Internet aerospace intelligence automatic extraction and structuring method and related device

Abstract

The application discloses an internet aerospace intelligence automatic extraction and structuring method and related equipment, relates to the technical field of aerospace technology, artificial intelligence, natural language processing and knowledge graph cross technology, and the method comprises the steps of: using a pre-established aerospace professional dictionary to perform normalization processing on aerospace field terms in processed intelligence text, using a large model fine-tuned by aerospace field data to obtain extracted information from the standardized text, obtaining extracted information associated with at least one space target unique identifier, then generating entity-relation-entity triplets and entity-attribute-value triplets, performing time sequencing, normalization and deduplication processing on all triplets, storing the processed triplets into a graph database, and forming a dynamically updated aerospace knowledge graph; providing a standardized query interface based on the aerospace knowledge graph, and outputting a structured knowledge base. The application provides effective data support for multi-satellite task planning, situation awareness and other engineering systems.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of aerospace technology, artificial intelligence, natural language processing and knowledge graphs, and in particular to an automatic extraction and structuring method and related equipment for Internet aerospace intelligence. Background Technology

[0002] Currently, global aerospace information sources are scattered, the data volume is huge, and the text is highly unstructured. Relying on manual processing is inefficient, has poor timeliness, and lacks standardization, which cannot meet the data needs of engineering systems such as multi-satellite mission planning and situational awareness.

[0003] In existing technologies, to address the aforementioned problems, general-purpose web crawlers are typically used to scrape aerospace information from the internet, and general-purpose natural language processing models are employed to extract information from the text. These methods attempt to reduce manual costs and increase data acquisition speed through automation.

[0004] However, existing technologies still have the following shortcomings: First, general-purpose web crawlers are not optimized for the aerospace field, resulting in a large amount of noise information unrelated to aerospace, which is difficult to clean and cannot normalize time formats or technical terms. Second, general-purpose natural language processing models have low accuracy in recognizing specialized entities such as satellites, orbits, payloads, launch sites, and mission numbers, easily leading to ambiguity and failing to accurately extract complex relationships, attributes, and dynamic events between entities. Third, the extracted results are fragmented, failing to form structured entity-relationship-attribute knowledge, making it difficult to support direct use in engineering systems. Fourth, the extracted spacecraft names, codes, and numbers cannot be automatically associated with the unique identifiers of space targets within the system, making it difficult to integrate intelligence data with engineering modules such as orbit calculation, collision warning, and mission planning. Fifth, the lack of a unified aerospace knowledge fusion and updating mechanism prevents the construction of a dynamic, reasonable, and searchable knowledge base, hindering the continuous evolution of intelligence data and its ability to provide timely support for upper-level applications. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies, specifically providing a method and related equipment for automatic extraction and structuring of internet-based aerospace intelligence, as detailed below: 1) In a first aspect, the present invention provides a method for automatic extraction and structuring of Internet aerospace intelligence, the specific technical solution of which is as follows: Multi-source aerospace intelligence is collected and processed to obtain processed intelligence text. Using a pre-established aerospace professional dictionary, aerospace terminology in the processed intelligence text is normalized to obtain standardized text; Information is extracted from standardized text using a large model finely tuned with data from the aerospace field. The spacecraft name, code, and number in the extracted information are matched with a pre-stored set of unique space target identifiers. Based on the matching results, entity disambiguation and association are performed to obtain the extracted information associated with at least one unique space target identifier. Based on the extracted information that has at least one unique identifier for a space target, entity-relationship-entity triples and entity-attribute-value triples are generated. All triples are processed by temporalization, normalization and deduplication. The processed triples are stored in a graph database to form a dynamically updated aerospace knowledge graph. Based on the aerospace knowledge graph, a standardized query interface is provided. The standardized query interface supports searching in the aerospace knowledge graph according to the unique identifier of the space target, mission number, time and region, and outputs a structured knowledge base.

[0006] The beneficial effects of the automatic extraction and structuring method for Internet aerospace intelligence provided by this invention are as follows: By selectively collecting and processing multi-source aerospace intelligence, processed intelligence texts can be obtained, effectively overcoming the problems of high information noise and difficulty in cleaning caused by general web crawlers. A pre-established aerospace professional dictionary is used to normalize aerospace terminology in the processed intelligence text, obtaining standardized text that eliminates ambiguity caused by term aliases, abbreviations, and foreign language translations, improving the accuracy of subsequent extraction. A large model fine-tuned with aerospace data is used to perform named entity recognition, relation extraction, attribute extraction, and event extraction from the standardized text, obtaining extracted information. This solves the shortcomings of general natural language processing models in terms of low accuracy and ambiguity in professional entity recognition, achieving high-precision intelligent extraction of entities, relations, attributes, and events. The spacecraft names, codes, and numbers in the extracted information are matched with a pre-stored set of unique space target identifiers. Based on the matching results, entity disambiguation and association are performed to obtain extracted information associated with at least one unique space target identifier. This achieves automatic association between the extracted results and the unique space target identifiers, allowing direct integration with engineering systems such as orbit calculation, collision warning, and mission planning. Based on the extracted information, entity-relation-entity triples and entity-attribute-value triples are generated. After time-sequencing, normalization, and deduplication, these are stored in a graph database to form a dynamically updated aerospace knowledge graph. This overcomes the shortcomings of fragmented extracted information and the lack of a unified knowledge fusion and updating mechanism. Based on this aerospace knowledge graph, a standardized query interface is provided, supporting retrieval by unique space target identifier, mission number, time, and region, and outputting a structured knowledge base. This provides high-timeliness, high-quality, and searchable data support for upper-level systems such as multi-satellite mission planning and situational awareness.

[0007] Based on the above scheme, the method for automatic extraction and structuring of Internet aerospace intelligence of the present invention can be further improved as follows.

[0008] Furthermore, multi-source aerospace intelligence is collected and processed in a targeted manner to obtain processed intelligence text, including: targeted crawling of multi-source aerospace intelligence, and performing webpage denoising, text extraction, data deduplication, and time normalization on the crawled results, and using the processed results as processed intelligence text.

[0009] The beneficial effects of adopting the above-mentioned further solution are as follows: By selectively crawling multi-source aerospace intelligence and performing webpage denoising, text extraction, data deduplication, and time normalization on the crawled results, the processed intelligence text can effectively reduce noise information irrelevant to the aerospace field, accurately extract the core text content from the webpage, avoid duplicate storage of the same intelligence, and unify various date and time formats from different sources into a standard format. This processing flow significantly improves the cleanliness, integrity, and consistency of the original intelligence, reduces the difficulty of subsequent text cleaning, and provides a high-quality data foundation for normalization processing using a pre-established aerospace professional dictionary and intelligent extraction using a large model fine-tuned from aerospace data, thereby improving the efficiency and reliability of the entire automated intelligence processing flow.

[0010] Furthermore, by utilizing a pre-established aerospace professional dictionary, the aerospace terminology in the processed intelligence text is normalized to obtain a standardized text. This includes: using a pre-established aerospace professional dictionary to normalize satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text, unifying the aliases, abbreviations, and foreign language translations of various aerospace terms, and using the normalized text as the standardized text.

[0011] The beneficial effects of adopting the above-mentioned further approach are as follows: By utilizing a pre-established aerospace professional dictionary, the satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text are normalized. This unifies the aliases, abbreviations, and foreign language translations of various aerospace terms. The normalized text serves as a standardized text, effectively eliminating ambiguity caused by multiple spellings of the same aerospace entity in different intelligence sources. It converts satellite designations, orbit types, payload names, launch site names, and agency abbreviations into unified predefined standard names. This process significantly improves the consistency and standardization of aerospace terminology in the text, reducing the risk of semantic confusion during subsequent named entity recognition, relation extraction, attribute extraction, and event extraction using large models fine-tuned with aerospace data. This improves the accuracy and recall rate of intelligent extraction, laying a solid foundation for generating high-quality structured extracted information.

[0012] Furthermore, by using a large model fine-tuned with data from the aerospace field, information is extracted from standardized text. This includes performing named entity recognition, relation extraction, attribute extraction, and event extraction on the standardized text using the large model fine-tuned with data from the aerospace field to obtain extracted information. The extracted information includes the identified entities, extracted relations, extracted attributes, and extracted events.

[0013] The beneficial effects of adopting the above-mentioned further approach are as follows: By utilizing a large model fine-tuned with aerospace data to perform named entity recognition, relation extraction, attribute extraction, and event extraction on standardized text, the resulting information includes identified entities, extracted relations, extracted attributes, and extracted events. This allows for accurate identification of aerospace-specific entities such as satellites, orbits, payloads, launch sites, and mission numbers; precise extraction of relationships between entities such as payload, launch, and affiliation; and the identification of attributes such as orbital altitude, inclination, and mass. Furthermore, it enables the identification of dynamic events such as launch, orbit insertion, and orbital changes. This processing overcomes the shortcomings of general natural language processing models in the aerospace field, such as low recognition accuracy and susceptibility to ambiguity. It transforms unstructured standardized text into structured extracted information containing rich semantic relationships, providing high-quality data input for subsequent space target unique identifier mapping, entity disambiguation and association, and knowledge graph construction.

[0014] Furthermore, the spacecraft name, code, and number in the extracted information are matched with a pre-stored set of unique space target identifiers. Based on the matching results, entity disambiguation and association are performed to obtain the extracted information associated with at least one unique space target identifier. This includes: performing similarity matching between the spacecraft name, code, and number in the extracted information and each entry in the pre-stored set of unique space target identifiers; determining candidate unique space target identifiers corresponding to the spacecraft name, code, and number based on the matching results; when multiple candidate identifiers exist, using the context information in the extracted information to perform entity disambiguation to select at least one unique space target identifier; and associating the selected at least one unique space target identifier with the extracted information to obtain the extracted information associated with at least one unique space target identifier.

[0015] The beneficial effects of adopting the above-mentioned further scheme are as follows: By performing similarity matching between the spacecraft name, code, and number in the extracted information and a pre-stored set of unique space target identifiers, candidate unique space target identifiers are determined. When multiple candidate identifiers exist, entity disambiguation is performed using contextual information in the extracted information to select the unique identifier. Finally, the selected unique space target identifier is associated with the extracted information to obtain extracted information associated with at least one unique space target identifier. This automatically establishes an accurate correspondence between spacecraft in the intelligence text and unique identifiers within the system. This process effectively solves the defect that extracted information cannot be automatically associated with unique space target identifiers, eliminates name ambiguity, and ensures that entities, relationships, attributes, and events in the intelligence data are all accompanied by clear unique space target identifiers. This allows for direct integration with engineering application systems such as orbit calculation, collision warning, and mission planning, thus establishing a data link between intelligence data and engineering systems.

[0016] Furthermore, based on the aerospace knowledge graph, a standardized query interface is provided to output a structured knowledge base. This includes: receiving query requests input according to at least one condition such as unique identifier of space target, mission number, time or region through the standardized query interface; retrieving triples that match the query request in the aerospace knowledge graph; outputting the retrieved triples in a structured data format; and using the output data as a structured knowledge base.

[0017] The beneficial effects of adopting the above-mentioned further solution are as follows: By receiving query requests input according to at least one condition—unique identifier of a space target, mission number, time, or region—through a standardized query interface, the system retrieves triples matching the query request from the aerospace knowledge graph and outputs the retrieved triples in a structured data format as a structured knowledge base. This provides a flexible and efficient data access method for upper-level application systems. Users can quickly locate the required aerospace intelligence based on single or combined conditions. For example, they can query all entities, relationships, and attributes associated with a specific space target's unique identifier, retrieve launch events by mission number, filter the latest intelligence by time interval, or search for launch sites and related missions by region. The structured data format output allows the search results to be directly parsed and loaded by engineering systems such as multi-satellite mission planning, situational awareness, and collision warning, significantly reducing data integration costs and improving the response speed and ease of use of intelligence services.

[0018] Furthermore, entity-relationship-entity triples and entity-attribute-value triples are generated based on the extracted information associated with at least one unique identifier of a space target. This includes: extracting each space entity and the relationships between each space entity from the extracted information associated with at least one unique identifier of a space target; combining each relationship and its two associated space entities into an entity-relationship-entity triple; and extracting each space entity and its corresponding attribute value from the extracted information associated with at least one unique identifier of a space target; combining each attribute and its corresponding space entity and attribute value into an entity-attribute-value triple.

[0019] The beneficial effects of adopting the above-mentioned further scheme are as follows: By extracting various aerospace entities and their relationships from the extracted information associated with at least one unique space target identifier, and combining each relationship and its two associated aerospace entities into an entity-relationship-entity triple, while simultaneously extracting each aerospace entity and its corresponding attribute value, and combining each attribute, its corresponding aerospace entity, and the attribute value into an entity-attribute-value triple, the originally fragmented extracted information can be transformed into a standardized triplet knowledge representation structure. The entity-relationship-entity triple clearly expresses the semantic connections between aerospace entities, and the entity-attribute-value triple accurately describes the inherent characteristics of each aerospace entity. This transformation process provides a unified data format for subsequent temporalization, normalization, and deduplication processing, and also lays a structured foundation for storing the triples in a graph database to construct a dynamically updated aerospace knowledge graph, ensuring the consistency and computability of knowledge representation.

[0020] Furthermore, the processed triples are stored in a graph database to form a dynamically updated aerospace knowledge graph. This includes: constructing nodes and edges in the graph database according to the subject and predicate of the processed triples; assigning a unique identifier to each space target; using the relation in each entity-relation-entity triple as a directed edge connecting two nodes; storing the attribute in each entity-attribute-value triple as attribute key-value pairs for nodes; establishing a time index in the graph database to support time-series queries; and incrementally updating existing nodes and edges each time a new triple is stored to form a dynamically updated aerospace knowledge graph.

[0021] The beneficial effects of adopting the above-mentioned further scheme are as follows: By constructing nodes and edges in the graph database according to the subject and predicate of the processed triples, each space target uniquely identifies a node, the relation in each entity-relation-entity triple is used as a directed edge connecting two nodes, and the attribute in each entity-attribute-value triple is stored as the attribute key-value pair of the node. This allows for the intuitive representation of complex relationships and attribute characteristics between aerospace entities in the graph database. The establishment of a time index supports efficient time-series queries based on the collection time label, meeting the retrieval requirements for timely intelligence. Incremental updates to existing nodes and edges are performed each time a new triple is stored, avoiding the performance overhead of full reconstruction and enabling the aerospace knowledge graph to continuously and in real-time integrate the latest intelligence data. This processing forms a dynamically updated aerospace knowledge graph, ensuring the timeliness, consistency, and scalability of knowledge, and providing fast and flexible knowledge retrieval capabilities for the upper-level standardized query interface.

[0022] Furthermore, all triples are subjected to temporalization, normalization, and deduplication, including: adding a collection time tag to each triple to achieve temporalization; converting attribute values ​​with the same attribute type but different units or different expressions in the triples after adding time tags into a unified dimension and unified format according to a preset standard to achieve normalization; and deleting duplicate triples in the normalized triples where the subject, predicate, object, and time tag are all the same, keeping only one copy to achieve deduplication.

[0023] The beneficial effects of adopting the above-mentioned further solutions are as follows: By adding a collection time tag to each triple to achieve temporalization, each piece of knowledge has a clear time dimension, supporting the tracing of intelligence evolution in chronological order. After adding the time tag, attribute values ​​with the same attribute type but different units or expressions in the triples are converted to a unified dimension and format according to a preset standard, eliminating differences in units and numerical expressions between data from different sources and ensuring that the same attribute value is comparable on the same benchmark. Duplicate triples with the same subject, predicate, object, and time tag are deleted from the normalized triples, leaving only one copy, effectively eliminating redundant data and avoiding storage waste and query ambiguity. This series of processes improves the data quality, consistency, and compactness of the triples, providing clean, standardized, and non-repeating input data for the subsequent construction of a graph database and the formation of a dynamically updated aerospace knowledge graph.

[0024] Furthermore, after forming a dynamically updated aerospace knowledge graph, the process also includes: repeatedly collecting and processing multi-source aerospace intelligence according to a preset time cycle; normalizing the intelligence using a pre-established aerospace professional dictionary; extracting information using a large model finely tuned with aerospace data; matching, disambiguating, and associating spacecraft names, codes, and numbers from the extracted information; generating triples and performing temporal normalization and deduplication on all triples; and storing the processed triples into a graph database. The newly stored triples are then merged with the already stored triples to achieve incremental updates to the aerospace knowledge graph.

[0025] The beneficial effects of adopting the above-mentioned further scheme are as follows: By repeatedly executing the complete process from targeted collection of multi-source aerospace intelligence to storing the processed triples into the graph database according to a preset time cycle, and merging the newly stored triples with the already stored triples, incremental updates to the aerospace knowledge graph are achieved. This process ensures that the knowledge graph can periodically and automatically acquire the latest aerospace intelligence published on the Internet, without the need for manual triggering or full reconstruction. Incremental updates only process newly generated triples, significantly reducing the computational resources and time costs of each update. When newly stored triples are merged with already stored triples, historical versions are preserved and the latest attributes and relationships are added, ensuring the continuity, consistency, and timeliness of the knowledge graph. The resulting dynamic update mechanism ensures that the aerospace knowledge graph always keeps pace with rapidly changing aerospace information, providing continuous and reliable knowledge services for upper-level application systems.

[0026] 2) Secondly, the present invention also provides an automatic extraction and structuring system for Internet aerospace intelligence, the specific technical solution of which is as follows: It includes a data acquisition and processing module, a normalization processing module, an extraction module, a matching and association module, a knowledge graph acquisition module, and an output module; The data acquisition and processing module is used to: collect and process multi-source aerospace intelligence in a targeted manner to obtain processed intelligence text; The normalization module is used to: normalize aerospace terminology in the processed intelligence text using a pre-established aerospace professional dictionary to obtain standardized text; The extraction module is used to extract information from standardized text using a large model finely tuned with data from the aerospace field. The matching and association module is used to: match the spacecraft name, code and number in the extracted information with the pre-stored set of unique space target identifiers, and perform entity disambiguation and association based on the matching results to obtain the extracted information associated with at least one unique space target identifier; The knowledge graph acquisition module is used to: generate entity-relationship-entity triples and entity-attribute-value triples based on the extracted information that is associated with at least one unique identifier of a space target; perform temporalization, normalization and deduplication on all triples; and store the processed triples in the graph database to form a dynamically updated aerospace knowledge graph. The output module is used to provide a standardized query interface based on the aerospace knowledge graph. The standardized query interface supports searching in the aerospace knowledge graph according to the unique identifier of the space target, mission number, time and region, and outputs a structured knowledge base.

[0027] 3) In a third aspect, the present invention also provides an electronic device, the electronic device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor, so as to enable the electronic device to implement any of the above-mentioned methods for automatic extraction and structuring of Internet aerospace intelligence.

[0028] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-mentioned methods for automatic extraction and structuring of Internet aerospace intelligence.

[0029] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below: Figure 1 This is a flowchart illustrating an automatic extraction and structuring method for Internet aerospace intelligence according to an embodiment of the present invention; Figure 2 This is a schematic diagram of an automatic extraction and structuring system for Internet aerospace intelligence according to an embodiment of the present invention. Detailed Implementation

[0031] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0032] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0033] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for automatic extraction and structuring of Internet aerospace intelligence, comprising the following steps: S1. Targeted collection and processing of multi-source aerospace intelligence to obtain processed intelligence text. Specifically, targeted crawling of multi-source aerospace intelligence is performed, and the crawled results are processed by webpage denoising, text extraction, data deduplication, and time normalization. The processed results are then used as the processed intelligence text. The specific implementation process is as follows: S10. Pre-configure a list of Uniform Resource Locators (URLs) for aerospace intelligence sources. This list contains the specific URLs of multiple target websites, such as launch forecast pages of aerospace news portals, data release pages of international space organizations, and news announcement pages of commercial space companies. After starting the crawling program, it sequentially accesses each URL in the list, sends a Hypertext Transfer Protocol (HTTP) request, and receives the raw Hypertext Markup Language (HTML) code of the webpage returned by the server. For each acquired webpage code, its source URL and collection timestamp are recorded, and the raw code is temporarily stored in a memory buffer as input data for subsequent processing.

[0034] S11. Processing multi-source space intelligence, specifically including: ① Perform webpage denoising: For each webpage's original Hypertext Markup Language (HTML) code, parse it into a Document Object Model (DOM) tree structure. Traverse each node of the DOM tree and identify noise nodes based on a pre-defined noise node feature rule library. Noise node features include node attributes containing keywords such as "advertisement," "navigation," "comment," and "share," or nodes located at the page edge with text density below a threshold. Remove all identified noise nodes from the DOM tree, retaining the main text container nodes and their child nodes. The DOM tree after noise removal is converted back to an HTML string, outputting the denoised webpage structure. ② Perform main text extraction: Read the denoised webpage structure and calculate the text density of each text block node in the DOM tree. Text density equals the number of Chinese characters within the node divided by the total number of characters within the node. Filter out consecutive text block nodes with a text density greater than a pre-defined threshold and a paragraph length exceeding a certain number of characters. Extract all text content from the filtered nodes and concatenate them in their original order into a continuous plain text string. This plain text string is the main text content of an intelligence report. Simultaneously, extract title information from webpage metadata or tags near the body text, and merge the title and body text content into a complete intelligence text record. ③ Perform data deduplication: Maintain a hash value index table of stored intelligence texts. For each output intelligence text record, calculate the hash value of the complete content of the record. Compare the calculated hash value with all existing hash values ​​in the index table. If the same hash value exists in the index table, the current intelligence text record is determined to be duplicate data, and the record is discarded without further processing. If the same hash value does not exist in the index table, the current intelligence text record is determined to be new data, the hash value of the record is added to the index table, and the intelligence text record is passed to the next step. ④ Perform time normalization processing: Receive non-duplicate intelligence text records, scan each character in the record, and use pre-compiled multiple time format regular expression patterns to match the date and time strings in the text. For each matched time string, call the date parsing function to convert it into an internally unified International Organization for Standardization (ISO) format string, in the format "year-year-year-month-day", and add a time zone offset based on the collection timestamp or the time zone identifier extracted from the webpage. Replace all occurrences of the time string in the original intelligence text with the converted, uniformly formatted string. After all replacements are complete, output the final plain text data and name it "Processed Intelligence Text".

[0035] S2. Using a pre-established aerospace professional dictionary, the aerospace terminology in the processed intelligence text is normalized to obtain standardized text. Specifically, using the pre-established aerospace professional dictionary, satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text are normalized to unify the aliases, abbreviations, and foreign language translations of various aerospace terms. The normalized text is then used as the standardized text. The specific implementation process is as follows: S20. Read the pre-built aerospace professional dictionary file from local storage or a database. This dictionary is organized in a structured manner, divided into five sub-tables according to terminology categories, corresponding to satellite designation mapping tables, orbital type mapping tables, payload mapping tables, launch site mapping tables, and agency abbreviation mapping tables, respectively. Each row in each sub-table contains a standard name field and a set of variant fields. The variant fields store all aliases, abbreviations, and foreign language translations corresponding to the standard name. Load the entire dictionary into memory and construct five hash mapping tables. The key of each mapping table is a variant string, and the value is the corresponding standard name string. In other words, the aerospace professional dictionary is a pre-built terminology knowledge base for the aerospace field. The dictionary is stored in the form of key-value pairs or mapping tables. Each entry contains a standard terminology name and its various variant forms. Variant forms include common aliases of the term, abbreviations in different languages, foreign language translations, and historically used old numbers. The aerospace professional dictionary covers categories such as satellite designations, orbital type names, payload names, launch site names, and abbreviations of aerospace agencies. The data in the dictionary comes from aerospace standard documents, publicly available launch records, satellite databases, and manually compiled terminology lookup tables.

[0036] S21. The processed intelligence text output from step S1 is used as the input data for this step. The complete content of the text is read and stored in a string variable. Simultaneously, the source identifier and collection timestamp of the text are recorded to maintain its association with the original intelligence after normalization processing. Then, the following operations are performed: ① Normalize satellite designations: After scanning the processed intelligence text string, perform pattern matching according to the common writing patterns of satellite designations. Satellite designations are usually composed of letters, numbers, and hyphens, such as "A-123" or "XYZ-01". For each matched candidate string, check if the string exists as a key in the satellite designation mapping table. If it exists, replace the string with the corresponding standard satellite designation in the mapping table. If it does not exist, leave the original string unchanged. Continue scanning the unprocessed parts of the text until all satellite designation candidates in the entire text have been checked and replaced.

[0037] ② Normalize the orbit type: Based on the text obtained after normalizing the satellite designation, continue scanning for keywords related to orbit types. Orbit types are usually represented by terms such as "Low Earth Orbit," "Geostationary Orbit," and "Polar Orbit," or their abbreviations such as "Low Orbit" and "Geostationary Orbit." Match the identified orbit type candidate strings with the orbit type mapping table. For each successfully matched string, replace it with the predefined standard orbit type name in the orbit type mapping table. For candidate strings that do not match, leave them as is.

[0038] ③ Payload Normalization: Read the text obtained after normalizing the orbit type and scan for candidate payload names. Payload names may appear as abbreviations or full names, such as the English abbreviation and full Chinese name of a remote sensor. Perform exact matching and substring matching on the payload mapping table. For exact matching, directly replace; for substring matching, if a continuous substring in the text is found to be completely identical to a variant in the payload mapping table, replace that substring with the standard payload name. After replacement, update the text content and continue scanning subsequent positions until the end of the text. A payload refers to a device or instrument mounted on a satellite or rocket to perform a specific mission. Payload names may include abbreviations, full names, different manufacturers' naming conventions, and translations in different languages. For example, the English abbreviation and the full Chinese name of a remote sensor may point to the same payload entity. Payload normalization requires associating multiple name variants with a unified standard payload name.

[0039] ④ Launch site normalization: Receive the text obtained after normalizing the payload and scan for candidate launch site names. Launch site names may include geographical location information, modifiers in the official name, and translations in different languages. Compare the candidate strings with the launch site mapping table. During comparison, prioritize matching longer strings to avoid false matches; for example, match "Baikonur Cosmodrome" first, then match "Baikonur". For successfully matched strings, replace them with the standardized launch site names found in the launch site mapping table. After the replacement operation is complete, save the updated text.

[0040] ⑤ Normalize Agency Abbreviations: Based on the text obtained after normalizing the launch site, scan for name fragments related to aerospace agencies. Agency abbreviations may appear as letter abbreviations or partial names. Match all suspected agency name fragments in the text with the agency abbreviation mapping table. A case-insensitive comparison method is used during matching. For successfully matched strings, replace them with the predefined standard full name of the agency in the agency abbreviation mapping table. After this step, all agency names in the text are unified to the same standard expression. Agency abbreviations refer to the abbreviations or simplified names of government agencies, space agencies, commercial companies, or international organizations responsible for space missions. The full name of an agency may be long, and abbreviations or unofficial abbreviations are often used in intelligence texts. For example, there are multiple ways to write the English abbreviation, Chinese abbreviation, and full name of a certain space agency. Normalizing agency abbreviations requires mapping these variations to a unified standard agency name.

[0041] S22. The final text content after completing the normalization process for the above five categories is output as standardized text. The standardized text no longer contains variant forms of satellite designations, orbit types, payloads, launch sites, or agency abbreviations, aliases, or foreign language translations. The standardized text is associated with and stored in relation to the identification information of its corresponding original processed intelligence text for use in subsequent step S3.

[0042] S3. Using a large model fine-tuned with aerospace data, information is extracted from standardized text. Specifically, the large model fine-tuned with aerospace data performs named entity recognition, relation extraction, attribute extraction, and event extraction on the standardized text to obtain extracted information. The extracted information includes the identified entities, extracted relations, extracted attributes, and extracted events. The specific implementation process is as follows: S30. Collect publicly released space intelligence texts from history, including satellite launch announcements, space press releases, orbital parameter reports, and mission status updates. Organize a manual annotation team to annotate these texts. Annotation content includes: marking the location and type of satellite, orbit, payload, launch site, and organizational entities in each text; marking the relationship type between each pair of entities; marking the attribute name and value for each entity; and marking the event type, event time, trigger words, and involved entities in each text. Organize the annotation results into the input / output format required for model fine-tuning, with each training sample containing the original standardized text and corresponding annotation labels.

[0043] S31. Load pre-trained parameters from a large open-source model library, such as BERT (Bidirectional Encoder Representations from Transformers), LLaMA 2 (Large Language Model MetaAI 2), or GLM (General Language Model). This model has already undergone self-supervised pre-training on a large-scale general corpus. Based on the pre-trained model, add four task output layers for named entity recognition, relation extraction, attribute extraction, and event extraction. The named entity recognition output layer uses a sequence labeling architecture, outputting an entity label for each character in the input text. The relation extraction output layer uses a classification architecture, outputting a relation type after inputting vector representations of two entities. The attribute extraction output layer uses a key-value pair prediction architecture. The event extraction output layer uses a multi-classification architecture combined with a pointer network. Configure hyperparameters such as learning rate, batch size, and number of training epochs during fine-tuning.

[0044] S32. Divide the prepared aerospace domain labeled dataset into training, validation, and test sets. For each sample in the training set, input standardized text into the large model, and perform forward propagation to obtain the prediction results of the four task output layers. Calculate the cross-entropy loss between the named entity recognition prediction result and the true label, the loss between the relation extraction prediction result and the true relation, the loss between the attribute extraction prediction result and the true attribute value, and the loss between the event extraction prediction result and the true event. Sum the four losses in a weighted manner to obtain the total loss. Use the backpropagation algorithm to calculate the gradient of the model parameters with respect to the total loss, and use the optimizer to update the model parameters. Repeat the above process, iterating multiple times on the training set. After each round, evaluate the model performance on the validation set, and save the model parameters with the best performance on the validation set as the fine-tuned large model.

[0045] Aerospace data fine-tuning refers to the process of performing additional supervised training on a pre-trained large model using a standardized text dataset labeled with entities, relationships, attributes, and events in the aerospace field. During fine-tuning, the model continues to update its weights based on the pre-trained parameters to minimize prediction errors on labeled aerospace data. After fine-tuning, the parameter distribution of the large model shifts towards the aerospace domain, enabling accurate identification of technical terms such as satellite designations, orbit types, and payload names, and extraction of the complex relationships between them. The training data required for fine-tuning comes from manually labeled aerospace intelligence samples and historical data labeled semi-automatically.

[0046] S33. The standardized text output from step S2 is used as the input data for this step. Read the complete content of the standardized text and segment it into several independent sentences according to sentence boundaries. The length of each sentence should not exceed the maximum input length limit of the large model. Package each sentence into the model input format and add necessary special markers. Then: ① Perform named entity recognition on standardized text: Each sentence in the standardized text is sequentially input into a large model fine-tuned with aerospace data. The model generates an entity label for each character in the sentence through a sequence labeling output layer. The entity label types include satellite entities, rocket entities, orbital entities, payload entities, launch site entities, space mission entities, and organizational entities. Based on the label sequence, consecutive character segments are merged into a complete entity, and the type of each entity is recorded. References to the same entity across sentences are temporarily stored independently, awaiting entity disambiguation and association processing in subsequent steps. After performing named entity recognition on all sentences, all entities and their type labels in the current standardized text are obtained. ② For all identified entities, traverse the entity pairs that appear simultaneously within each sentence. For each entity pair, concatenate the two entity vectors and the text segment vector between the two entities, and input them into the relation extraction output layer of the large model. The relation extraction output layer outputs a probability distribution representing the probability that the entity pair belongs to each predefined relation type. The relation type with the highest probability exceeding a preset threshold is selected as the extraction result. Output relation triples, where the first element is the head entity, the second element is the relation type, and the third element is the tail entity. ③ Perform attribute extraction on standardized text: For each identified entity, scan the sentence containing the entity and the feature phrases describing the entity in adjacent sentences. Input the entity's vector representation and the context window text containing the entity into the attribute extraction output layer of the large model. The attribute extraction output layer outputs an attribute name and its corresponding attribute value. Combine the entity, attribute name, and attribute value into an attribute triple. For example, for a satellite entity, attribute extraction can output "orbital altitude" and "500 km" as the attribute name and attribute value. Collect all attribute triples to form an attribute extraction result list. ④ Perform event extraction on standardized text: Input each sentence in the standardized text into a large model fine-tuned with aerospace data. The model's event extraction output layer first determines whether the current sentence contains a predefined event type; if so, it outputs an event type label. Then, the model extracts the event trigger words from the sentence. Trigger words are usually verb phrases such as "launch," "enter orbit," and "change orbit." The model further identifies the subject and object entities involved in the event. For example, in a launch event, the subject entity is a rocket or satellite, and the object entity is the launch site. The model also extracts the timing information of the event from the context. The event type, trigger words, subject entity, object entity, and event time are organized into an event structure and added to the event list from which the information is extracted.

[0047] S34. Merge and package the obtained entity list, relation triplet list, attribute triplet list, and event list into a single extracted information. In the extracted information, each entity records its original text position and type label; each relation triplet records the head entity, relation type, and tail entity; each attribute triplet records the entity, attribute name, and attribute value; and each event records the event type and related elements. Use this extracted information as the input data for step S4.

[0048] S4. Match the spacecraft name, code, and number in the extracted information with a pre-stored set of unique space target identifiers, and perform entity disambiguation and association based on the matching results to obtain extracted information associated with at least one unique space target identifier. Specifically, perform similarity matching between the spacecraft name, code, and number in the extracted information and each entry in the pre-stored set of unique space target identifiers. Based on the matching results, determine the candidate unique space target identifiers corresponding to the spacecraft name, code, and number. When there are multiple candidate identifiers, use the context information in the extracted information to perform entity disambiguation to select at least one unique space target identifier, and associate the selected at least one unique space target identifier with the extracted information to obtain extracted information associated with at least one unique space target identifier. The specific implementation process is as follows: S40. Receive the extraction information output from step S3, which contains a list of identified entities. Traverse the entity list and filter out all entries of entity type satellite or spacecraft. For each spacecraft entity, extract three fields from its original text location: spacecraft name, code, and number. The spacecraft name is usually taken from the entity's main name text. The code is identified by matching short strings of alphanumeric combinations using regular expressions. The number is identified by matching the standard format of international satellite identifiers, such as a four-digit year followed by a three-digit serial number. Store each extracted spacecraft name, code, and number as an independent string variable, and record the same spacecraft entity to which they belong. The code is a short identification string assigned to a spacecraft during its development, launch, or operation. The code is usually composed of letters and numbers combined according to certain rules, such as the model code of a series of satellites. The code often does not contain complete semantic information but has a high degree of uniqueness. The same satellite may have different codes in different countries or organizations. The number refers to the spacecraft serial number officially registered and published by international organizations or national authorities. Common satellite identification numbers include the International Satellite Identifier, which consists of the launch year, launch sequence number, and launch component number. These numbers have a globally recognized format and are unique, serving as a crucial basis for spacecraft identification. The difference between a number and a designation lies in the fact that the number's format is more standardized and officially assigned.

[0049] S41. Read the pre-stored set of unique space target identifiers from the local database or index file. This set is stored in tabular form, with each row corresponding to a known spacecraft. Each row contains a primary key field storing the unique space target identifier, and several alias variant fields storing the spacecraft's standard name, various designations, various serial numbers, and other previously used names. Load the entire set into memory and build an inverted index so that any alias variant string can be quickly located to the corresponding unique space target identifier.

[0050] S42. For each extracted spacecraft name string, iterate through all alias variant entries in the pre-stored set. For each entry, calculate the edit distance between the spacecraft name string and the entry string. Edit distance refers to the minimum number of single-character edit operations required to convert one string to another. Divide the edit distance by the maximum length of the two strings to obtain a normalized difference value, and then subtract the difference value from 1 to obtain the similarity score. Collect the unique space target identifiers corresponding to all entries with similarity scores exceeding a preset threshold, such as 0.8, and add these identifiers to the candidate identifier list. If the candidate identifier list is empty, lower the similarity threshold and rematch.

[0051] S43. For each code string extracted in step S40, perform matching using the same method as in step S42. The difference is that the codes are usually shorter, so exact matching is prioritized, i.e., determining whether the code string is exactly equal to a certain code variant in the pre-stored set. If an exact match fails, edit distance similarity matching is used. Add the successfully matched spatial target unique identifier to the candidate identifier list. If the same code matches multiple different spatial target unique identifiers, add all of these identifiers to the candidate list for use in subsequent disambiguation steps.

[0052] S44. For each number string extracted in step S40, first verify whether the number string conforms to the format specifications of international satellite identifiers. If the verification passes, perform an exact match in the number field of the pre-stored set, as the number has global uniqueness. If the exact match is successful, the corresponding unique identifier of the space target is directly used as the matching result, and multiple candidates are not generated. If the exact match fails, attempt to correct the format of the number string, such as by adding missing padding characters, and then match again. Successfully matched identifiers are added to the candidate identifier list.

[0053] S45. Collect all candidate space target unique identifiers obtained in steps S42 to S44, merge and remove duplicates to form a candidate identifier list. Check the length of the candidate identifier list. If the list is empty, it means no known spacecraft was matched, and the current spacecraft entity is marked as an unknown target, and no further association is performed. If the list length is 1, directly determine the unique identifier as the final matching identifier. If the list length is greater than 1, it means there are multiple candidate identifiers, and proceed to the entity disambiguation step. Association refers to recording the unique space target identifier in the form of key-value pairs into the data structure of the extracted information after determining the unique space target identifier, so that every entity, relationship, attribute, or event in the extracted information can be traced back to the specific space target unique identifier. After association is completed, the extracted information is no longer an isolated text fragment, but has established a direct link with the engineering data in the system.

[0054] S46. Extract the context information of the current spacecraft entity from the extracted information. Context information includes the spacecraft entity's launch time, orbital altitude, orbital inclination, payload name, affiliated organization name, and mission description keywords. Read the known attribute values ​​corresponding to each candidate identifier from the pre-stored set of unique space target identifiers, including the spacecraft's official launch time, standard orbital parameters, official payload list, and affiliated organization. For each candidate identifier, calculate the matching degree between each item in the context information and the corresponding known attribute value, and sum the weighted matching degrees to obtain a comprehensive matching score. Select the candidate identifier with the highest comprehensive matching score as the selected unique space target identifier. If the highest score is lower than a preset threshold, disambiguation is deemed a failure, and the spacecraft entity is marked as requiring manual confirmation.

[0055] S47. After obtaining the unique identifier of the selected space target, process the entire extracted information as a data object. Add an identifier field at the root level of the extracted information, setting the field name to the unique identifier of the space target and the field value to the selected identifier string. Simultaneously, iterate through all relation triples, attribute triples, and event entries related to the spacecraft entity in the extracted information, appending the same unique identifier of the space target as a foreign key to each entry. Store the associated extracted information in a temporary data structure.

[0056] S48. Repeat steps S40 to S47 for all spacecraft entities in the extracted information. For spacecraft entities that cannot be matched or fail disambiguation, their association identifier field remains empty. Package all processed extracted information and output it as extracted information associated with at least one unique space target identifier. This output data is passed to step S5 for subsequent knowledge graph construction. Extracted information associated with at least one unique space target identifier refers to an enhanced data structure formed by adding one or more unique space target identifier fields to the original extracted information after the above similarity matching, candidate identifier determination, entity disambiguation, and association operations. Each spacecraft entity in this extracted information corresponds to a clear internal identifier, which can be directly used by engineering modules such as orbit calculation, collision warning, and mission planning.

[0057] S5. Generate entity-relationship-entity triples and entity-attribute-value triples based on the extracted information that has at least one unique identifier for a space target. Perform time-sequencing, normalization, and deduplication on all triples. Store the processed triples in the graph database to form a dynamically updated aerospace knowledge graph. The process of generating entity-relationship-entity triples and entity-attribute-value triples based on extracted information associated with at least one unique identifier of a space target includes: extracting each space entity and the relationships between them from the extracted information associated with at least one unique identifier of a space target; combining each relationship and its two associated space entities into an entity-relationship-entity triple; and extracting each space entity and its corresponding attribute value from the extracted information associated with at least one unique identifier of a space target; combining each attribute, its corresponding space entity, and the attribute value into an entity-attribute-value triple. The specific implementation process is as follows: 1) Receive the extracted information output from step S4, which is associated with at least one unique identifier for a space target. This extracted information already contains an entity list, where each item records the entity name, entity type, entity position in the original text, and the associated unique identifier for a space target. Traverse the entity list and filter out entries whose entity type belongs to the category of aerospace entities, including satellite entities, rocket entities, orbital entities, payload entities, launch site entities, space mission entities, and organizational entities. For each filtered aerospace entity, extract its entity name string and the unique identifier field for a space target, and store them as a single aerospace entity record in a temporary entity set. Simultaneously, generate an internal temporary node number for each entity for reference during subsequent triplet construction. A space entity refers to a space-related object with independent semantics and objective existence identified from aerospace intelligence text. Types of space entities include satellite entities, rocket entities, orbital entities, payload entities, launch site entities, space mission entities, and organizational entities. In the extracted information associated with at least one unique identifier for a space target, each aerospace entity has been appended with a unique identifier field for a space target, allowing the entity to uniquely correspond to a specific space target in the system. Space entities are the source of nodes in the knowledge graph.

[0058] 2) Read the list of relation triples from the extracted information. This list is the raw result obtained from the relation extraction operation in step S3. Each relation record contains three fields: header entity name, relation type, and tail entity name. Iterate through each relation record and search for the corresponding aerospace entity record in the temporary entity set based on the header and tail entity names. During the search, it is necessary to compare the entity name and the unique identifier of the space target to ensure that both the header and tail entities are extracted and confirmed aerospace entities. If both the header and tail entities exist in the temporary entity set, mark this relation record as a valid relation and retain its header entity reference, relation type string, and tail entity reference.

[0059] 3) For each valid relation record, create a new data structure to store entity-relation-entity triples. This data structure contains three fields: the first field stores the name of the head entity and a unique spatial target identifier (a string); the second field stores the relation type string; and the third field stores the name of the tail entity and a unique spatial target identifier (a string). Also store the node numbers of the head and tail entities in the temporary entity set within the triples for direct node location during subsequent graph database import. Add the completed triples to a new triple list variable, specifically for storing entity-relation-entity type triples.

[0060] 4) Read the attribute triplet list from the extracted information. This list is the raw result obtained from the attribute extraction operation in step S3. Each attribute record contains three fields: space entity name, attribute name, and attribute value. Iterate through each attribute record and search for the corresponding space entity record in the temporary entity set based on the space entity name. During the search, it is also necessary to match the unique identifier of the space target to ensure the uniqueness of the entity. If a corresponding space entity is found, mark this attribute record as a valid attribute and retain the space entity reference, attribute name string, and attribute value string. The attribute value may be a pure number, a quantity with units, or a text description; maintain its original format without conversion.

[0061] 5) For each attribute record marked as valid in the previous step, create a new data structure to store entity-attribute-value triples. This data structure contains three fields: the first field stores the name of the space entity and a unique string combining its name and space target identifier; the second field stores the attribute name string; and the third field stores the attribute value string. Also store the node number of the space entity in the temporary entity set into the triple, so that this attribute can be stored as a key-value pair for nodes when importing into the graph database later. Add the completed triples to another new triple list variable, which is specifically used to store entity-attribute-value type triples.

[0062] 6) Merge and package the entity-relationship-entity triple list with the entity-attribute-value triple list generated in the previous step as the output data for this step. This output data will be passed to the subsequent operations in step S5 of the final technical solution, namely, to perform temporalization, normalization, and deduplication on all triples, and store them in a graph database to form a dynamically updated aerospace knowledge graph.

[0063] The process involves temporalization, normalization, and deduplication of all triples. This includes: adding a collection time tag to each triple to achieve temporalization; converting attribute values ​​with the same attribute type but different units or expressions in the time-tagged triples to a unified dimension and format according to a preset standard to achieve normalization; and deleting duplicate triples with the same subject, predicate, object, and time tag from the normalized triples, keeping only one copy to achieve deduplication. The specific implementation process is as follows: 1) Receive all entity-relationship-entity triples and entity-attribute-value triples passed from the previous step. Read the collection timestamp corresponding to each original intelligence text from the crawling records in step S1. This collection timestamp is accurate to the second and uses the International Organization for Standardization (ISO) format. Iterate through each triple. Since the identifier of its source intelligence text was recorded when the triple was generated, the corresponding collection timestamp can be found based on this identifier. Add a field to the data structure of each triple, named "Collection Timestamp," with the value set to the aforementioned collection timestamp. Temporarily store the triples with added timestamps in a new list. At this point, all triples have completed the time-series processing.

[0064] 2) Traverse all triples and identify their types. For entity-relation-entity triples, no normalization is performed; they are passed to subsequent steps as is, as relational strings do not have issues with unit and expression diversity. For entity-attribute-value triples, further check whether their predicate attribute names belong to a predefined list of attribute types that need normalization. This list includes numerical attributes such as track height, track inclination, resolution, mass, and design life. Filter out the entity-attribute-value triples that need normalization to form a list to be normalized.

[0065] 3) Read the unit conversion table from the preset standard. This table specifies the base unit and conversion factors from various other units to the base unit for each attribute type. Iterate through each entity-attribute-value triple in the list to be normalized, extracting its attribute name and attribute value string. Parse the attribute value string, separating the numerical part and the unit part. The unit part may be Chinese units, English units, or unit symbols. Look up the unit conversion table based on the attribute name to determine if the unit of the current attribute value is consistent with the base unit. If not, multiply or divide the numerical part by the corresponding factor according to the conversion factor to obtain the value in the base unit. Retain appropriate precision for the converted value and replace the unit part with the full Chinese name of the base unit.

[0066] 4) On the basis of completing the unified dimension processing, continue to process the same triple. Read the format specification table in the preset standard, which stipulates the writing rules for numbers, including not using thousand separators, retaining a maximum of two decimal places, not using scientific notation, and not using Chinese numerals. Extract the converted numeric part string, and delete the thousand-separator commas therein. If the numeric value contains Chinese numerals such as "thirty-six thousand", call a Chinese character-to-number function to convert it into Arabic numerals. If the numeric value is expressed in scientific notation, expand it into an ordinary decimal string. Retain the specified number of decimal places for the numeric value according to the specification, and discard the excess digits. Splice the processed numeric value and the reference unit string into a complete attribute value string, with a space added therebetween. Replace the object field in the original triple with this new string.

[0067] 5) After completing the unified dimension and unified format conversion of all entity-attribute-value triples to be normalized, merge the converted triples with the entity-relation-entity triples that do not require normalization and other entity-attribute-value triples that do not require normalization to form a complete normalized triple list. Pass this list to the next step for deduplication processing.

[0068] 6) Receive the normalized triple list, create an empty result triple list and a hash set for recording the signatures of triples that have already appeared. Traverse each triple, and extract the values of four fields from the triple: subject string, predicate string, object string, and collection time tag. Splice these four strings in a fixed order to form a unique signature key, for example, "subject|predicate|object|time tag". Check whether the signature key already exists in the hash set. If it does not exist, add the current triple to the result list, and add the signature key to the hash set. If it already exists, skip the current triple and do not add it to the result list. Continue to process the next triple until all triples have been checked. After the processing is completed, only one copy is retained for each duplicate group in the result list.

[0069] 7) Use the result triple list obtained in the previous step as the output data of this operation step. All triples in this output data have undergone timing processing, normalization and deduplication, and can be directly used for subsequent graph database storage operations. Pass this output data to the next operation of step S5 in the final technical solution, that is, "storing the processed triples into a graph database to form a dynamically updated aerospace knowledge graph".

[0070] The process involves storing the processed triples in a graph database to form a dynamically updated aerospace knowledge graph. This includes: constructing nodes and edges in the graph database based on the subject and predicate of the processed triples; assigning a unique identifier to each space target node; using the relation in each entity-relation-entity triple as a directed edge connecting two nodes; storing the attribute in each entity-attribute-value triple as a node's attribute key-value pair; establishing a time index in the graph database to support time-series queries; and incrementally updating existing nodes and edges each time a new triple is stored, thus forming a dynamically updated aerospace knowledge graph. The specific implementation process is as follows: 1) Connect to the pre-installed graph database service, which is already started and configured with storage paths and memory parameters. Check if a graph space or graph name exists in the database for storing the aerospace knowledge graph. If not, execute the create command to create a new blank graph and set a primary key index for fast lookup of node identifiers. Simultaneously, prepare a memory buffer for temporarily storing nodes and edges to reduce frequent disk write operations.

[0071] 2) Receive the processed list of triples passed from the previous step. Iterate through the entire list, extracting the subject field from each triple. The subject field contains the name of the space entity and the unique identifier of the associated space target. Parse the unique identifier of the space target from the subject string and add it to a temporary set. The set is automatically deduplicated, ultimately yielding a list of all unique identifiers of space targets to be created or updated for this database insertion.

[0072] 3) Iterate through the list of unique space target identifiers obtained in the previous step. For each identifier in the list, perform a node lookup operation in the graph database to determine if a node with that identifier as its unique key already exists. If the node does not exist, create a new node and set the unique space target identifier as the node's primary key attribute. Attach a fixed label type to the node; the label name can be "spacecraft" or dynamically determined based on the entity type. If the node already exists, skip the creation step and directly locate that node to prepare for subsequent attribute updates. Record the mapping relationship between each unique space target identifier and the node's internal identifier.

[0073] 4) Traverse the processed triple list and filter entries of type entity-relation-entity triples. For each such triple, extract the unique spatial target identifiers for the head entity and tail entity. Using the mapping relationship established in the previous step, find the graph database nodes corresponding to these two identifiers. Check if a directed edge of the same type as the relation type in the triple already exists between the source node and the target node. If not, create a new directed edge, setting the source node to the head entity node, the target node to the tail entity node, and the edge's type label to the relation type string in the triple. Simultaneously, attach the collection timestamp (i.e., collection timestamp) as an attribute to this directed edge. If a directed edge of the same type already exists, merge the current collection timestamp into the edge's attribute list, retaining the earliest and latest collection timestamps.

[0074] 5) Traverse the processed triple list and filter out entries of type entity-attribute-value triples. For each such triple, extract the unique spatial target identifier carried in the subject and find the corresponding graph database node. Read the attribute name and attribute value from the triple. Check if the node already has an attribute key with the same name. If not, add the attribute name as the key and the attribute value as the value to the node's attribute set. If an attribute key with the same name already exists, compare the existing attribute value with the current attribute value. If they do not match, determine which attribute value is more recent based on the collection time stamp, or merge and store the two attribute values ​​in a list format, while recording their respective time stamps.

[0075] 6) Create a dedicated time index in the graph database for the data collection time tag fields stored in all nodes and edges. The time index is created using the database's provided indexing command, specifying the index type as a time range index. Indexes are created on the data collection time tags in the node attributes, and also on the data collection time tag attributes of the edges. After the indexes are created, verify their effectiveness by executing a simple time range query to test the response time. The time index enables rapid location of target data when subsequently retrieving triples based on time conditions.

[0076] 7) Submit all new nodes, new directed edges, new attribute key-value pairs, and new timestamps processed in this operation to the graph database. The commit operation uses a transactional approach to ensure data consistency. For existing nodes, only attribute key-value pairs are merged or updated; existing attributes are not deleted. For existing directed edges, only new timestamp attributes are added or the edge's attribute list is updated; directed edges of the same type are not recreated. The entire update process does not involve deleting or fully rebuilding existing nodes and edges; only insertion and merging operations are performed. The current database connection is closed after the transaction is committed.

[0077] 8) After completing all the above steps, the graph database contains the triplet data processed this time and is integrated with the previously existing knowledge data. Record the number of triples added, the number of newly added nodes, the number of newly added directed edges, and the number of updated nodes. Mark the graph database's running status as normal and prepare it for the next batch processing. After multiple incremental updates, the aerospace knowledge stored in the graph database continues to be enriched, forming a dynamically updated aerospace knowledge graph. This knowledge graph can be directly used in the standardized query interface provided in step S6.

[0078] Graph databases are database management systems specifically designed for storing and querying graph-structured data. They use nodes and edges as basic data units, with nodes representing entities and edges representing relationships between entities. Graph databases support fast traversal of nodes and edges, making them particularly suitable for storing and querying complex relationships in knowledge graphs. A node is the most basic storage unit in a graph database, representing an independent entity. Each node in a graph database has a unique internal identifier and can have multiple attribute key-value pairs attached. Nodes are connected by directed edges. In this technical solution, each spatial target uniquely identifies a node in the graph database. An edge is a directed line connecting two nodes in a graph database, representing the semantic relationship between them. Each edge has a type label indicating the type of relationship. Edges point from the source node to the target node, exhibiting directionality. In this technical solution, the relationship in each entity-relationship-entity triple is converted into a directed edge. A directed edge is an edge with a clear direction. A directed edge originates from one node and points to another; the starting point is called the source node, and the ending point is called the target node. Directionality expresses the asymmetry of relationships. For example, in the "satellite-borne payload" relationship, the satellite is the source node and the payload is the target node; they cannot be reversed. Attribute key-value pairs are a data storage format consisting of attribute names and attribute values. The attribute name serves as the key, and the attribute value as the value, with a one-to-one correspondence. In a graph database, each node can have zero or one or more attribute key-value pairs. Attribute key-value pairs are directly appended to the nodes without needing to be connected by edges. In this technical solution, the attributes in each entity-attribute-value triple are converted into an attribute key-value pair for the node. A time index is a database index structure specifically built for the time stamp field. The time index organizes the storage location of triples in chronological order, enabling the database to quickly locate data records within a specified time range or point in time. Time indexes can be implemented using tree structures or skip list structures, supporting equality queries and range queries on the time field. Incremental updates refer to inserting newly generated processed triples into the existing graph database and updating the affected nodes and edges without rebuilding the entire knowledge graph. During incremental updates, for existing nodes, only their attribute key-value pairs are added or modified; for existing edges, only unique edges are added; and for non-existent nodes and edges, insertion operations are performed. Incremental updates avoid rebuilding the entire graph every time data is added to the database.

[0079] S6. A standardized query interface based on the aerospace knowledge graph is provided. This interface supports retrieval within the aerospace knowledge graph based on the unique identifier of the space target, mission number, time, and region, outputting a structured knowledge base. Specifically, providing the standardized query interface based on the aerospace knowledge graph and outputting the structured knowledge base includes: receiving query requests input based on at least one condition (unique identifier of the space target, mission number, time, or region); retrieving triples matching the query request from the aerospace knowledge graph; outputting the retrieved triples in a structured data format; and using the output data as the structured knowledge base. The specific implementation process is as follows: S60. Predefine the calling specifications for the standardized query interface. These specifications include a unified resource locator path, request method, request parameter names and types, and a structured data format for the response body. The query interface supports four request parameters: a unique spatial target identifier parameter, a task number parameter, a time parameter, and a region parameter. The time parameter can include two subfields: start time and end time. The region parameter can include four subfields: minimum latitude, maximum latitude, minimum longitude, and maximum longitude. Users must provide at least one of these four parameters. The interface response body uses a key-value pair object representation format. Each response object includes a subject field, a predicate field, an object field, a data collection time stamp field, and an optional confidence level field. Publish this specification document to the upper-layer calling system. A standardized query interface refers to a set of predefined, uniformly formatted application programming interfaces (APIs). This interface specifies the syntax format, parameter names, data types, and return result format for external callers to input query requests. The standardized query interface is not dependent on a specific programming language and can be called via the Hypertext Transfer Protocol (HTTP). The returned data uses a common structured data exchange format. The standardized query interface enables upper-level multi-satellite mission planning systems, space situational awareness systems, and other systems to obtain data from the aerospace knowledge graph in a unified manner.

[0080] S61. Start the service process of the standardized query interface and listen for Hypertext Transfer Protocol (HTTP) requests on the specified port. When a query request is received, parse the parameters in the request. If the request contains a spatial target unique identifier parameter, extract the value of the parameter, which may be a single identifier string or a comma-separated list of multiple identifiers. Use each spatial target unique identifier as a search condition. For each identifier, search the graph database for all triples with that identifier as the subject and all triples with that identifier as the object. Temporarily store the matched triples in the result set.

[0081] S62. If the query request includes a task number parameter, extract its value. The task number might be a string such as "2025-001". Perform a retrieval operation in the graph database, searching for entries in all entity-relationship-entity triples where the subject or object field contains the task number. Since the task number may be associated with a spatial target unique identifier, utilize the mapping table in the spatial target unique identifier set to convert the task number into a corresponding list of spatial target unique identifiers, and then retrieve the triples again according to the spatial target unique identifier method. Combine the triples obtained from the two retrievals into the result set. After the retrieval is complete, record the number of matched triples.

[0082] S63. If the query request includes a time parameter, further parse the start and end times within the time parameter. The start and end times use the International Organization for Standardization (ISO) date and time format. Utilize the time index built on the aerospace knowledge graph to perform a time range query in the graph database. The time range query targets the collection time tag field of all triples, filtering out all triples whose collection time tag is greater than or equal to the start time and less than or equal to the end time. Add the triples that meet the time criteria to the result set. If the result set already contains triples added in the previous steps, merge the currently matched triples with the existing results and take the union.

[0083] S64. If the query request includes a region parameter, extract the minimum latitude, maximum latitude, minimum longitude, and maximum longitude values. Search the graph database for all triplets related to geographic location, specifically including the latitude and longitude coordinates of the launch site entity, the region parameter of the orbital coverage, and the coordinate information of the event location. Filter out entries in these triplets whose latitude and longitude coordinates fall within the bounding box specified in the request. For launch site entities, compare whether the launch site coordinates are within the region; for event triplets, determine whether the coordinates of the event location are within the region. Add the triplets that meet the region criteria to the result set.

[0084] S65. If a query request contains two or more conditions, including a unique spatial target identifier, mission number, time, and region, these conditions are combined using a logical AND operation. First, a search is performed based on the first condition to obtain an intermediate result set. Then, based on this intermediate result set, subsequent conditions are used for filtering. For example, if the request specifies both a unique spatial target identifier and a time range, the system first finds the associated triples based on the unique spatial target identifier, then filters out triples whose collection time labels fall within the specified range. Finally, a set of triples that satisfy all conditions is obtained.

[0085] S66. Convert the set of triples obtained in the previous step into a structured data format. Iterate through each triple in the set and create a dictionary object. For entity-relation-entity type triples, the dictionary object contains four key-value pairs: the subject key corresponds to the combination of the name of the head entity and its unique spatial target identifier, the predicate key corresponds to the relation type string, the object key corresponds to the combination of the name of the tail entity and its unique spatial target identifier, and the collection timestamp key corresponds to the timestamp string appended to the triple. For entity-attribute-value type triples, the predicate key of the dictionary object corresponds to the attribute name, and the object key corresponds to the attribute value. Place all dictionary objects into an array object. Structured data format refers to a data representation form with clearly defined fields and hierarchical organization. Structured data formats support direct parsing and processing by programming languages ​​without the need for additional text cleaning or format conversion. Common structured data formats include key-value object representation or Extensible Markup Language (EXPLAIN). In structured data formats, the field names of each record are fixed and the field types are explicit. For example, a triplet outputs a dictionary structure containing four fields: subject, predicate, object, and collection time tag.

[0086] S67. Use the array object generated in the previous step as the message body of the Hypertext Transfer Protocol (HTTP) response, and set the content type in the response header to application format. Return the response to the client that called the standardized query interface. The data received by the caller is the structured knowledge base. Each record in the structured knowledge base has a clearly defined field name and type, and can be directly parsed and loaded into the memory of the mission planning system or situational awareness system. Simultaneously, record the request parameters and response data volume of this query in the log file for subsequent performance analysis and interface optimization. A structured knowledge base refers to a set of triples retrieved from the aerospace knowledge graph, output in a structured data format. The structured knowledge base is not the original internal storage format of a graph database, but rather a data view oriented towards application systems. Each piece of knowledge in the structured knowledge base has a clearly defined field label and can be directly imported into relational databases, data tables, or application memory objects for subsequent analysis or decision-making.

[0087] Optionally, in the above technical solution, after forming a dynamically updated aerospace knowledge graph, it also includes: S7. The following steps are repeated according to a preset time cycle: targeted collection and processing of multi-source aerospace intelligence; normalization using a pre-established aerospace professional dictionary; extraction of information using a large model fine-tuned with aerospace data; matching, disambiguation, and association of spacecraft names, codes, and numbers in the extracted information; generating triples and performing temporal normalization and deduplication on all triples; and storing the processed triples in a graph database. The newly stored triples are then merged with the already stored triples to achieve incremental updates to the aerospace knowledge graph. The specific implementation process is as follows: S70. During startup, the configuration file is read. The configuration file stores preset time period values ​​in key-value pairs, for example, a period of 360 seconds. This value is parsed into an integer and stored in a global variable. Simultaneously, the first execution delay parameter is read, which determines how long the system waits after startup before initiating the first execution. After completing the configuration read, the system enters a waiting state.

[0088] S71. The internal timer counts down according to a preset time period. When the timer reaches zero, an execution signal is triggered. Upon receiving the signal, the operation of directional collection and processing of multi-source aerospace intelligence in step S1 begins. Steps S1, S2, S3, S4, and S5 are executed sequentially, including generating triples, performing time-series normalization and deduplication on all triples, and storing the processed triples in the graph database. The graph database may be empty during the first execution; nodes and edges are created according to the normal process.

[0089] S72. During step S5, while storing the processed triples into the graph database, each triple written to the graph database is simultaneously copied to a temporary list. This temporary list is specifically used to record newly stored triples in this repeated execution cycle. An execution round number and the start timestamp of this execution are appended to the temporary list. The triples in the temporary list have undergone temporalization, normalization, and deduplication processing, and have been successfully written to the graph database.

[0090] S73. After the processed triples are stored in the graph database in each repeated execution cycle, the fusion process is executed. Since the incremental merging at the node and edge level has already been achieved in step S5 ("incrementally updating existing nodes and edges each time a new triple is stored"), the fusion operation in this step focuses more on checking data consistency across cycles. Each newly stored triple in the temporary list recorded in step S72 is traversed, and the graph database is searched to see if there are any stored triples with the same subject, predicate, and object but different collection time tags. If so, multiple time tag versions are retained without deletion. For entity-attribute-value type triples, if the attribute value in the newly stored triple is more up-to-date than the attribute value in the already stored triple, the attribute value is updated to the latest value, while the historical attribute value is retained as a version record. After fusion is completed, the temporary list is cleared.

[0091] S74. After completing all operations in the first repetitive execution cycle, reset the timer to the preset time period value. Enter the waiting state again, and the timer starts counting down again. During the waiting period, no data acquisition or processing operations are performed to free up computing resources. Maintain the connection to the graph database and continuously listen for external stop signals.

[0092] S75. When the timer resets to zero again, a second execution signal is triggered. All steps S71 to S73 are repeated. In each subsequent cycle, the new intelligence text collected during step S1 does not overlap with the content collected in the previous cycle. The storage to the graph database operation in step S5 automatically performs incremental updates, inserting only new nodes, edges, and attributes, without re-inserting existing content. The fusion operation in step S4 further ensures that the time version and attribute values ​​of the newly stored triples maintain a reasonable sequential relationship with the already stored triples. After each cycle is completed, the new content added to the aerospace knowledge graph precisely reflects the newly emerging aerospace intelligence within the preset time period. This process is repeated infinitely until a termination command is received.

[0093] S76. At the beginning and end of each repeated execution cycle, a record is written to the log file. The record includes the execution round number, start timestamp, end timestamp, number of intelligence texts collected this time, number of newly stored triples, and number of attributes updated during the fusion process. If any step S1 to S5 in a cycle encounters an exception, such as a network connection failure or the large model service being unavailable, the exception is captured and the error information is recorded. Then, the remaining steps of the current cycle are skipped, and the timer is reset to wait for the next cycle. The exception handling mechanism ensures that the failure of a single cycle will not affect the normal execution of subsequent cycles. After multiple repeated executions, the aerospace knowledge graph continues to receive incremental updates, always maintaining synchronization with the latest aerospace intelligence on the Internet.

[0094] This invention establishes a fully automated processing workflow through multi-source aerospace intelligence gathering, text preprocessing, intelligent extraction of a large model fine-tuned from aerospace data, unique identifier mapping and alignment of space targets, construction of a dynamically updated aerospace knowledge graph, and output of a structured knowledge base. The specific technical implementation steps are as follows: S101. Targeted Acquisition of Multi-Source Aerospace Intelligence. This involves targeted crawling of multi-source aerospace intelligence, including aerospace news, launch announcements, publicly available information from international organizations, commercial aerospace information, and official documents. The process includes webpage denoising, text extraction, data deduplication, and time normalization. A pre-configured list of Uniform Resource Locators (URLs) for aerospace intelligence sources is used to sequentially access and obtain the original Hypertext Markup Language (HTML) code of the webpages. The code is parsed into a Document Object Model (DOM) tree, and noisy nodes such as navigation bars, advertisements, and comment sections are removed based on a noise node feature rule base. The text density is calculated from the denoised structure, and continuous text blocks with a density exceeding a threshold are extracted as the main text. The hash value of the main text is calculated and compared with a stored hash value index table, discarding duplicate records. Regular expressions are used to match various date and time formats in the text, uniformly converting them to the International Organization for Standardization (ISO) format and adding a time zone offset. The output is the processed intelligence text.

[0095] S102. Text Cleaning and Aerospace Standardization. Using a pre-established aerospace professional dictionary, the satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text are normalized. This standardizes the aliases, abbreviations, and foreign language translations of various aerospace terms, and the normalized text becomes the standardized text. Five mapping tables, categorized by terminology, are loaded: satellite designation mapping table, orbit type mapping table, payload mapping table, launch site mapping table, and agency abbreviation mapping table. Each mapping table uses variant strings as keys and standard names as values. The input text is scanned, and satellite designations, orbit types, payloads, launch sites, and agency abbreviations are matched and replaced sequentially to eliminate multiple spellings of the same entity, outputting the standardized text.

[0096] S103. Intelligent Extraction Based on a Fine-Tuned Large Model from the Aerospace Field. Using a large model fine-tuned with aerospace data, named entity recognition, relation extraction, attribute extraction, and event extraction are performed on standardized text to obtain extracted information. The pre-trained large model is fine-tuned using a pre-labeled aerospace dataset, adding four task output layers. Standardized text is segmented into sentences and input into the model. Named entity recognition outputs sequence labels, identifying satellite entities, rocket entities, orbital entities, payload entities, launch site entities, space mission entities, and organizational entities. Relation extraction outputs the relation type for each pair of co-occurring entities, such as satellite payload, mission orbit, launch site, and spacecraft affiliation. Attribute extraction outputs attribute names and values ​​for each entity, such as orbital altitude, inclination, resolution, mass, and design life. Event extraction identifies event types, including launch, orbit insertion, orbit change, collision warning, mission completion, as well as trigger words, subject entities, object entities, and event time. The above results are combined into extracted information, which includes a list of entities, a list of relation triples, a list of attribute triples, and a list of events.

[0097] S104. Space Target Unique Identifier Mapping and Entity Disambiguation. The spacecraft names, designations, and serial numbers extracted from the information are automatically matched, associated, and disambiguated with the system's internal space target unique identifiers, ensuring that the intelligence can be directly integrated with engineering modules such as orbit calculation, collision warning, and mission planning. The spacecraft names, designations, and serial numbers are extracted from the extracted information. A pre-stored set of space target unique identifiers is loaded, containing the standard name, various designations, serial numbers, and alias variations for each known spacecraft. Similarity matching is performed on the names, designations, and serial numbers respectively, using edit distance for names and designations and exact matching for serial numbers, resulting in a list of candidate space target unique identifiers. If multiple candidate identifiers exist, contextual information, including launch time, orbital parameters, payload name, and affiliated organization, is used to perform a weighted match with the known attribute values ​​of each candidate identifier, selecting the identifier with the highest overall score. The selected space target unique identifier is added to each relevant entity, relationship, and attribute of the extracted information, outputting the extracted information associated with at least one space target unique identifier.

[0098] S105. Knowledge Fusion and Knowledge Graph Construction. Standard entity-relationship-entity triples and entity-attribute-value triples are generated, and after temporalization, normalization, and deduplication, they are stored in a graph database to form a dynamically updated aerospace knowledge graph. Each aerospace entity and its relationships are extracted from information associated with unique space target identifiers and combined into entity-relationship-entity triples; each aerospace entity and its attribute values ​​are extracted and combined into entity-attribute-value triples. A collection time tag is added to each triple to achieve temporalization. For triples with the same attribute type, the units are standardized according to a preset standard, such as converting orbital altitudes in different units to kilometers, and the format is standardized, such as removing thousands separators and converting Chinese numerals to Arabic numerals. Duplicate triples are removed based on the four fields of subject, predicate, object, and time tag. The processed triples are used to construct graph database nodes according to the subject, with each unique space target identifier corresponding to one node. The relationship in the entity-relationship-entity triple is used as a directed edge connecting two nodes. The attributes in the entity-attribute-value triple are stored as attribute key-value pairs for nodes. A time index is built in the graph database to support time-series queries. Each time a new triple is stored, existing nodes and edges are incrementally updated without deleting the original content; only the newly added attributes and time tags are merged, thus forming a dynamically updated aerospace knowledge graph.

[0099] S106. Structured Knowledge Base Output and Services. Provides standardized query interfaces, supporting the retrieval of intelligence data by unique space target identifier, mission number, time, and region, providing data support for upper-level systems such as multi-satellite mission planning, situational awareness, and collision warning. Defines interface specifications, supporting Hypertext Transfer Protocol (HTTP) calls. Request parameters include at least one of the following: unique space target identifier, mission number, start and end times in the time frame, and minimum, maximum, minimum, and maximum latitude in the region. Based on these parameters, the system performs searches in the graph database using time indexes and spatial conditions: by unique space target identifier, it searches for all triples containing that identifier as the subject or object; by mission number, it searches for triples containing that number, converts them to a unique space target identifier, and then searches; by time, it filters triples whose time tags are within a specified interval using the time index; by region, it filters geographical locations, such as launch site coordinates or event location coordinates, for triples within the bounding box. Logical AND filtering is used for combinations of multiple conditions. The retrieved triples are converted into a structured data format, such as key-value pair object representation. Each triple outputs a subject, predicate, object, and collection time tag field, forming a structured knowledge base that is returned to the caller.

[0100] The technical effects of this invention are as follows: 1) By constructing a large-scale model fine-tuned with aerospace data for the aerospace vertical field, fully automated crawling and preprocessing of open-source aerospace information from the internet was achieved. Compared with manual sorting or general crawling solutions, this invention uses targeted crawling, webpage denoising, text extraction, data deduplication, and time normalization processing, which can efficiently acquire multi-source aerospace intelligence, significantly reduce the manpower cost and time delay of information collection, and improve the automation rate and timeliness of intelligence acquisition.

[0101] 2) Utilizing a large-scale model fine-tuned with aerospace data, named entity recognition, relation extraction, attribute extraction, and event extraction are performed on standardized text, achieving high-precision intelligent extraction of entities, relationships, attributes, and events in the aerospace field. This model, fine-tuned with labeled aerospace data, can accurately identify entities such as satellites, rockets, orbits, payloads, launch sites, missions, and institutions; extract relationships such as satellite-payload, mission-orbit, launch-time, and target-affiliated institutions; and attributes such as orbital altitude, inclination, resolution, mass, and design life. It also identifies events such as launch, orbit insertion, orbit change, collision warning, and mission completion. The extraction results are output in a standardized, structured data format, providing high-quality input for subsequent knowledge fusion.

[0102] 3) The spacecraft names, codes, and serial numbers extracted from the information are matched for similarity with a pre-stored set of unique space target identifiers. Entity disambiguation and association are then performed using contextual information, achieving automatic mapping and alignment between the extraction results and the unique space target identifiers. The extracted information, after association, can be directly integrated with engineering application systems such as orbit calculation, collision warning, and mission planning, breaking down data barriers between intelligence data and engineering systems.

[0103] 4) By generating entity-relation-entity triples and entity-attribute-value triples, and then performing temporal sequencing, normalization, and deduplication, the data is stored in a graph database, forming a dynamically updated aerospace knowledge graph. This knowledge graph is characterized by temporal sequencing, standardization, and no redundancy. It supports retrieval based on unique space target identifiers, mission numbers, time, and region, and can provide a highly timely, high-quality, and directly usable structured knowledge base for upper-level systems such as multi-satellite mission planning, space situational awareness, and collision warning.

[0104] 5) This invention integrates multi-source intelligence gathering, text preprocessing, intelligent extraction of large models, unique identifier mapping and alignment, knowledge graph construction, and knowledge base output into an automated processing flow, and supports repeated incremental updates according to a preset time period. This design significantly improves the automation rate of intelligence processing, reduces the cost of manual intervention, and enhances the data support capabilities for space situational awareness and multi-satellite mission planning, thus possessing significant engineering application value.

[0105] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation. The scheme after adjusting the order is also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0106] like Figure 2 As shown, an embodiment of the present invention provides an automatic extraction and structuring system for Internet aerospace intelligence, comprising a data acquisition and processing module 201, a normalization processing module 202, an extraction module 203, a matching and association module 204, a knowledge graph acquisition module 205, and an output module 206. The data acquisition and processing module 201 is used to: collect and process multi-source aerospace intelligence in a targeted manner to obtain processed intelligence text; The normalization processing module 202 is used to: normalize the aerospace terminology in the processed intelligence text using a pre-established aerospace professional dictionary to obtain standardized text; The extraction module 203 is used to: extract information from standardized text using a large model finely tuned with data from the aerospace field; The matching and association module 204 is used to: match the spacecraft name, code and number in the extracted information with the pre-stored set of unique space target identifiers, and perform entity disambiguation and association based on the matching results to obtain the extracted information associated with at least one unique space target identifier; The knowledge graph acquisition module 205 is used to: generate entity-relationship-entity triples and entity-attribute-value triples based on the extracted information that is associated with at least one unique identifier of a space target; perform temporalization, normalization and deduplication on all triples; and store the processed triples into a graph database to form a dynamically updated aerospace knowledge graph. The output module 206 is used to: provide a standardized query interface based on the aerospace knowledge graph. The standardized query interface supports retrieval in the aerospace knowledge graph according to the unique identifier of the space target, mission number, time and region, and outputs a structured knowledge base.

[0107] Optionally, in the above technical solution, the data acquisition and processing module 201 is specifically used to: perform targeted crawling of multi-source aerospace intelligence, and perform webpage denoising, text extraction, data deduplication and time normalization processing on the crawling results, and use the processed results as processed intelligence text.

[0108] Optionally, in the above technical solution, the normalization processing module 202 is specifically used to: use a pre-established aerospace professional dictionary to normalize the satellite code, orbit type, payload, launch site and agency abbreviation in the processed intelligence text, unify the aliases, abbreviations and foreign language translations of various aerospace terms, and use the normalized text as a standardized text.

[0109] Optionally, in the above technical solution, the extraction module 203 is specifically used to: use a large model finely tuned by data from the aerospace field to perform named entity recognition, relation extraction, attribute extraction and event extraction on standardized text to obtain extracted information, which includes the identified entities, extracted relations, extracted attributes and extracted events.

[0110] Optionally, in the above technical solution, the matching and association module 204 is specifically used to: perform similarity matching between the spacecraft name, code, and number in the extracted information and each entry in the pre-stored set of unique space target identifiers; determine the candidate unique space target identifiers corresponding to the spacecraft name, code, and number based on the matching results; when there are multiple candidate identifiers, perform entity disambiguation using the context information in the extracted information to select at least one unique space target identifier; and associate the selected at least one unique space target identifier with the extracted information to obtain the extracted information associated with at least one unique space target identifier.

[0111] Optionally, in the above technical solution, the output module 206 is specifically used to: receive a query request input according to at least one condition such as unique identifier of space target, mission number, time or region through a standardized query interface, retrieve triples that match the query request in the aerospace knowledge graph, output the retrieved triples in a structured data format, and use the output data as a structured knowledge base.

[0112] Optionally, in the above technical solution, the knowledge graph acquisition module 205 is specifically used to: extract each aerospace entity and the relationship between each aerospace entity from the extraction information associated with at least one unique identifier of a space target, combine each relationship and its two associated aerospace entities into an entity-relationship-entity triple, and extract each aerospace entity and its corresponding attribute value from the extraction information associated with at least one unique identifier of a space target, and combine each attribute and its corresponding aerospace entity and attribute value into an entity-attribute-value triple.

[0113] Optionally, in the above technical solution, the knowledge graph acquisition module 205 is further specifically used to: construct nodes and edges in the graph database according to the subject and predicate of the processed triples, with each spatial target uniquely identified as a node, the relation in each entity-relation-entity triple as a directed edge connecting two nodes, the attribute in each entity-attribute-value triple as an attribute key-value pair for storage, establish a time index in the graph database to support time-series queries, and incrementally update the existing nodes and edges each time a new triple is stored, forming a dynamically updated aerospace knowledge graph.

[0114] Optionally, in the above technical solution, the knowledge graph acquisition module 205 is further specifically used to: add a collection time tag to each triple to achieve temporalization; convert attribute values ​​with the same attribute type but different units or different expressions in the triples after adding time tags into a unified dimension and unified format according to a preset standard to achieve normalization; and delete duplicate triples in the normalized triples where the subject, predicate, object and time tag are the same, keeping only one copy to achieve deduplication.

[0115] Optionally, the above technical solution also includes an update module, which is used to: after forming a dynamically updated aerospace knowledge graph, repeatedly perform the following steps according to a preset time cycle: targeted collection and processing of multi-source aerospace intelligence; normalization processing using a pre-established aerospace professional dictionary; extracting information using a large model finely tuned with aerospace data; matching, disambiguating, and associating spacecraft names, codes, and numbers in the extracted information; generating triples and performing temporal normalization and deduplication processing on all triples; and storing the processed triples into a graph database. The newly stored triples are then merged with the already stored triples to achieve incremental updates to the aerospace knowledge graph.

[0116] It should be noted that the beneficial effects of the Internet aerospace intelligence automatic extraction and structuring system 200 provided in the above embodiments are the same as those of the Internet aerospace intelligence automatic extraction and structuring method described above, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0117] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned methods for automatic extraction and structuring of Internet aerospace intelligence.

[0118] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described methods for automatic extraction and structuring of Internet aerospace intelligence.

[0119] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0120] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for automatic extraction and structuring of internet-based aerospace intelligence, characterized in that, include: Multi-source aerospace intelligence is collected and processed to obtain processed intelligence text. Using a pre-established aerospace professional dictionary, the aerospace terminology in the processed intelligence text is normalized to obtain standardized text; Information is extracted from the standardized text using a large model finely tuned with data from the aerospace field. The spacecraft name, code, and number in the extracted information are matched with a pre-stored set of unique space target identifiers, and entity disambiguation and association are performed based on the matching results to obtain extracted information associated with at least one unique space target identifier. Based on the extracted information that has at least one unique identifier for a space target, entity-relationship-entity triples and entity-attribute-value triples are generated. All triples are processed by temporalization, normalization and deduplication. The processed triples are stored in a graph database to form a dynamically updated aerospace knowledge graph. Based on the aforementioned aerospace knowledge graph, a standardized query interface is provided. This interface supports searching within the aerospace knowledge graph by unique space target identifier, mission number, time, and region, and outputs a structured knowledge base.

2. The method for automatic extraction and structuring of Internet aerospace intelligence according to claim 1, characterized in that, The process involves targeted collection and processing of multi-source aerospace intelligence to obtain processed intelligence text. This includes targeted crawling of multi-source aerospace intelligence, and performing webpage denoising, text extraction, data deduplication, and time normalization on the crawled results. The processed results are then used as the processed intelligence text.

3. The method for automatic extraction and structuring of Internet aerospace intelligence according to claim 1, characterized in that, Using a pre-established aerospace professional dictionary, the aerospace terminology in the processed intelligence text is normalized to obtain a standardized text. This includes: using a pre-established aerospace professional dictionary to normalize satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text, unifying the aliases, abbreviations, and foreign language translations of various aerospace terms, and using the normalized text as the standardized text.

4. The method for automatic extraction and structuring of Internet aerospace intelligence according to claim 1, characterized in that, Using a large model fine-tuned with aerospace data, information is extracted from the standardized text, including: using the large model fine-tuned with aerospace data to perform named entity recognition, relation extraction, attribute extraction, and event extraction on the standardized text to obtain extracted information, which includes the identified entities, extracted relations, extracted attributes, and extracted events.

5. An automatic extraction and structuring system for internet-based aerospace intelligence, characterized in that, It includes a data acquisition and processing module, a normalization processing module, an extraction module, a matching and association module, a knowledge graph acquisition module, and an output module; The data acquisition and processing module is used to: collect and process multi-source aerospace intelligence in a targeted manner to obtain processed intelligence text; The normalization processing module is used to: normalize the aerospace terminology in the processed intelligence text using a pre-established aerospace professional dictionary to obtain standardized text; The extraction module is used to: extract information from the standardized text using a large model finely tuned with data from the aerospace field; The matching and association module is used to: match the spacecraft name, code and number in the extracted information with a pre-stored set of unique space target identifiers, and perform entity disambiguation and association based on the matching results to obtain extracted information associated with at least one unique space target identifier; The knowledge graph acquisition module is used to: generate entity-relationship-entity triples and entity-attribute-value triples based on the extracted information that is associated with at least one unique identifier of a space target; perform temporalization, normalization and deduplication on all triples; and store the processed triples in the graph database to form a dynamically updated aerospace knowledge graph. The output module is used to: provide a standardized query interface based on the aerospace knowledge graph. The standardized query interface supports searching in the aerospace knowledge graph according to the unique identifier of the space target, mission number, time and region, and outputs a structured knowledge base.

6. The Internet-based aerospace intelligence automatic extraction and structuring system according to claim 5, characterized in that, The data acquisition and processing module is specifically used to: crawl multi-source aerospace intelligence in a targeted manner, and perform webpage denoising, text extraction, data deduplication, and time normalization on the crawled results, and use the processed results as the processed intelligence text.

7. The Internet-based aerospace intelligence automatic extraction and structuring system according to claim 5, characterized in that, The normalization module is specifically used to: use a pre-established aerospace professional dictionary to normalize the satellite designations, orbit types, payloads, launch sites, and agency abbreviations in the processed intelligence text, unify the aliases, abbreviations, and foreign language translations of various aerospace terms, and use the normalized text as the standardized text.

8. The Internet-based aerospace intelligence automatic extraction and structuring system according to claim 5, characterized in that, The extraction module is specifically used to: use a large model finely tuned with aerospace data to perform named entity recognition, relation extraction, attribute extraction and event extraction on the standardized text to obtain extraction information, which includes the identified entities, extracted relations, extracted attributes and extracted events.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for automatic extraction and structuring of Internet aerospace intelligence as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for automatic extraction and structuring of Internet aerospace intelligence as described in any one of claims 1 to 4.

Images