Methods, systems, and storage media for constructing knowledge graphs in the civil aviation service sector

Abstract

This invention discloses a method, system, and storage medium for constructing a knowledge graph in the civil aviation service field. The method includes: S1, using a BERT-BiLSTM-CRF algorithm model to extract entities and obtain interrelated entity vector sequences, feature vector sequences, and annotation sequences; S2, using a convolutional neural network model to extract sentence vectors and their contained entity vectors, and employing n filters to identify and extract a database of entity-relationship-entity triples; S3, using a conditional random field entity node integration model to integrate the annotation information and store it in the corresponding entities as entity attribute values; S4, using the triple database and the integrated entity attribute values ​​to link and fuse them to construct a civil aviation knowledge graph. This invention can obtain a comprehensive and accurate civil aviation knowledge graph of entity relationships based on a civil aviation knowledge text database. It can not only satisfy passenger knowledge question answering and querying needs but also serve as a training and educational resource, improving the overall service level.

Description

Technical Field

[0001] This invention relates to the field of civil aviation data processing and application, and in particular to a method, system and storage medium for constructing a knowledge graph in the field of civil aviation services. Background Technology

[0002] As a vital component of global economic and social development, the civil aviation industry has a profound and far-reaching impact on modern society. With the continuous growth of global air traffic, the complexity and diversity of civil aviation services are also constantly increasing. To ensure efficient and reliable flight operations, timely and accurate acquisition and processing of a large volume of NOTAMs, regulations, and standards has become a crucial task for airlines and civil aviation authorities. However, the current civil aviation service sector faces numerous challenges, particularly significant deficiencies in information management and knowledge acquisition.

[0003] The civil aviation industry's pursuit of "sincere service" and the development of smart civil aviation necessitate further improvements in service quality and increased passenger satisfaction. With the rise of intelligent question-and-answer robots and AI big data, various aspects of civil aviation, such as intelligent service robots in airport waiting halls, AI-powered question-and-answer and information retrieval on airline websites, displays in front of aircraft seats for passenger learning, and others, are crucial for enhancing service. Constructing a complete knowledge graph, or obtaining a comprehensive one, is a fundamental requirement for intelligent question-and-answer robots and AI big data.

[0004] In the civil aviation sector, there is a massive amount of data, which presents significant challenges for civil aviation staff in terms of knowledge acquisition and information retrieval. Ordinary passengers, on the other hand, know very little about it, creating an information gap. For example, passengers cannot quickly obtain the necessary civil aviation knowledge through simple searches, hindering the scientific and accurate decision-making process for their travel. Furthermore, airlines cannot better manage and utilize various NOTAMs and standard information to improve operational efficiency and service quality. This severely restricts the improvement of civil aviation services and hinders the enhancement of their intelligence and informatization levels.

[0005] Constructing a knowledge graph for querying and answering questions about civil aviation knowledge, and relying on this knowledge graph to form important links for massive amounts of data content, facilitates more comprehensive and efficient access to the data content corresponding to the entity nodes of interest. In addition to satisfying passengers' knowledge-based question-and-answer and query needs (for ordinary passengers, due to a lack of professional background, they often feel confused and at a loss when faced with complex aviation service information and professional terminology), it also provides important data support for civil aviation staff in knowledge learning and retrieval. This is also a technical problem that major airlines and related institutions urgently need to solve. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problems pointed out in the background art, and to provide a method, system and storage medium for constructing a knowledge graph in the field of civil aviation services. The invention extracts entity vectors and labels through the BERT-BiLSTM-CRF algorithm model, then uses a convolutional neural network model with an attention mechanism to identify the correspondence between entities in sentence vectors and construct entity-relation-entity triples. The labeled information is integrated and stored as the attribute values ​​of entities through a conditional random field entity node integration model, and then linked and fused to obtain a civil aviation knowledge graph.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] A method for constructing a knowledge graph in the civil aviation service domain, the method comprising:

[0009] S1. Construct the BERT-BiLSTM-CRF algorithm model and the civil aviation knowledge text database. The BERT-BiLSTM-CRF algorithm model performs civil aviation-related entity extraction processing on the civil aviation knowledge text database and obtains interrelated entity vector sequences, feature vector sequences, and annotation sequences.

[0010] S2. Construct a convolutional neural network model based on an attention mechanism. The convolutional neural network model extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database and uses n filters to identify the local feature matrix C = (c1, c2, ..., c...) of the entity correspondence in the extracted sentence vectors. n This constructs a triplet database of entity-relationship-entity;

[0011] S3. Construct a conditional random field entity node integration model based on semi-supervised learning and / or weakly supervised learning, with the following formula: Where x represents the labeled sequence with step S1 as input, y represents the integrated labeled sequence, Z(x) represents the normalization factor, and f k λ represents the feature function that extracts features from the input labeled sequence. k Describing the characteristic function f k The weights, y j-1 This represents the labeled data at position j-1, y j This represents the annotation data at position j, where j represents the position of the input annotation sequence; the annotation information in the integrated annotation sequence is stored in the entity and used as the attribute value of the entity;

[0012] S4. Using the entity-relationship-entity and attribute values ​​of the integrated entity in the triplet database from step S2, link and merge them to construct the civil aviation knowledge graph.

[0013] To better achieve the present invention, the present invention also includes the following methods:

[0014] A. Calculate the similarity of entities in different sentence vectors using the cosine similarity method:

[0015] Where similarity(e, c) represents the similarity judgment result between e and c, e represents the selected entity, and c represents the entity used for similarity comparison, i.e., the candidate entity. The vector representing entity e, A vector representing entity C;

[0016] B. Perform attribute value and relationship fusion processing on entity C with similarity judgment greater than the similarity threshold or / and the highest similarity judgment with entity e.

[0017] Preferably, the sentence vectors and the entity vectors contained in the civil aviation knowledge text database are extracted using sliding windows, and the entity vectors contained in the sentence vectors are extracted using n sliding windows corresponding to the filters.

[0018] Preferably, the local feature matrix also includes the corresponding weight matrix W∈R w*d Where R represents the real number system, w represents the sliding window size, and d represents the dimension after embedding and connecting; the sentence vector is denoted as the sequence (q1, q2, ..., q... i , ..., q m ), q i The data at position i in the sentence is used to perform a convolution operation to obtain a sequence, as shown in the following formula.

[0019] c i =Wq i-w+1：i +b, Wq i-w+1：i q i-w+1 to q i The weights, b represents the bias vector, and c i This represents the local features at position i in the sentence.

[0020] The local feature matrix of entity correspondence in the sentence vector is extracted with entities in the sentence vector as the core.

[0021] Preferably, the model update expression for the semi-supervised learning method in the conditional random field entity node fusion model is as follows:

[0022] P(y | x;θ)=Model(x;θ) , where θ represents the model parameters;

[0023] θ new =argmaxlogP(Y|X;θ1), θ new θ1 represents the model parameters after the iteration update, θ1 represents the model parameters before the iteration update, X represents the original labeled data set, and Y represents the integrated labeled data set.

[0024] Preferably, the entity vector sequence of the BERT-BiLSTM-CRF algorithm model is arranged in sequence according to the entity vectors, and the feature vectors in the feature vector sequence are the features corresponding to the entity vectors extracted from the text data and arranged in sequence according to the feature vectors. The annotation sequence includes semantic dependency information between entities.

[0025] Preferably, in step S1, the BERT-BiLSTM-CRF algorithm model includes a BERT masked language model, a BiLSTM network, and a Conditional Random Field (CRF) module. The BERT masked language model uses a masked language method to generate deep bidirectional language representations of civil aviation knowledge text data in the civil aviation knowledge text database. The BiLSTM network uses forward and backward methods to capture bidirectional long-distance semantic dependencies. The CRF module extracts entity nodes to obtain interrelated entity vector sequences and uses a set of feature functions to obtain interrelated feature vector sequences and label sequences.

[0026] Preferably, the present invention further includes the following method:

[0027] S5. Continuously update and collect civil aviation knowledge data and construct a civil aviation knowledge database. Users input keywords or entity words corresponding to entities. Following steps S1 to S4, obtain the civil aviation knowledge graph of each entity in the civil aviation knowledge database. Use the civil aviation knowledge graph to perform retrieval analysis on the civil aviation knowledge database based on keywords or entity words corresponding to entities, and obtain the retrieval results corresponding to the entities.

[0028] A knowledge graph construction system for the civil aviation service domain includes a BERT-BiLSTM-CRF algorithm model, a convolutional neural network model, a conditional random field entity node integration model, and a civil aviation knowledge graph model structure. The BERT-BiLSTM-CRF algorithm model internally contains a civil aviation knowledge text database. It performs high-relevance entity extraction processing on the civil aviation knowledge text database, obtaining interrelated entity vector sequences, feature vector sequences, and annotation sequences. The convolutional neural network model is built based on an attention mechanism. It extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database and uses n filters to identify the local feature matrix C = (c1, c2, ..., c...) of the entity correspondences in the extracted sentence vectors. n A triplet database of entity-relation-entity is constructed. The conditional random field entity node integration model is constructed based on semi-supervised learning and / or weakly supervised learning methods. The conditional random field entity node integration model integrates the labeled information in the labeled sequence, as shown in the following formula: Where x represents the labeled sequence with step S1 as input, y represents the integrated labeled sequence, Z(x) represents the normalization factor, and f k λ represents the feature function that extracts features from the input labeled sequence. k Describing the characteristic function f k The weights, y j-1 This represents the labeled data at position j-1, y j This represents the labeled data at position j, where j represents the position of the input labeled sequence; the conditional random field entity node integration model stores the labeled information in the integrated labeled sequence into the entity and uses it as the attribute value of the entity; the civil aviation knowledge graph model structure uses the entity-relationship-entity of the triple database and the attribute values ​​of the integrated entity to link and fuse them to obtain the civil aviation knowledge graph.

[0029] A storage medium storing a computer program, which, when executed by a processor, implements the steps of the knowledge graph construction method in the civil aviation service field of the present invention.

[0030] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0031] (1) This invention extracts entity vectors and labels through the BERT-BiLSTM-CRF algorithm model, then uses a convolutional neural network model with attention mechanism to identify the entity correspondence in the sentence vector and construct entity-relationship-entity triples. The label information is integrated and stored as the attribute value of the entity through the conditional random field entity node integration model, and then linked and fused to obtain the civil aviation knowledge graph.

[0032] (2) This invention can obtain a comprehensive and accurate civil aviation knowledge graph of relationships between entities based on the civil aviation knowledge text database. This facilitates the subsequent use of the civil aviation knowledge graph to provide valuable reference information for civil aviation managers and decision-makers through data analysis, assisting them in making scientific decisions. It can also use the structured data of the civil aviation knowledge graph to conduct in-depth data mining, discover new knowledge and patterns hidden behind the data, thereby promoting the innovative development of the industry.

[0033] (3) The civil aviation knowledge graph of this invention covers multi-level and multi-dimensional information on civil aviation service terminology and general conditions of transport for various airlines, enabling efficient integration and sharing of information and providing technical support for the high-quality development of civil aviation services. This civil aviation knowledge graph not only satisfies passengers' knowledge-based question-and-answer and query needs, but also provides important data support for civil aviation staff's knowledge learning and retrieval. It can also serve as a training and education resource, providing systematic learning materials for practitioners and improving overall service levels. Furthermore, this invention not only enhances the efficiency of information utilization but also promotes the development of various industries, laying a solid foundation for continuous innovation. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the method for constructing a knowledge graph in the civil aviation service field according to the present invention.

[0035] Figure 2 This is a schematic diagram showing the partial list of entities in step S1 of the embodiment;

[0036] Figure 3 This is a schematic diagram illustrating the extraction of the annotation sequence corresponding to a portion of the listed entities in step S1 of the embodiment.

[0037] Figure 4 This is a schematic diagram illustrating the relationships between excerpts of the civil aviation knowledge graph in the example.

[0038] Figure 5 This is a schematic diagram of the principle structure of the knowledge graph construction system in the civil aviation service field of the present invention.

[0039] Figure 6 This is a usage scenario diagram of the civil aviation knowledge graph for civil aviation data retrieval in the embodiment. Detailed Implementation

[0040] The present invention will be further described in detail below with reference to embodiments:

[0041] Example

[0042] like Figure 1 As shown, a method for constructing a knowledge graph in the civil aviation service domain includes the following steps:

[0043] S1. Construct a BERT-BiLSTM-CRF algorithm model and a civil aviation knowledge text database. The civil aviation knowledge text data in the database comes from civil aviation laws, regulations, notices, management regulations, etc. (including text data obtained from websites or web pages). It covers multi-level and multi-dimensional information such as civil aviation service terminology and general conditions of transport for various airlines. Entity nodes are extracted using this invention. The BERT-BiLSTM-CRF algorithm model performs entity extraction processing on the civil aviation knowledge text database with high civil aviation relevance (entities are associated with the civil aviation field) and obtains interrelated entity vector sequences, feature vector sequences, and annotation sequences. The entity vector sequences, feature vector sequences, and annotation sequences are all arranged in sequence with entity nodes as key association nodes (that is, the entity vector sequences, feature vector sequences, and annotation sequences all include entity nodes and are arranged in order to form a sequence set). In this embodiment, a total of nearly 20,000 entity nodes (referred to as entities, some examples of entities are shown below) are extracted. Figure 2As shown in the figure, the relationship between entities provides basic data and relational label information to support the subsequent construction of a civil aviation knowledge graph in the civil aviation field. The entity vector sequence of the BERT-BiLSTM-CRF algorithm model is arranged in sequence according to the entity vectors. The feature vector sequence consists of features extracted from the text data corresponding to the entity vectors and arranged in sequence according to the feature vectors. The label sequence includes semantic dependency information between entities and is also formed by sorting the entity nodes to form a sequence set.

[0044] In some embodiments, the BERT-BiLSTM-CRF algorithm model includes a BERT masked language model, a BiLSTM network, and a Conditional Random Field (CRF) module. The BERT masked language model uses a masked language method to generate deep bidirectional language representations of civil aviation knowledge text data in the civil aviation knowledge text database (mining word vectors with deep bidirectional contextual relationships as the basis for further filtering of entity vectors). The BiLSTM network uses forward and backward methods to capture bidirectional long-distance semantic dependencies of information (the BiLSTM network has a stronger ability to extract contextual features and can capture bidirectional long-distance semantic dependencies of information). The CRF module extracts entity nodes (selecting entity nodes from word vectors) to obtain a sequence of interrelated entity vectors, and uses a set of feature functions (including several feature functions) to obtain a sequence of interrelated feature vectors and a labeled sequence.

[0045] S2. Construct a convolutional neural network model based on an attention mechanism. The convolutional neural network model extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database and uses n filters to identify the local feature matrix C = (c1, c2, ..., c...) of the entity correspondence in the extracted sentence vectors. nThis constructs a triplet database of entity-relation-entity. Preferably, sentence vectors and their contained entity vectors in the civil aviation knowledge text database are extracted using sliding windows. The entity vectors contained in the sentence vectors are extracted using n sliding windows corresponding to filters, thus sequentially extracting entity vectors from the sentence vectors using sliding windows. Preferably, the subject-verb-object entities are also emphasized in the sentence vectors. The convolutional neural network model of this invention incorporates an attention mechanism module to further extract relations from entity vectors in sentence vectors. The entity vector sequence strategy in sentence vectors is built on the attention mechanism, which can obtain a pure, noise-free combined vector representation. This invention constructs a civil aviation knowledge graph model structure to assist in the embedding, editing, and optimization of civil aviation knowledge graphs. First, each sentence is processed by a convolutional neural network model and transformed into a feature vector representation (with entity vectors at its core). The feature vector corresponding to the sentence with the highest support is identified, and these vectors are gradually accumulated. Preferably, a specific set is constructed, containing all combinations of feature vectors from sentences involving the same entity pair. Then, the ability score of each set of feature vectors mapped to relation labels in a multi-class classification task is calculated, thereby identifying the feature vector combination with the highest score. The set of feature vectors with the highest score integrates the semantic information of positive instances and the features of potentially noisy sentences, providing a solid foundation for training a classifier for target entity pairs, thus constructing a comprehensive and suitable model framework for relation extraction tasks. During entity relation extraction, the lengths of sentences vary, which allows civil aviation service domain texts to reveal the relationships between entity nodes at any position within a sentence (see...). Figure 3 As shown, some entity correspondences and annotation sequences are listed; therefore, this invention can adopt the strategy of extracting diverse local features from sentences to infer the corresponding relationship types of civil aviation service entities, and introduce convolution operations to process these extracted local features.

[0046] In some embodiments, the local feature matrix C = (c1, c2, ..., c n The data also includes the corresponding weight matrix W∈R. w*d (During convolution operations, a randomly initialized weight can be defined first), where R represents the real number system, w represents the sliding window size, and d represents the dimension after embedding and connection; the sentence vector is denoted as the sequence (q1, q2, ... q i , ..., q m This sequence is a sequence centered around entity vectors, q i The data at position i in the sentence is used to perform a convolution operation to obtain the sequence:

[0047] c i =Wq i-w+1：i +b, Wq i-w+1：iq i-w+1 to q i The weights, b represent the bias vector (the formula in this invention includes the bias vector b, which helps the model better fit the training data; of course, the formula may not include the bias vector b), c i This represents the local feature at position i in the sentence (thus obtaining all local features in the sentence vector; multiple dissimilar filters are used to extract local features, and the convolutional layer ultimately outputs a set of matrices (c1, c2, ..., c...). n This invention can further identify all positive instances that embody label relationships and extract local feature matrices of entity correspondences in sentence vectors, with entities in the sentence vectors as the core. Because this invention employs an attention mechanism, when sentence length is insufficient to meet convolution window requirements or when sentences are too long and may lead to information loss, the attention mechanism module can identify and focus on key parts of the sentence, thereby better capturing label relationships when constructing feature vectors without strictly relying on the sliding window size.

[0048] In some embodiments, after obtaining the entity vector in each sentence vector, the present invention further includes the following method:

[0049] A. Calculate the similarity of entities in different sentence vectors using the cosine similarity method. (After information processing, the data is stored in a data layer and a schema layer. Although this achieves entity extraction from unstructured and semi-structured civil aviation service data, there is still a lot of redundant and erroneous information. Data aggregation processing is needed to eliminate redundancy and errors, thereby improving the accuracy of entity-relationship-entity. The entity linking process involves selecting a group of entities to be determined from the civil aviation service entity pool, calculating similarity based on attribute values, and linking them to the correct entity objects.)

[0050] Where similarity(e, c) represents the similarity judgment result between e and c, e represents the selected entity, and c represents the entity used for similarity comparison, i.e., the candidate entity. The vector representing entity e, A vector representing entity c;

[0051] B. Perform attribute value and relationship fusion processing on entity c with the highest similarity score and / or entity e (if the relationship between the two entities is represented differently in multiple data sources, fusion can be performed by calculating the consistency of the relationship).

[0052] The further method is as follows: First, determine a selected entity e. Based on the annotation information in the annotation sequence of the BERT-BiLSTM-CRF algorithm model, select a set of entities for similarity comparison (hereinafter referred to as the set of entities to be compared). In the set of entities to be compared, use the cosine similarity method to calculate the similarity of each entity C. Select each entity C with a similarity judgment greater than the similarity threshold and fuse its attribute values ​​and relations with entity e (for multiple entity words with the same meaning); or select the entity c with the highest similarity judgment and fuse its attribute values ​​and relations with entity e (for two entity words with the same meaning).

[0053] S3. Construct a conditional random field entity node integration model based on semi-supervised learning and / or weakly supervised learning, with the following formula: Where x represents the labeled sequence with step S1 as input, y represents the integrated labeled sequence, Z(x) represents the normalization factor, and f k This represents the feature function used to extract features from the input labeled sequence (in step S1, feature extraction is performed using a feature function set containing multiple feature functions, f). k (where λ is the k-th eigenfunction in the set of eigenfunctions) k Describing the characteristic function f k The weights, y j-1 This represents the labeled data at position j-1, y j This represents the annotation data at position j, where j represents the position of the input annotation sequence; the annotation information in the integrated annotation sequence is stored in the entity and used as the attribute value of the entity.

[0054] In some embodiments, the model update expression for the semi-supervised learning method in the conditional random field entity node ensemble model is as follows:

[0055] P(y|x;θ)=Nodel(x;θ), where θ represents the model parameters. The above method is used to identify the attributes corresponding to entity nodes. Labeling refers to assigning a known target value or label to samples in entity nodes. In supervised learning, the model learns the relationship between input data (features) and their corresponding output labels, enabling it to predict whether new data matches entity nodes. Part-of-speech tagging is then performed to help identify the descriptive attributes of entity nouns. A semi-supervised learning model is used, training the model with partially labeled data and a large amount of unlabeled data. A common method in this step is self-training. The basic idea is to first train a base model using labeled data, then use this model to predict unlabeled data. A second training iteration can be performed using a portion of accurately predicted unlabeled data to update the semi-supervised learning model and make it more adaptable to civil aviation data text. Data with high prediction confidence is selected from the unlabeled training results and continuously updated iteratively.

[0056] In self-trained semi-supervised learning methods, model updates can be simplified to the following form.

[0057] θ new =argmaxlogP(Y|X;θ1), θ new θ1 represents the model parameters after iterative updates, θ2 represents the model parameters before iterative updates, X represents the original labeled data set, and Y represents the integrated labeled data set. Semi-supervised learning is suitable for training on data with high labeling accuracy. In the process of automated labeling, it can be used for data text annotation in the knowledge graph of the civil aviation field, and can then be integrated with the Conditional Random Field (CRF) module. In this stage, the CRF module can be used again to score the obtained entity attributes. The final score is obtained through weighted summation, thus identifying attributes with higher relevance to the corresponding entities and discarding attributes with lower scores.

[0058] S4. Using the entity-relationship-entity pairs from the triplet database in step S2, and the attribute values ​​of the integrated entities, a civil aviation knowledge graph is constructed by linking and fusing them (see...). Figure 3 (This section extracts and lists a portion of the entity relationship annotations for display.) See also: Figure 4 This section extracts a portion of the civil aviation knowledge graph, which contains nearly 20,000 entities and multiple links between them (as well as annotation information, etc.). This facilitates knowledge sharing among personnel in the civil aviation field and other fields, as well as the efficient integration and retrieval of civil aviation data (e.g.,...). Figure 6As shown in the illustration, this invention's civil aviation knowledge graph is used for rapid retrieval applications in civil aviation databases. It facilitates users' quick understanding of civil aviation entity nodes and their connections. The civil aviation knowledge graph not only provides civil aviation managers and passengers with references and information summaries of civil aviation knowledge relationships to support scientific decision-making, but also uncovers hidden knowledge and patterns, driving industry innovation. Furthermore, it can serve as a training and educational resource, providing systematic learning materials for practitioners and improving overall service levels. This invention not only enhances information utilization efficiency but also promotes the development of various industries, laying a solid foundation for continuous innovation.

[0059] This invention constructs entity-relation-entity triples through multi-level integration and entity linking of entities, and further builds a knowledge graph using the Neo4j graph database. Through the Cypher query language, we successfully imported nearly 20,000 entity nodes and their relationships. Compared with traditional relational databases, Neo4j is more suitable for representing complex entity (node) and relation (edge) structures. Cypher, as Neo4j's query language, is designed specifically for interacting with graph data and can better express queries of graph structures.

[0060] In some embodiments, the present invention further includes the following method:

[0061] S5. Continuously update and collect civil aviation knowledge data and construct a civil aviation knowledge database. Users input keywords or entity terms corresponding to entities. Following steps S1 to S4, a civil aviation knowledge graph for each entity in the database is obtained. The civil aviation knowledge graph is then used to perform keyword or entity term retrieval analysis on the database, yielding the corresponding retrieval results. See [link to relevant documentation]. Figure 6 Taking the entity term "passenger ticket" as an example, the search results are truncated as follows: Figure 6 As shown.

[0062] like Figure 5 As shown, a knowledge graph construction system for the civil aviation service domain includes a BERT-BiLSTM-CRF algorithm model, a convolutional neural network model, a conditional random field entity node integration model, and a civil aviation knowledge graph model structure. The BERT-BiLSTM-CRF algorithm model internally contains a civil aviation knowledge text database. It performs high-relevance entity extraction processing on the civil aviation knowledge text database to obtain interrelated entity vector sequences, feature vector sequences, and annotation sequences. The convolutional neural network model is built based on an attention mechanism. It extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database and uses n filters to identify the local feature matrix C = (c1, c2, ..., c...) of the entity correspondences in the extracted sentence vectors. nA triplet database of entity-relation-entity is constructed. The conditional random field entity node integration model is constructed based on semi-supervised learning and / or weakly supervised learning methods. The conditional random field entity node integration model integrates the labeled information in the labeled sequence, as shown in the following formula: Where x represents the labeled sequence with step S1 as input, y represents the integrated labeled sequence, Z(x) represents the normalization factor, and f k λ represents the feature function that extracts features from the input labeled sequence. k Describing the characteristic function f k The weights, y j-1 This represents the labeled data at position j-1, y j This represents the labeled data at position j, where j represents the position of the input labeled sequence; the conditional random field entity node integration model stores the labeled information in the integrated labeled sequence into the entity and uses it as the attribute value of the entity; the civil aviation knowledge graph model structure uses the entity-relationship-entity of the triple database and the attribute values ​​of the integrated entity to link and fuse them to obtain the civil aviation knowledge graph.

[0063] A storage medium storing a computer program, which, when executed by a processor, implements the steps of the knowledge graph construction method in the civil aviation service field of the present invention.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a knowledge graph in the civil aviation service field, characterized in that: The methods include: S1. Construct the BERT-BiLSTM-CRF algorithm model and the civil aviation knowledge text database. The BERT-BiLSTM-CRF algorithm model performs civil aviation-related entity extraction processing on the civil aviation knowledge text database and obtains interrelated entity vector sequences, feature vector sequences, and annotation sequences. The entity vector sequence of the BERT-BiLSTM-CRF algorithm model is arranged in sequence according to the entity vectors. In the feature vector sequence, the feature vectors are the features corresponding to the entity vectors extracted from the text data and arranged in sequence according to the feature vectors. The annotation sequence includes semantic dependency information between entities. S2. Construct a convolutional neural network model based on an attention mechanism. The convolutional neural network model extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database, and uses n filters to identify the local feature matrix of entity correspondences in the extracted sentence vectors. A database of entity-relation-entity triples is constructed; the entity correspondences extracted from sentence vectors are identified using n filters, including the following methods: A. Calculate the similarity of entities in different sentence vectors using the cosine similarity method: ,in express and The similarity judgment result, Indicates the selected entity. This refers to the entities used for similarity comparison, i.e., candidate entities. Representing entities The vector, Representing entities ; B. Entities with similarity scores greater than the similarity threshold or / and the highest similarity scores. With entity Perform attribute value and relationship fusion processing; if the relationship between two entities is represented differently in multiple data sources, fusion is performed by calculating the consistency of the relationship; first, determine a selected entity. Based on the annotation information in the labeled sequence of the BERT-BiLSTM-CRF algorithm model, a set of entities for similarity comparison is selected. Then, the cosine similarity method is used sequentially to compare the similarity of each entity in the set to be compared. For similarity calculation, given multiple entity words with the same meaning, select those entities with a similarity score greater than a similarity threshold for comparison. associate its attribute values ​​and relationships with the entity Perform fusion processing; or for two entity words with the same meaning, select the entity with the highest similarity score. With entity Perform attribute value and relationship fusion processing; S3. Construct a conditional random field entity node integration model based on semi-supervised learning and / or weakly supervised learning, with the following formula: ,in This represents the labeled sequence with step S1 as input. This represents the integrated labeled sequence. Represents the normalization factor. This represents the feature function that extracts features from the input labeled sequence. Characteristic function The weight, Indicates the first Data labeled at each location, Indicates the first Data labeled at each location, Indicates the position of the input annotation sequence; stores the annotation information in the integrated annotation sequence in the entity and uses it as the attribute value of the entity; S4. Using the entity-relationship-entity and attribute values ​​of the integrated entity in the triplet database from step S2, link and merge them to construct the civil aviation knowledge graph. S5. Continuously update and collect civil aviation knowledge data and construct a civil aviation knowledge database. Users input keywords or entity words corresponding to entities. Following steps S1 to S4, obtain the civil aviation knowledge graph of each entity in the civil aviation knowledge database. Use the civil aviation knowledge graph to perform retrieval analysis on the civil aviation knowledge database based on keywords or entity words corresponding to entities, and obtain the retrieval results corresponding to the entities.

2. The method for constructing a knowledge graph in the civil aviation service field according to claim 1, characterized in that: Sentence vectors and their contained entity vectors are extracted from the civil aviation knowledge text database using sliding windows. The entity vectors contained in the sentence vectors are extracted using n sliding windows corresponding to the filters.

3. The method for constructing a knowledge graph in the civil aviation service field according to claim 1 or 2, characterized in that: The local feature matrix also includes the corresponding weight matrix. Where R represents the real number system, w represents the sliding window size, and d represents the dimension after embedding and connecting; the sentence vector is denoted as a sequence. , The data at position i in the sentence is used to perform a convolution operation to obtain a sequence, as shown in the following formula. , express arrive The weight, This represents the bias vector. This represents the local features at position i in the sentence. The local feature matrix of entity correspondence in the sentence vector is extracted with entities in the sentence vector as the core.

4. The method for constructing a knowledge graph in the civil aviation service field according to claim 1, characterized in that: The model update expression for the semi-supervised learning method in the entity node ensemble model of Conditional Random Fields is as follows: ,in Indicates model parameters; , This represents the model parameters after iterative updates. This represents the model parameters before the iterative update. Represents the original labeled data set. This represents the integrated set of labeled data.

5. The method for constructing a knowledge graph in the civil aviation service field according to claim 1, characterized in that: In step S1, the BERT-BiLSTM-CRF algorithm model includes a BERT masked language model, a BiLSTM network, and a Conditional Random Field (CRF) module. The BERT masked language model uses a masked language method to generate deep bidirectional language representations of civil aviation knowledge text data in the civil aviation knowledge text database. The BiLSTM network uses forward and backward methods to capture bidirectional long-distance semantic dependencies. The CRF module extracts entity nodes to obtain interrelated entity vector sequences and uses a set of feature functions to obtain interrelated feature vector sequences and label sequences.

6. A knowledge graph construction system for the civil aviation service domain, implementing the knowledge graph construction method for the civil aviation service domain as described in claim 1, characterized in that: The model comprises the BERT-BiLSTM-CRF algorithm, a convolutional neural network model, a conditional random field entity node integration model, and a civil aviation knowledge graph model. The BERT-BiLSTM-CRF algorithm internally contains a civil aviation knowledge text database. It extracts entities from this database with high civil aviation relevance, obtaining interrelated entity vector sequences, feature vector sequences, and annotation sequences. The convolutional neural network model, built on an attention mechanism, extracts sentence vectors and their contained entity vectors from the civil aviation knowledge text database and uses n filters to identify the local feature matrix of entity correspondences within the extracted sentence vectors. A triplet database of entity-relation-entity is constructed. The conditional random field entity node integration model is constructed based on semi-supervised learning and / or weakly supervised learning methods. The conditional random field entity node integration model integrates the labeled information in the labeled sequence, as shown in the following formula: ,in This represents the labeled sequence with step S1 as input. This represents the integrated labeled sequence. Represents the normalization factor. This represents the feature function that extracts features from the input labeled sequence. Characteristic function The weight, Indicates the first Data labeled at each location, Indicates the first Data labeled at each location, The position of the input annotation sequence is indicated; the conditional random field entity node integration model stores the annotation information in the integrated annotation sequence in the corresponding entity and uses it as the attribute value of the entity; the civil aviation knowledge graph model structure uses the entity-relationship-entity of the triple database and the attribute value of the integrated entity to link and fuse them to obtain the civil aviation knowledge graph.

7. A storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 5.

Images