Method and system for identifying named entities in aerospace field
By building clear entity categories and corpus in the aerospace field and using hybrid neural network models for naming entity recognition, the problem of low accuracy in entity recognition in the aerospace field is solved and efficient key entity extraction is achieved.
Patent Information
- Application Number
- CN202510043742.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-10
- Publication Date
- 2025-05-30
AI Technical Summary
The aerospace field has problems such as blurred entity naming boundaries, diverse entity name abbreviations and aliases, and lack of public corpus, resulting in the existing models performing poorly in terms of accuracy, recall and F1 values.
By dividing entity categories and setting entity boundaries clear rules, the space corpus Space-Corpus is built, and a hybrid neural network model is used, including BERT, Bi-LSTM and CRF layers, is used to identify named entities.
It realizes efficient identification of key entities in the statement text of user demand in the aerospace field, improves the accuracy and recall of entity recognition, and overcomes the problem of unclear entity boundary division in traditional methods.
Smart Images
Figure CN120068870A_ABST
Abstract
Description
Technical Field
[0001] This document relates to the technical field of named entity recognition, and particularly to a method and system for named entity recognition in the aerospace field. Background Art
[0002] Named entity recognition is a hot research direction in natural language processing technology. Its task is to automatically identify the required entities from unstructured text and label them as predefined categories. Named entity recognition in the aerospace field can be considered as automatically identifying specific categories of entities within the aerospace field (such as spacecraft, space launch sites, aerospace agencies, etc.) from various unstructured texts, and it is a key primary task for aerospace text information extraction, user semantic understanding in the aerospace field, aerospace knowledge graph construction, etc.
[0003] Compared with the general field, due to the characteristics of the aerospace field, there are problems in named entity recognition such as fuzzy entity naming boundaries, diverse abbreviations and aliases of entity names, and lack of public corpora, which pose obstacles to the construction of aerospace corpora. In similar fields, combat command texts, joint operation scenario texts, and command post exercise scenario documents are generally used as entity recognition corpora. However, due to the large differences between the language characteristics of user requirements in the aerospace field and professional texts, problems such as colloquial expressions, abbreviations, and non-standard entity names like aliases are relatively rare in the above-mentioned corpora, so they are not suitable as typical representatives of user requirement statements.
[0004] The problem of named entity recognition in the aerospace field for open-source unstructured text is a problem with relatively strong professionalism and immature neural network models. Conventional entity recognition models in similar fields mainly use a single unsupervised training model or a simple hybrid neural network model for processing, and these entity recognition models usually perform mediocrely in terms of accuracy, recall rate, and F1 value. Therefore, an efficient named entity recognition model for the aerospace field is needed to identify key entities in user requirement texts. Summary of the Invention
[0005] One or more embodiments of this specification provide a method for named entity recognition in the aerospace field, including:
[0006] Dividing entity categories and setting clear rules for entity boundaries;
[0007] According to the divided entity categories and the clear rules for entity boundaries, after data cleaning, setting annotation strategies, and implementing annotation, construct the aerospace corpus Space-Corpus;
[0008] Construct a mathematical model of the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0009] Furthermore, the method of classifying entity categories and setting entity boundary clarification rules specifically includes:
[0010] By summarizing entity categories in the aerospace field and combining big data analysis of common user requirement description statements, classify the entity categories often involved in the text of customer requirement statements;
[0011] Set reasonable entity boundary clarification rules to clearly distinguish entities and provide a unified standard for subsequent entity annotation.
[0012] Furthermore, the entity categories specifically include: person names, aerospace agencies, countries, time, space environment, functional uses, and related entities.
[0013] Furthermore, the entity boundary clarification rules specifically include:
[0014] When a country is connected to an aerospace agency, due to the uniqueness of the aerospace agency, label the country and the agency separately;
[0015] When the space environment is connected to a related entity, if the entity has a clear reference, label them separately, otherwise label them together as the space environment;
[0016] When a country is connected to a related entity, if the entity is unique to that country, label them separately, otherwise label them together as a related entity;
[0017] When labeling time and related entities, label them according to the accurate principle;
[0018] When characters such as numbers, letters, and short dashes are connected to related entities, label them together as an entity.
[0019] Furthermore, the specific method of constructing the aerospace corpus Space-Corpus after data cleaning, setting annotation strategies, and implementing annotation according to the divided entity categories and entity boundary clarification rules is as follows:
[0020] Perform data cleaning on the obtained original data text, perform regularization processing on the original data text, store it in the form of a TXT text in UTF-8 encoding format, and perform sentence-level division on the text using punctuation marks "。", "!", "?" as sentence separation marks;
[0021] According to the set entity boundary clarification rules, use the BMEO annotation strategy for entity annotation to obtain the aerospace corpus Space-Corpus.
[0022] Furthermore, the specific method of constructing the mathematical model of the named entity recognition method is as follows:
[0023] Construct the BERT layer of the entity recognition model to obtain the semantic encoding information of the input text, and convert the text information into corresponding word vectors and embed them into the model.
[0024] Construct the Bi-LSTM layer of the entity recognition model, and process the input information through three-layer structures of the forget gate, input gate, and output gate in sequence. The Bi-LSTM layer can convert the input vectors and forms into a token sequence with deep context information.
[0025] Construct the CRF layer of the entity recognition model, which is used to calculate the named entity category with the highest possibility score, obtain the sequence with the highest probability by maximizing the log-likelihood function, realize conditional constraints on the output of the Bi-LSTM layer, and output the sequence with the highest probability to obtain the globally optimal sequence.
[0026] Furthermore, the word vectors include: character vectors, sentence vectors, and position vectors.
[0027] One or more embodiments of this specification provide a named entity recognition system in the aerospace field, including:
[0028] Rule setting module: used to divide entity categories and set clear rules for entity boundaries;
[0029] Corpus construction module: used to construct the aerospace corpus Space-Corpus after data cleaning, setting annotation strategies, and implementing annotations according to the divided entity categories and clear rules for entity boundaries;
[0030] Model recognition module: used to construct a mathematical model of the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0031] One or more embodiments of this specification provide an electronic device, including:
[0032] A processor; and,
[0033] A memory arranged to store computer-executable instructions that, when executed, cause the processor to implement the steps of the above-mentioned named entity recognition method in the aerospace field.
[0034] One or more embodiments of this specification provide a storage medium for storing computer-executable instructions that, when executed, implement the steps of the above-mentioned named entity recognition method in the aerospace field.
[0035] By adopting the embodiments of the present invention, through analyzing and summarizing the named entities in the field, a corpus Space-Corpus with relatively clear entity categories and reasonable distribution is independently constructed by means of data preprocessing, clarifying entity boundaries, adopting annotation strategies and software, etc.; at the same time, by constructing a hybrid neural network named entity recognition model suitable for small-scale corpora, the advantages of multiple neural networks are concentrated, and the goal of extracting key entity information in the semantics of user requirements is realized. Compared with the traditional aerospace field entity recognition method, the entity boundary is more clearly divided, and the entity recognition accuracy is higher, and the key entity names can be extracted more efficiently from the text of the requirements statements of aerospace field users.
[0036] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specifically gives the specific embodiments of the present invention. Brief Description of the Drawings
[0037] In order to more clearly illustrate the technical solutions in one or more embodiments of this specification or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0038] Figure 1 It is a flowchart of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0039] Figure 2 It is a schematic diagram of the construction process of the aerospace corpus Space-Corpus of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0040] Figure 3 It is a partial content display diagram of the aerospace corpus Space-Corpus of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0041] Figure 4 It is an overview diagram of a named entity recognition model of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0042] Figure 5 It is a process diagram of generating word vectors based on BERT of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0043] Figure 6 The BERT model structure diagram of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0044] Figure 7 The LSTM cell structure diagram of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification;
[0045] Figure 8 The composition schematic diagram of a named entity recognition system in the aerospace field provided for one or more embodiments of this specification;
[0046] Figure 9 The structure schematic diagram of an electronic device provided for one or more embodiments of this specification. Detailed implementation manners
[0047] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will clearly and completely describe the technical solutions in one or more embodiments of this specification with reference to the accompanying drawings in one or more embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this document.
[0048] Method embodiments
[0049] According to an embodiment of the present invention, a named entity recognition method in the aerospace field is provided. Figure 1 The flowchart of a named entity recognition method in the aerospace field provided for one or more embodiments of this specification, as Figure 1 shown, the named entity recognition method in the aerospace field according to the embodiment of the present invention specifically includes:
[0050] S1. Divide entity categories and set entity boundary clarification rules.
[0051] By summarizing the entity categories in the aerospace field and combining big data analysis of common demand description statements of users, the entity categories often involved in the text of customer demand statements are divided.
[0052] The specific entity categories include: person names, space agencies, countries / regions, time, space environment, functions and uses, and related entities. To label as many useful entities as possible while avoiding overly complex category settings, the scope of coverage of some categories is explained and expanded as necessary here: Space agencies mainly include launch sites, research and development units of spacecraft, etc.; Countries / regions include the full names and abbreviations of countries and regions; Related entities refer to natural celestial bodies such as stars, planets, and satellites, and also include various artificial spacecraft entities. In addition, the series names (such as the Long March series of launch vehicles), nicknames, and abbreviations of spacecraft should also be included; The space environment includes space orbit environments, such as low Earth orbit, polar circular orbit, etc.; Functions and uses refer to the functions and roles of the above-mentioned space agencies and related entities. For example, the functions of remote sensing satellites include national land resource surveys, crop yield estimates, and disaster prevention and mitigation, etc.
[0053] Set reasonable entity boundary clarification rules to clearly distinguish entities and provide a unified standard for subsequent entity labeling. In this embodiment, five entity boundary clarification rules are set, specifically including:
[0054] When a country / region is connected to a space agency, due to the uniqueness of the space agency, the country / region and the agency are labeled separately; such as "Japan, Kagoshima Space Center", "Russia, Vostochny Cosmodrome".
[0055] When the space environment is connected to a related entity, if the entity has a clear reference, they are labeled separately, otherwise they are labeled together as the space environment; such as "Earth-Moon transfer orbit satellite", "Geostationary orbit, Fengyun-2 satellite".
[0056] When a country / region is connected to a related entity, if the entity is unique to that country, they are labeled separately, otherwise they are labeled together as a related entity; such as "USA, GPS navigation satellite", "Russian launch vehicle".
[0057] When labeling time and related entities, label according to the accurate principle; for example, "May 24th" should not be labeled as "May, 24th", and "AsiaSat 6D communication satellite" should not be labeled as "AsiaSat 6D, communication satellite".
[0058] When characters such as numbers, letters, and short dashes are connected to a related entity, they are labeled together as an entity. Such as "FY-2A meteorological satellite", "GSAT-30 communication satellite", etc.
[0059] S2. According to the divided entity categories and entity boundary clarification rules, after data cleaning, setting the labeling strategy, and implementing the labeling, construct the space corpus Space-Corpus.
[0060] Such as Figure 2As shown, obtain the original data text from channels such as Aihangtian.com, People's Daily Online - Aerospace, National Space Administration official website, China Aerospace Science Popularization Network, China Aerospace Science and Technology Knowledge Base, and NASA Chinese website. Clean the obtained original data text, perform regularization processing on the original data text, delete data such as texts containing advertisements, links, incomplete expressions, and meaningless texts that do not contain aerospace information through regularization means, store it in the form of TXT text using UTF-8 encoding format, and use punctuation marks "。", "!", "?" as clause markers to perform sentence-level division on the text;
[0061] According to the set entity boundary definition rules, adopt the BMEO annotation strategy for entity annotation to obtain the aerospace corpus Space-Corpus.
[0062] BMEO annotation is to perform character-level position annotation on each entity in the corpus, with advantages such as complete annotation structure and clear process. Among them: B annotates the starting character of the entity; M represents the middle part of the entity; E represents the ending character of the entity; O represents non-entity content.
[0063] Through the BMEO annotation strategy, it is possible to perform position annotation on all entities in the text stored in TXT form. The annotated text is presented in a vertical column. After each character is the entity classification situation of that character and the description of the positional relationship type in the entity name, including four forms: non-entity (O) of the character, starting character (B) of the entity name, middle part (M) of the entity name, and ending character (E) of the entity name. The text containing the category and position information of the annotated entity obtained through the above steps constitutes the aerospace corpus Space-Corpus, as Figure 3 shown, and can be used to train the entity recognition model.
[0064] S3. Construct a mathematical model of the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0065] Overview diagram of the named entity recognition model is as Figure 4As shown, first, build the BERT layer of the entity recognition model to obtain the semantic encoding information of the input text, and convert the text information into corresponding word vectors and embed them into the model. The BERT model uses two unsupervised prediction tasks, namely the masked language model MLM and the next sentence prediction NSP, for pre-training. In the MLM task, some characters in the training set are randomly replaced with [mask], and the content of [mask] is predicted through the context of [mask] during training, so as to improve the prediction accuracy. The task of NSP is to judge the relationship between two adjacent sentences, so as to determine the rationality of the sentence order. Through the two tasks of MLM and NSP, the BERT model can fully understand the context relationship of characters in the text and sentence-level features. As Figure 5 shown, the BERT layer maps each word in the training text into a low-dimensional dense vector sum form through the Transformer model it contains. This vector is composed of word vectors, sentence vectors, and position vectors, so as to obtain the vector sum form of each word in the above three dimensions. The Y=(Y 1 ,Y 2 ,...,Y n ) obtained after the BERT layer is processed is calculated by multiple layers of Transformer to get C=(C 1 ,C 2 ,...,C n ) and output to the Bi-LSTM for further learning of context features, and its structure is as Figure 6 shown.
[0066] Build the Bi-LSTM layer of the entity recognition model, and process the input information through three-layer structures of the forget gate, input gate, and output gate in turn. The forget gate is used to judge which parts of the information need to be discarded; the input gate is used to judge which information can be updated to the next unit structure; the output gate is used to judge which part of the information can be output. The Bi-LSTM layer can convert the input vector sum form into a tag sequence with deep context information. As Figure 7 shown, the cell state of the LSTM at time t can be obtained by equations (1)-(5):
[0067] F t =σ(W f ×[h t-1 ,x t +b f ) (1)
[0068] I t =σ(W i ×[h t-1 ,x t +b i ) (2)
[0069] Ot = σ(W o × [h t-1 , x t + b o ) (3)
[0070] C t = F t * C t-1 + I t * tanh(W C × [h t-1 , x t + b C ) (4)
[0071] Where: F, I, and O are the input gate, output gate, and forget gate; W is the weight matrix; b is the bias matrix; h t-1 is the output at time t - 1; x t is the input to the LSTM at time t; C t is the state of the LSTM at time t; σ and tanh are activation functions; O t is the generation matrix.
[0072] Then the output of the LSTM at time t can be expressed as:
[0073] h t = O t * tanh(C t ) (5)
[0074] The structure of the LSTM enables it to obtain information from the previous context, but it cannot utilize information from the subsequent context. However, it is particularly crucial to obtain bidirectional information from both the front and back. Therefore, the author concatenates two sets of LSTMs, the forward and backward ones, in a head-to-tail manner to obtain the Bi-LSTM model to acquire deep information from both the front and back contexts. The Bi-LSTM layer can convert the input vector and form into a token sequence with deep information from both the front and back contexts.
[0075] Construct the CRF layer of the entity recognition model. The CRF layer is a discriminative classifier used to calculate the named entity category with the highest likelihood score, which is used to calculate the named entity category with the highest likelihood score. By maximizing the log-likelihood function, the sequence with the highest probability is obtained, realizing conditional constraints on the output of the Bi-LSTM layer. It not only considers the classification probability of a single word but also combines the category information of adjacent words and the structure of the entire sentence to adjust the output of the Bi-LSTM layer. Finally, the sequence with the highest probability is output to obtain the globally optimal sequence and complete the entity recognition task.
[0076] The CRF algorithm can output the optimal tag sequence in the full text by utilizing the relationship between adjacent annotation results. Its basic algorithm is as follows:
[0077] Assume that for a statement of length n, the score matrix of the output layer is P ∈ R n* k, where k is the number of label types, and the matrix element P ij is the score of the i-th word under the j-th label. Then for the input sentence X = (X 1 , X 2 , …, X n ), the score of the output label sequence Y = (Y 1 , Y 2 , …, Y n ) is:
[0078]
[0079] In the formula, B is the output score matrix of the Bi-LSTM.
[0080] By normalizing all possible sequence paths, the probability distribution of the output sequence Y can be obtained as:
[0081]
[0082] To maximize the above probability, first solve the logarithmic likelihood function of Y* for the correct label:
[0083]
[0084] where represents all sequences that conform to the correct annotation rules. During training, the output sequence Y with the maximum overall probability is obtained by maximizing the logarithmic likelihood function *
[0085] Y * = argmax S(X, Y) (9)
[0086] By conditionally constraining the outputs of all Bi-LSTM layers through the CRF layer, and then selecting the sequence with the highest score from the CRF layer for the final output, the probability of incorrect prediction sequences can be effectively reduced, and the globally optimal sequence can be obtained.
[0087] The beneficial effects of the present invention are as follows:
[0088] By adopting the embodiments of the present invention, through analyzing and summarizing the named entities in the field, a corpus Space-Corpus with relatively clear entity categories and reasonable distribution is independently constructed by means of data preprocessing, clarifying entity boundaries, adopting annotation strategies and software, etc. At the same time, by constructing a hybrid neural network named entity recognition model suitable for small-scale corpora, the advantages of multiple neural networks are concentrated, and the goal of extracting key entity information in the semantics of user requirements is achieved. Compared with the traditional entity recognition method in the aerospace field, the entity boundary is more clearly divided, and the entity recognition accuracy is higher, and the key entity names can be extracted more efficiently from the text of the requirements statements of aerospace field users.
[0089] System embodiment
[0090] According to an embodiment of the present invention, a named entity recognition system in the aerospace field is provided. Figure 8 The composition schematic diagram of a named entity recognition system in the aerospace field provided for one or more embodiments of this specification is as follows Figure 8 As shown, the named entity recognition system in the aerospace field according to the embodiment of the present invention specifically includes:
[0091] Rule setting module 80: used to divide entity categories and set rules for clarifying entity boundaries;
[0092] Corpus construction module 82: used to construct the aerospace corpus Space-Corpus after data cleaning, setting annotation strategies and implementing annotation according to the divided entity categories and rules for clarifying entity boundaries;
[0093] Model recognition module 84: used to construct a mathematical model of the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0094] The embodiment of the present invention is a system embodiment corresponding to the above method embodiment, and the specific operations of each module can be understood with reference to the description of the method embodiment, and will not be elaborated here.
[0095] Device embodiment 1
[0096] The embodiment of the present invention provides an electronic device, as Figure 9 shown, including: a memory 90, a processor 92, and a computer program stored on the memory 90 and executable on the processor 92. When the computer program is executed by the processor 92, the following method steps are implemented:
[0097] S1. Divide entity categories and set rules for clarifying entity boundaries;
[0098] S2. According to the entity category division and entity boundary clarification rules, after data cleaning, setting annotation strategies, and implementing annotation, construct the aerospace corpus Space-Corpus;
[0099] S3. Construct a mathematical model for the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0100] Device Embodiment Two
[0101] An embodiment of the present invention provides a computer-readable storage medium, on which an implementation program for information transmission is stored. When the program is executed by a processor 92, the following method steps are implemented:
[0102] S1. Divide entity categories and set entity boundary clarification rules;
[0103] S2. According to the entity category division and entity boundary clarification rules, after data cleaning, setting annotation strategies, and implementing annotation, construct the aerospace corpus Space-Corpus;
[0104] S3. Construct a mathematical model for the named entity recognition method, train the constructed model based on the aerospace corpus, and obtain the globally optimal sequence through the trained model.
[0105] The computer-readable storage medium in this embodiment includes, but is not limited to: ROM, RAM, magnetic disk, optical disc, etc.
[0106] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for named entity recognition in the aerospace field, characterized in that: include: Classify entities and set clear rules for entity boundaries; According to the entity categories and entity boundary rules, after data cleaning, setting annotation strategies and implementing annotation, the aerospace corpus Space-Corpus was constructed; A mathematical model of a named entity recognition method is constructed, the constructed model is trained based on the aerospace corpus, and a global optimal sequence is obtained through the trained model.
2. The method according to claim 1, characterized in that The classification of entities and setting clear rules for entity boundaries specifically include: By summarizing the entity categories in the aerospace field and combining big data analysis of users' commonly used demand description sentences, the entity categories frequently involved in customer demand statement texts are divided; Set reasonable entity boundaries and clear rules to clearly distinguish entities and provide a unified standard for subsequent entity labeling.
3. The method according to claim 2, characterized in that The entity categories specifically include: personnel name, space agency, country, time, space environment, functional purpose and related entities.
4. The method according to claim 2, characterized in that: The entity boundary clearing rules specifically include: When a country is linked to a space agency, the country and agency are marked separately due to the uniqueness of the space agency; When the spatial environment is connected to a related entity, if the entity has a clear reference, it is marked separately, otherwise it is marked together as the spatial environment; When a country is linked to a related entity, if the entity is specific to that country, it is marked separately, otherwise it is marked together as a related entity; When marking time and related entities, follow the principle of accuracy; When characters such as numbers, letters, and hyphens are connected to related entities, they are marked together as entities.
5. The method according to claim 1, characterized in that According to the entity categories and entity boundary clarification rules, after data cleaning, setting annotation strategies and implementing annotation, the specific method of constructing the Space-Corpus is as follows: The obtained raw data text is cleaned, the raw data text is regularized, and stored in the form of TXT text in UTF-8 encoding format, and the punctuation marks ".", "!" and "?" are used as sentence markers to divide the text into sentence levels; According to the set entity boundary clarification rules, the BMEO annotation strategy is used to perform entity annotation and obtain the aerospace corpus Space-Corpus.
6. The method according to claim 1, characterized in that The specific method of constructing the mathematical model of the named entity recognition method is: Build the BERT layer of the entity recognition model, obtain the semantic encoding information of the input text, convert the text information into corresponding word vectors and embed them into the model. The Bi-LSTM layer of the entity recognition model is constructed, and the input information is processed in sequence through a three-layer structure of a forget gate, an input gate, and an output gate. The Bi-LSTM layer can convert the input vector and form into a tag sequence with deep context information. The CRF layer of the entity recognition model is constructed to calculate the named entity category with the highest probability score. The sequence with the highest probability is obtained by maximizing the log-likelihood function, and the output of the Bi-LSTM layer is conditionally constrained. The sequence with the highest probability is output to obtain the global optimal sequence.
7. The method according to claim 1, characterized in that The word vector includes: character vector, sentence vector and position vector.
8. A named entity recognition system in the aerospace field, characterized in that: include: Rule setting module: used to classify entities and set clear rules for entity boundaries; Corpus construction module: used to define rules according to the divided entity categories and entity boundaries, and to build the Space-Corpus after data cleaning, setting annotation strategies and implementing annotations; Model recognition module: used to construct a mathematical model of a named entity recognition method, train the constructed model based on the aerospace corpus, and obtain a global optimal sequence through the trained model.
9. An electronic device, characterized in that: include: processor; as well as, A memory arranged to store computer executable instructions, wherein when the computer executable instructions are executed, the processor implements the steps of the method for named entity recognition in the aerospace field as claimed in any one of claims 1 to 7.
10. A storage medium, characterized in that: Used to store computer executable instructions, which, when executed, implement the steps of the aerospace field named entity recognition method as described in any one of claims 1 to 7.