aerospace equipment; Combined with the aviation equipment environmental test standard entity relationship similarity and attribute similarity, weighted average is carried out, specifically: ； wherein, is the overall similarity; a is a first weight coefficient; and β is a second weight coefficient.

7. The method of claim 1, wherein the method is characterized by: Step S5 is specifically: The aviation equipment environmental test standard entity clustering similarity is calculated as: ； wherein, is an environmental test standard entity clustering similarity for aerospace equipment; is a semantic similarity clustering result; is a graph relationship similarity clustering result; is an environmental test standard entity semantic clustering category for aerospace equipment; is an environmental test standard entity graph relationship clustering category for aerospace equipment; According to the aviation equipment environment test standard entity clustering similarity Carrying out clustering extraction, obtaining an aviation equipment environment test standard entity fusion clustering set ; determining a final aviation equipment environment test standard entity list .

8. The method of claim 1, wherein the method is based on semantic consistency of an aviation equipment environment test knowledge graph. Step S2 is specifically: S21: Extracting aviation equipment environmental test standard entities using a large language model, outputting a list of entity types in the aviation equipment environmental test standard containing entity names and entity types Including: test items, test equipment and test parameters; S22: using large language model extracts aviation equipment environmental test standard entity relationship, including relationship type of test project, test equipment and test parameter; S23: using large language model extracts aviation equipment environmental test standard entity attribute, extracts physical quantity attribute related to test condition, including test equipment size, test temperature and test humidity; S24: The list of the environmental test standards entity types of the aviation equipment extracted in step S21 According to the environmental test standards entity relationship and attribute extraction results, the environmental test standards entity extraction results are supplemented.

9. The method of claim 1, wherein the method is based on semantic consistency of an aviation equipment environment test knowledge graph. Step S1 is specifically: S11: Obtain the aviation equipment environmental test standard file in the portable document format (PDF) form, perform optical character recognition, and obtain the original recognized aviation equipment environmental test standard text ; S12: presetting the aviation equipment environmental test standard entity type list; S13: presetting the aviation equipment environmental test standard entity relationship; S14: presetting the aviation equipment environmental test standard entity attribute; S15: The list of the entity types of the aviation equipment environmental test standard obtained according to steps S12-S14 , entity relations and entity attributes , to construct the aviation equipment environmental test standard knowledge base.

10. A knowledge graph construction system for the semantic consistency-based aviation equipment environment test knowledge graph construction method of any one of claims 1 to 9, characterized in that, It comprises: Entity recognition module, relationship recognition module, attribute recognition module and data processing module; The entity recognition module extracts the entity according to the entity extraction prompt word template of the aviation equipment environment test standard, uses a large language model to perform aviation equipment environment test standard entity extraction on the input text, extracts all aviation equipment environment test standard entities that meet the conditions, generates an aviation equipment environment test standard entity list, and realizes automatic identification of key entities in the aviation equipment environment test standard text; The relationship recognition module identifies the association between the aviation equipment environment test standard entities, forms the relationship edges in the knowledge graph, and enhances the structured representation of knowledge. The semantic relationship between the aviation equipment environment test standard entities is extracted from the aviation equipment environment test standard text and the identified aviation equipment environment test standard entities. The attribute recognition module includes detailed information of the aviation equipment environment test standard entity, perfects the feature description of the aviation equipment environment test standard entity in the knowledge graph, thereby supporting more fine-grained knowledge query and analysis, and is responsible for extracting aviation equipment environment test standard entity attribute information from the aviation equipment environment test standard text and the identified aviation equipment environment test standard entities. The data processing module is responsible for integrating, cleaning, and optimizing the semantic consistency of the aviation equipment environment test standard entity, relationship, and attribute data output by the preceding modules.

Images