An open source intelligence-oriented weapon equipment knowledge graph construction method, system, device and storage medium

Abstract

The application discloses a weapon equipment knowledge graph construction method based on open source intelligence, obtains military text data from open source resources, pre-processes the military text data to obtain standardized military text data, carries out labeling processing on the standardized military text data to obtain a training set (a weapon equipment entity identification data set to be trained, a weapon equipment attribute extraction data set to be trained and a weapon equipment entity linking data set to be trained), improves a model through training of the model, and carries out recognition and attribute extraction on weapon equipment based on the improved model, and finally constructs a knowledge graph, so that the problems of low utilization rate of military open source information, time-consuming and laborious query and inconvenience caused by the reasons of scattering, uneven quality and large data volume of current open source military information are solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the research field of knowledge graph construction, and particularly relates to a weapon equipment knowledge graph construction method, system and device for open source intelligence and a storage medium. BACKGROUND

[0002] Big data and artificial intelligence technology are widely applied in various industries of the whole society, and are an important force to promote the intelligent transformation of the whole society. The in-depth application of related technology in the military field is becoming an irreversible trend. Compared with its application in civilian fields, the complexity, security and countermeasures of related technology in the military field put higher requirements.

[0003] In the military field, the realization of intelligent combat relies on more in-depth mining, analysis, distribution and utilization of data. The knowledge graph technology, with its strong ability to integrate and correlate knowledge, is exactly what military strategic analysis needs. Military texts, as an important carrier of military intelligence, are the cornerstone of subsequent work. In practice, due to the large size of intelligence data, the discrete distribution of data sources and the inconsistency of data formats, there are many problems in data mining and application, which seriously hinder the development of artificial intelligence technology, especially the large-scale application of knowledge graph technology in the military field.

[0004] Overall, the related technology of knowledge graph has gradually matured, and domestic and foreign enterprises, research institutions and universities have successfully applied knowledge graph in various industries and brought significant benefits. Foreign countries started earlier in this field and conducted relatively in-depth research. Knowledge graph technology has been successfully applied in search technology, recommendation system, user portrait, intelligence analysis and many other fields. While China continues to study the application of knowledge graph in information retrieval, education, transportation, finance, medical treatment and other fields, it should continue to study the application mode of knowledge graph in the military and focus on solving the problems of complex multi-data structure and difficult data selection exposed by knowledge graph. In the general field, represented by Baidu, since the world conference in 2017, knowledge graph has been constructed, which is the largest knowledge graph in the world, including more than 50 billion entities and 550 billion events. Knowledge graph not only affects the way people search for information on PC and mobile devices to get answers, but also continuously optimizes the demand experience in various product lines such as maps, education and academic search. In the vertical field such as medical treatment, Pingan Good Doctor launched a medical knowledge graph containing more than 3,000 diseases, which provides a theoretical basis for intelligent consultation.

[0005] Google, IBM, Microsoft and other foreign large companies have invested a lot of effort in the research of knowledge graph related technology. Among them, the MAKG proposed by Microsoft contains an RDF dataset of 8 billion triples to provide entity embedding for more than 2 million papers, and Google's open source knowledge graph provides a structured knowledge representation of hundreds of billions of web pages, making search directly lead to answers. At the same time, in the past 5 years, the US military and major military power agencies such as the US Defense Advanced Research Projects Agency (DARPA) have carried out and supported a large number of big data research and development projects, including text deep mining, large multi-dimensional data set analysis, big data set collection and analysis, etc. At the same time, the Russian army started to develop a combat command information system in 2016, and made great progress in the research of battlefield situation big data. For the analysis and mining of big data, various countries have actively tried and made significant progress. In the military field, Nanjing Camera Star Intelligence launched a defense knowledge graph with a million entities, but the knowledge type and application point are relatively single, and the high customization for application makes it difficult to promote and apply. In general, the current knowledge graph research in the domestic military field is still in the early stage.

[0006] In summary, compared with traditional civilian big data, military data often has higher requirements for data quality. In practice, there are also problems such as uneven quality and difficult to distinguish between true and false of data scattered on the network, and due to the requirement of secrecy, a large amount of accurate military equipment information cannot be disclosed. Only from public channels such as encyclopedias, news, and forum websites can related technical parameters and related events of military equipment be obtained. How to filter high-quality data from massive low-quality data and lack of data samples through data analysis technology, and how to improve the applicability of the model in real scenarios where both data quantity and quality are lacking, are the keys to improving intelligent data analysis capabilities.

[0007] For military equipment dynamic information knowledge graph, how to mine valuable information from scattered and uneven quality open source information is the key to the successful application of this technology in the military field. However, in the field of military application, there is still a lack of an integrated method or platform that can quickly acquire, process military equipment dynamic data, accurately query and diversely display data. SUMMARY

[0008] In order to solve the deficiencies of the prior art, the present application provides a weapon equipment knowledge graph construction method for open source intelligence, to solve the problem of low utilization rate, time-consuming and laborious query, and inconvenience of military open source information caused by scattering, uneven quality, and large data volume.

[0009] Based on the above purpose, one or more embodiments of the present specification provide a weapon equipment knowledge graph construction method for open source intelligence, the method comprising:

[0010] Obtaining military text data, preprocessing the military text data to obtain standardized military text data;

[0011] According to the standardized military text data, determine the military text data to be trained, and perform label processing on the military text data to be trained to obtain a weapon equipment entity recognition data set, a weapon equipment attribute extraction data set and a weapon equipment entity linking data set to be trained;

[0012] The weapon equipment entity recognition data set to be trained is input into the corresponding model for training processing to obtain a trained entity recognition model; the weapon equipment attribute extraction data set to be trained is input into the corresponding model for training processing to obtain a trained relation extraction model; the weapon equipment entity linking data set to be trained is input into the corresponding model for training processing to obtain a trained entity linking model;

[0013] Based on the entity recognition model and the relation extraction model, the standardized military text data is subjected to weapon equipment recognition and weapon equipment attribute extraction operations respectively to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set;

[0014] Based on the entity linking model, the weapon equipment entity data set and the existing entity in the weapon equipment knowledge graph are subjected to entity linking to obtain a target candidate entity data set, the target attribute data set of the weapon equipment is determined according to the target candidate entity data set and the weapon equipment attribute data set, the target attribute data set is stored in the database, and a weapon equipment knowledge graph is obtained.

[0015] Optionally, the step of performing weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set comprises:

[0016] The semantic representation data set of the input standardized military text data is obtained through the Bert pre-training language model, the entity label sequence with the maximum probability is output through the CRF layer, and the weapon equipment entity data set is determined based on the entity label sequence; the weapon equipment attribute data set is obtained by performing weapon equipment attribute extraction operation on the standardized military text data based on the relation extraction model.

[0017] Optionally, the step of inputting the weapon equipment attribute extraction data set to be trained into the corresponding model for training processing to obtain a trained relation extraction model comprises:

[0018] The weapon equipment attribute extraction data set to be trained is converted into an attribute vector sequence by Bert Embedding, the attribute vector sequence is input into DGCNN for encoding, and an encoded attribute vector sequence is obtained;

[0019] The encoded vector sequence is input into Self-Attention, the output result is spliced with prior features, a spliced attribute vector sequence is obtained, the spliced attribute vector sequence is input into CNN and Dense layer, the head and tail positions of the weapon equipment are predicted, and an intermediate relationship extraction model is obtained;

[0020] Any weapon equipment is input into the intermediate relationship extraction model for training, and the trained relationship extraction model is obtained.

[0021] Optionally, the step of obtaining the trained relationship extraction model by inputting any weapon equipment into the intermediate relationship extraction model includes:

[0022] Any weapon equipment is sampled, and a sub-attribute vector sequence corresponding to the weapon equipment is input into LSTM, and an attribute encoding vector sequence of the weapon equipment is obtained;

[0023] The encoded attribute vector sequence is input into another layer of Self-Attention, the output result is spliced with the attribute encoding vector sequence of the weapon equipment, and a target attribute vector sequence is obtained;

[0024] The target attribute vector sequence is input into CNN and Dense, the head and tail positions of the weapon equipment are predicted, and the trained relationship extraction model is obtained.

[0025] Optionally, the step of obtaining the target candidate entity data set by performing entity linking on the weapon equipment entity data set and the existing entities in the weapon equipment knowledge graph based on the entity linking model includes:

[0026] The candidate entity linking data set is obtained by crawling based on Baidu Encyclopedia and Wikipedia according to the standardized military text data, and the candidate entity linking data set is obtained.

[0027] The candidate entity linking data set and the corresponding entity referents are spliced to obtain a spliced candidate entity linking data set, and the candidate entity and the entity referents are separated by a separator.

[0028] The concatenated candidate entity linking dataset is input into each BERT pre-training model, and after the concatenated candidate entity linking dataset enters a binary classification Dence layer through a Dropout layer, the output result is concatenated and input into a Dence layer to obtain a plurality of concatenated candidate entity linking datasets;

[0029] The plurality of concatenated candidate entity linking datasets are input into a Softmax, and the candidate entity dataset with the highest score is taken as a target candidate entity dataset.

[0030] Optionally, the method further comprises:

[0031] Based on the entity name of the target candidate entity dataset, a search is performed in a knowledge graph database, and a first query result is returned.

[0032] If the first query result is NIL, the target candidate entity dataset and the corresponding target attribute dataset are stored in the knowledge graph database.

[0033] Optionally, the method further comprises:

[0034] Based on the attribute name of the target attribute dataset, a search is performed in the knowledge graph database, and a second query result is returned. If the second query result is NIL, the attribute is stored in the attribute database of the corresponding entity of the knowledge graph. Otherwise, the attribute is merged with the attribute of the corresponding entity, and is stored in the attribute database of the corresponding entity of the knowledge graph.

[0035] One or more embodiments of the present specification provide an open source intelligence-oriented weapon equipment knowledge graph construction system, characterized in that the system comprises:

[0036] An acquisition module is configured to acquire military text data, and preprocess the military text data to obtain standardized military text data.

[0037] A labeling module is configured to determine to-be-trained military text data according to the standardized military text data, and label the to-be-trained military text data to obtain a to-be-trained weapon equipment entity recognition dataset, a to-be-trained weapon equipment attribute extraction dataset, and a to-be-trained weapon equipment entity linking dataset.

[0038] The computing module is configured to input the to-be-trained weapon equipment entity recognition data set into a corresponding model for training processing to obtain a trained entity recognition model, input the to-be-trained weapon equipment attribute extraction data set into a corresponding model for training processing to obtain a trained relation extraction model, and input the to-be-trained weapon equipment entity linking data set into a corresponding model for training processing to obtain a trained entity linking model; and perform weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set.

[0039] The determining module is configured to perform entity linking on the weapon equipment entity data set and an existing entity in a weapon equipment knowledge graph based on the entity linking model to obtain a target candidate entity data set, determine a target attribute data set of the weapon equipment based on the target candidate entity data set and the weapon equipment attribute data set, store the target attribute data set in a database, and obtain the weapon equipment knowledge graph.

[0040] One or more embodiments of the present specification provide a weapon equipment knowledge graph construction device for open source intelligence, and the device comprises:

[0041] The obtaining unit is configured to obtain military text data, pre-process the military text data, and obtain standardized military text data.

[0042] The labeling unit is configured to determine to-be-trained military text data based on the standardized military text data, and perform labeling processing on the to-be-trained military text data to obtain a to-be-trained weapon equipment entity recognition data set, a to-be-trained weapon equipment attribute extraction data set, and a to-be-trained weapon equipment entity linking data set.

[0043] The computing unit is configured to input the to-be-trained weapon equipment entity recognition data set into a corresponding model for training processing to obtain a trained entity recognition model, input the to-be-trained weapon equipment attribute extraction data set into a corresponding model for training processing to obtain a trained relation extraction model, and input the to-be-trained weapon equipment entity linking data set into a corresponding model for training processing to obtain a trained entity linking model; and perform weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set.

[0044] The determining unit is configured to perform entity linking on the weapon equipment entity data set and existing entities in the weapon equipment knowledge graph based on an entity linking model to obtain a target candidate entity data set, determine a target attribute data set of the weapon equipment according to the target candidate entity data set and the weapon equipment attribute data set, store the target attribute data set in a database, and obtain the weapon equipment knowledge graph.

[0045] The one or more embodiments of the present specification provide a storage medium having computer instructions stored thereon, characterized in that: the computer instructions perform the steps of the weapon equipment knowledge graph construction method for open source intelligence when running.

[0046] It can be seen from the above that the weapon equipment knowledge graph construction method based on open source intelligence provided by one or more embodiments of the present specification obtains military text data from open source resources, pre-processes the military text data to obtain standardized military text data, and changes the disordered data into standard and unified available data after processing. According to the standardized military text data, determine the military text data to be trained, and perform annotation processing on the military text data to be trained to obtain weapon equipment entity recognition data set, weapon equipment attribute extraction data set and weapon equipment entity linking data set to be trained, and process the data required for model training. The weapon equipment entity recognition data set to be trained is input into the corresponding model for training processing to obtain the trained entity recognition model. The weapon equipment attribute extraction data set to be trained is input into the corresponding model for training processing to obtain the trained relation extraction model. The weapon equipment entity linking data set to be trained is input into the corresponding model for training processing to obtain the trained entity linking model. Then, based on the trained model, recognition and attribute extraction are performed, that is, based on the entity recognition model and the relation extraction model, weapon equipment recognition and weapon equipment attribute extraction operations are performed on the standardized military text data respectively to obtain weapon equipment entity data set and corresponding weapon equipment attribute data set. Finally, based on the entity linking model, the weapon equipment entity data set and the existing entity in the weapon equipment knowledge graph are linked to obtain a target candidate entity data set, and the target attribute data set of the weapon equipment is determined according to the target candidate entity data set and the weapon equipment attribute data set. The target attribute data set is stored in the database to obtain the weapon equipment knowledge graph. The method can improve the model by training the model based on the standardized open source data, and can recognize and extract the attributes of the weapon equipment based on the improved model. Finally, the knowledge graph is constructed, which solves the problems of low utilization rate of military open source information, time-consuming and laborious query and inconvenience caused by scattered, uneven quality and large amount of data of current open source military information. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0048] Figure 1A flowchart of a weapon equipment knowledge graph construction method for open source intelligence in an embodiment of the present application;

[0049] Figure 2 A flowchart of another weapon equipment knowledge graph construction method for open source intelligence in an embodiment of the present application;

[0050] Figure 3 A flowchart of an open source intelligence acquisition workflow in an embodiment of the present application;

[0051] Figure 4 An attribute extraction data example graph in an embodiment of the present application;

[0052] Figure 5 A BERT-based entity linking model graph in an embodiment of the present application;

[0053] Figure 6 A structure schematic diagram of a weapon equipment knowledge graph construction device for open source intelligence provided in an embodiment of the present application

[0054] Figure 7 A structure schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described below in detail with specific embodiments and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0056] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present application should be understood as the usual meaning understood by those skilled in the art to which the present disclosure belongs. The terms "first", "second" and the like used in one or more embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and the like mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and the like are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0057] Knowledge graph: Knowledge graph is a theory of the present age which combines the theories and methods of mathematics, graphics, information visualization technology, information science and other disciplines with citation analysis, co-occurrence analysis and other methods, and uses visualized graph to show the core structure, development history, frontier field and overall knowledge architecture of a discipline to achieve the purpose of multi-disciplinary integration.

[0058] Entity recognition: Entity recognition is an information extraction technology. This technology can obtain entity-related data such as names, places, and organizations from text data through deep learning-related methods.

[0059] Entity relationship extraction: Entity relationship recognition refers to the task of extracting the implied relationship between entities in the text during natural language processing. The relationship between the extracted entities can be expressed in different forms of language or form, such as relational databases, XML, etc.

[0060] Entity linking: Entity linking refers to the task of mapping a string of text containing a specific meaning in a knowledge base to the corresponding entity. This technology is one of the key technologies for knowledge graph construction and knowledge updating, and realizes the function of mapping ambiguous named reference items to the entities they refer to, solving the problem of polysemy in knowledge graph construction.

[0061] Entity attribute alignment: Entity attribute alignment aims to determine whether multiple sources of attributes refer to the same attribute, and integrates information with the same meaning from different sources to enrich the knowledge base.

[0062] The inventors find that due to the huge size of intelligence data, the discrete distribution of data sources and the inconsistency of data formats, many problems are presented when data mining applications are carried out, which seriously hinders the large-scale application of knowledge graph technology in the military field. At present, the open source military information has the problems of low utilization rate, time-consuming and laborious query and inconvenience due to the reasons of scattering, uneven quality and large data volume. Therefore, the inventors provide a weapon equipment knowledge graph construction method for open source intelligence, obtain military text data from open source resources, preprocess the military text data to obtain standardized military text data, and change the disordered data into standard and unified available data after processing. The standardized military text data are used to determine the military text data to be trained, and the military text data to be trained are labeled to obtain the weapon equipment entity recognition data set to be trained, the weapon equipment attribute extraction data set to be trained and the weapon equipment entity linking data set to be trained, so that the data required for model training is well processed. The weapon equipment entity recognition data set to be trained is input into the corresponding model for training processing to obtain the trained entity recognition model. The weapon equipment attribute extraction data set to be trained is input into the corresponding model for training processing to obtain the trained relation extraction model. The weapon equipment entity linking data set to be trained is input into the corresponding model for training processing to obtain the trained entity linking model. Then, based on the trained model, recognition and attribute extraction are carried out, that is, the standardized military text data are subjected to weapon equipment recognition and weapon equipment attribute extraction operations based on the entity recognition model and the relation extraction model, to obtain the weapon equipment entity data set and the corresponding weapon equipment attribute data set. Finally, based on the entity linking model, the weapon equipment entity data set is subjected to entity linking with the existing entities in the weapon equipment knowledge graph to obtain a target candidate entity data set, the target attribute data set of the weapon equipment is determined based on the target candidate entity data set and the weapon equipment attribute data set, the target attribute data set is stored in a database, and a weapon equipment knowledge graph is obtained. The open source data after standardization is labeled to obtain the training set (the weapon equipment entity recognition data set to be trained, the weapon equipment attribute extraction data set to be trained and the weapon equipment entity linking data set to be trained), the model is trained to improve the model, the weapon equipment is recognized and the attribute is extracted based on the improved model, and finally the knowledge graph is constructed, so that the problems of low utilization rate, time-consuming and laborious query and inconvenience of the open source military information due to the reasons of scattering, uneven quality and large data volume are solved, the information is mined from the unstructured text, the related data and resources are integrated, the military text data set is expanded, the informatization and intelligentization of the traditional science and technology / military intelligence work are realized, and the auxiliary decision-making ability for the intelligence analysis work of science and technology intelligence workers is provided.

[0063] Various non-limiting embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0064] Referring to Figure 1 , a weapon equipment knowledge graph construction method for open source intelligence in an embodiment of the present application is shown, and the method comprises:

[0065] S301: Obtain military text data, pre-process the military text data to obtain standardized military text data, determine the military text data to be trained according to the standardized military text data, and perform annotation processing on the military text data to be trained to obtain a weapon equipment entity recognition data set, a weapon equipment attribute extraction data set and a weapon equipment entity linking data set to be trained.

[0066] Specifically, the military text data is obtained, and the obtaining manner can be an automatic obtaining manner. Specifically, the military text data can be obtained from an open source obtaining channel through network crawler, timer, multi-threading, resource pool and other technologies, such as obtaining open source military text data from news websites and government websites.

[0067] To more accurately obtain the military text data, an embodiment provides an open source intelligence obtaining workflow, which is specifically shown in Figure 3 The obtained military text data includes but is not limited to tables, library table data, and institutional analysis is performed on these data. The obtaining channel can be website data, access history data, billion obtaining data, or data obtained by using a crawler. The data obtaining can also be in a real-time obtaining state, and real-time data is provided, supporting data subscription, machine automatic flow obtaining and other functions.

[0068] Based on data obtaining, weapon equipment and performance related information are automatically identified from the data, thereby providing support for the function of constructing a weapon equipment knowledge graph.

[0069] For example, the news and government websites involved can be more than 60, and the websites are shown in Table 1 as follows, but the open source obtaining channel is not limited to the following websites. As long as the website is public and accessible, it can be an obtaining channel of the present application. Open source intelligence obtaining provides military text data support for the present application, and is a data source for the weapon equipment knowledge graph.

[0070] Table 1 website list example

[0071]

[0072] Data preprocessing preliminarily processes the military text in the open source intelligence into a data form required by each model of the knowledge graph, thereby providing a data set for the present application.

[0073] The data preprocessing mainly includes two parts of original data preprocessing and training set labeling. The original data (military text data) preprocessing realizes standardization processing of original intelligence data, mainly including data cleaning, word segmentation, sentence segmentation and the like. The military text data is preprocessed to obtain standardized military text data. According to the standardized military text data, the military text data to be trained is determined, and the military text data to be trained is labeled to obtain the weapon equipment entity recognition data set to be trained, the weapon equipment attribute extraction data set to be trained and the weapon equipment entity linking data set to be trained.

[0074] The data labeling tool provided by the application can label the entities, entity categories and relationships between entities on the standardized military text data, construct the weapon equipment entity recognition data set to be trained, the weapon equipment attribute extraction data set to be trained, and the weapon equipment entity linking data set to be trained, and then combine the search results of Wikipedia and Baidu Encyclopedia to construct the weapon equipment entity linking data set to be trained.

[0075] The entity categories mainly include weapon equipment, organization, person, country and the like.

[0076] The entity relationship (attribute) mainly includes a plurality of relationships such as number, equipment alias, code, side number, use country, belonging country, previous type, modified type, width, height, diameter, draught, wingspan, weight, production time, research and development time, service time, researcher, research and development unit, production unit, speed, loading, number of personnel, number of crew members, range, derivative type, muzzle velocity, maximum flight speed, maximum range, payload, combat radius, first flight time, launch mode, endurance, boiler, fuel, type, displacement, full load displacement, submerged displacement, submerged depth, hit accuracy, guidance system, aspect ratio, ammunition capacity, self-sustaining force, waterline width, economic speed, warhead weight, action distance, cruising altitude, bullet diameter and the like. In the entity linking data set, each data includes a plurality of candidate entities and a true result of the data, and the plurality of candidate entities can refer to 20, 30 or the like.

[0077] Optionally, the military text data is expanded. Since the open source intelligence related data contains limited military text data with available information, and there is no mature data set with large data volume at present, how to expand the data set to improve the accuracy of the model has been a problem to be solved. The data expansion of the application provides more sufficient data volume.

[0078] The application expands the data set by replacing words with the same word but different meanings. That is, by first labeling a part of the real military text data, and then replacing the entities in the text with other known entities of the same type. For example: in the sentence "B-type large anti-submarine escort ship is part of the A-type ship's drawing based on the transfer of a certain country, designed by K Institute." Contains the "certain" country entity and the "B-type large anti-submarine escort ship" weapon equipment entity, we can replace "certain country" with "certain country", and replace "B-type large anti-submarine escort ship" with "C ship" to construct a new data. Although this data is not real in terms of meaning, it can help train model parameters through sentence structure and context semantics.

[0079] S302: input the to-be-trained weapon equipment entity recognition data set into the corresponding model for training processing to obtain a trained entity recognition model; input the to-be-trained weapon equipment attribute extraction data set into the corresponding model for training processing to obtain a trained relationship extraction model; and input the to-be-trained weapon equipment entity linking data set into the corresponding model for training processing to obtain a trained entity linking model.

[0080] Specifically, the application uses a Bert-BiLSTM-CRF model to identify weapons and equipment and weapons and equipment related information in military text. The semantic representation data set of the input standardized military text data is obtained through the Bert pre-training language model, the CRF layer outputs the entity label sequence with the maximum probability, and the weapon equipment entity data set is determined based on the entity label sequence; the relationship extraction model is used to perform weapon equipment attribute extraction operation on the standardized military text data to obtain the weapon equipment attribute data set. Weapon equipment related entities include DATE, COUNTRY, PER, WEAPON, ORG, etc.

[0081] Table 2 weapon equipment entity recognition data example

[0082]

[0083]

[0084] In the data set of the application, the weapon equipment and related attribute information can be displayed in the form of triplets, and the example data is shown in Figure 4The input of the relation extraction model is a military intelligence text, and the output is all relation triples contained in the sentence. The triple is in the form of ["s", "p", "o"], where s represents the subject, o represents the object, that is, the attribute or target entity of the relation, and p represents the predicate, that is, the relationship type between the two entities. When performing relation extraction, all predicates are agreed in advance.

[0085] To train the relation extraction model, real data needs to be labeled. Optionally, the labeled data can be 2w pieces of labeled data, and the accuracy of attribute extraction is related to the quality of these labeled data. The relation extraction uses a Bert-CNN-Attention model to identify the subject s first, and then input s to identify the attribute o and the relationship type p at the same time. In the step of inputting the weapon equipment attribute extraction data set to be trained into the corresponding model for training processing to obtain the trained relation extraction model, specifically, the weapon equipment attribute extraction data set to be trained is converted into an attribute vector sequence through Bert Embedding, and the attribute vector sequence is input into DGCNN for encoding to obtain an encoded attribute vector sequence.

[0086] The coded vector sequence is input into Self-Attention, the output result is spliced with the prior feature to obtain a spliced attribute vector sequence, and the spliced attribute vector sequence is input into CNN and Dense layer to predict the head and tail positions of the weapon equipment, and obtain an intermediate relation extraction model. According to the intermediate relation extraction model, any weapon equipment is input to obtain the trained relation extraction model.

[0087] Specifically, any weapon equipment is sampled, the encoded attribute vector sequence corresponding to the sub-attribute vector sequence of the any weapon equipment is input into LSTM to obtain the attribute encoding vector sequence of the any weapon equipment; the encoded attribute vector sequence is input into another layer of Self-Attention, and the output result is spliced with the attribute encoding vector sequence of the any weapon equipment to obtain a target attribute vector sequence; the target attribute vector sequence is input into CNN and Dense to predict the head and tail positions of the weapon equipment, thereby obtaining the trained relation extraction model. For each type of relation (p), a "half-pointer-half-labeled" structure is constructed to predict the head and tail positions of the corresponding o, and thus o and p are predicted.

[0088] After the relation extraction model is trained, the model can be used for weapon equipment attribute extraction to construct a weapon equipment knowledge graph.

[0089] S303: Based on the entity recognition model and the relation extraction model, the standardized military text data are subjected to weapon equipment recognition and weapon equipment attribute extraction operations respectively, to obtain a weapon equipment entity dataset and a corresponding weapon equipment attribute dataset.

[0090] The standardized military text data are input into the trained entity recognition model and the relation extraction model respectively, to perform weapon equipment recognition and weapon equipment attribute extraction operations, to obtain a weapon equipment entity dataset and a corresponding weapon equipment attribute dataset. At this time, the rudimentary data of the weapon equipment knowledge graph has been formed, and after entity linking and attribute alignment, it becomes the standardized weapon equipment knowledge graph.

[0091] S304: Based on the entity linking model, the weapon equipment entity dataset is subjected to entity linking with the existing entities in the weapon equipment knowledge graph, to obtain a target candidate entity dataset, and based on the target candidate entity dataset and the weapon equipment attribute dataset, a target attribute dataset of the weapon equipment is determined, the target attribute dataset is stored into a database, and a weapon equipment knowledge graph is obtained.

[0092] Entity linking is used to solve the problem of multi-word one meaning or one word multiple meanings, and provides a basis for new equipment discovery and attribute completion function. The model construction process mainly includes candidate entity generation and candidate entity sorting.

[0093] Candidate entity generation: In this application, by combining Baidu Encyclopedia and Wikipedia, mention (in the entity linking task, we call the detected word (or word group) to be linked as mention.) is taken as the input of Baidu Encyclopedia (Wikipedia), the first n data of the search results in the category of weapon equipment (after preliminary screening, the advertisements and other irrelevant items are deleted) are crawled as the candidate entity to construct the entity linking dataset, wherein n is selected from any integer, such as n = 20 or 30.

[0094] Candidate entity sorting: In this application, feature extraction is performed by using the BERT pre-training model, so as to complete the candidate entity sorting. The model structure of the BERT (Bidirectional Encoder Representations from Transformers) entity linking model used in this application is shown in Figure 5 .

[0095] Specifically, based on Baidu Encyclopedia and Wikipedia, the standardized military text data are crawled to obtain a candidate entity linking dataset, and the candidate entity linking dataset includes a weapon equipment entity dataset and a corresponding weapon equipment attribute dataset.

[0096] The candidate entity linking data set and its corresponding entity mention are spliced to obtain a spliced candidate entity linking data set, and the candidate entity and the entity mention are separated by a separator;

[0097] The spliced candidate entity linking data set is input into each BERT pre-training model, the spliced candidate entity linking data set enters a two-class Dense layer after a Dropout layer, the output result is spliced and input into a Dense layer to obtain a plurality of spliced candidate entity linking data sets;

[0098] The plurality of spliced candidate entity linking data sets are input into a Softmax, and the candidate entity data set with the highest score is taken as a target candidate entity data set.

[0099] The entity attribute alignment provides the function of finding the corresponding attribute in the knowledge graph corresponding to the entity attribute information in the military text, and provides support for weapon equipment attribute completion.

[0100] In the process of attribute alignment, since the schema of entity relation extraction has been determined, the entity attribute alignment process of the present application is as follows:

[0101] Optionally, the extracted entity relation is standardized by some rules, for example, date class is standardized as "YYYY year MM month DD day", "YYYY year early" is standardized as "YYYY year 01 month 01 day", "YYYY year 1 month middle" is standardized as "YYYY year 01 month 15 day", etc.

[0102] Optionally, the weapon equipment knowledge graph construction method for open source intelligence further includes: based on the entity name of the target candidate entity data set, searching in the knowledge graph database and returning a first query result; if the first query result is NIL, it means that there is no record about the entity in the original knowledge graph database, and a new record needs to be supplemented to the knowledge graph database, then the target candidate entity data set and the corresponding target attribute data set are stored in the knowledge graph database.

[0103] Optionally, the weapon equipment knowledge graph construction method for open source intelligence further includes: based on the attribute name of the target attribute data set, searching in the knowledge graph database and returning a second query result, if the second query result is NIL, storing the attribute in the attribute database of the corresponding entity of the knowledge graph; otherwise, merging the attribute and the attribute of the corresponding entity, and storing in the attribute database of the corresponding entity of the knowledge graph. Finally, the completion and update of the entity attribute are completed, and the automatic construction of the weapon equipment knowledge graph is realized.

[0104] In an embodiment of the present application, referring toFigure 2 , a detailed drawing of an example flowchart of a weapon equipment knowledge graph construction method for open source intelligence is drawn, an open source intelligence acquisition module accesses intelligence websites (a network resource pool) through a web crawler to obtain original intelligence text (military text data), and stores the original intelligence text into a military text intelligence resource pool, then pre-processes and labels the original text to obtain labeled weapon equipment entity recognition data sets, weapon equipment attribute extraction data sets and weapon equipment entity linking data sets to be trained; by inputting the trained data into the corresponding model for training, the trained entity recognition model, relationship extraction model and entity linking model are obtained; based on the entity recognition model and the relationship extraction model, the standardized military text data is subjected to weapon equipment recognition and weapon equipment attribute extraction operations respectively to obtain weapon equipment entity data sets and corresponding weapon equipment attribute data sets;

[0105] Based on the entity linking model, the weapon equipment entity data sets and the existing entities in the weapon equipment knowledge graph are subjected to entity linking to obtain target candidate entity data sets, the target attribute data sets of the weapon equipment are determined according to the target candidate entity data sets and the weapon equipment attribute data sets, the target attribute data sets are stored into a database, and the weapon equipment knowledge graph is obtained. The weapon equipment entity recognition, attribute extraction, entity linking and attribute alignment of the embodiment are consistent with the foregoing.

[0106] Corresponding to the weapon equipment knowledge graph construction method for open source intelligence, the embodiment of the present application provides a weapon equipment knowledge graph construction system for open source intelligence, the system comprising:

[0107] An acquisition module is configured to:

[0108] Acquire military text data, pre-process the military text data to obtain standardized military text data;

[0109] A labeling module is configured to:

[0110] Determine the military text data to be trained according to the standardized military text data, and label the military text data to be trained to obtain weapon equipment entity recognition data sets, weapon equipment attribute extraction data sets and weapon equipment entity linking data sets to be trained respectively;

[0111] A calculation module is configured to:

[0112] inputting the to-be-trained weapon equipment entity recognition data set into a corresponding model for training processing to obtain a trained entity recognition model; inputting the to-be-trained weapon equipment attribute extraction data set into a corresponding model for training processing to obtain a trained relation extraction model; inputting the to-be-trained weapon equipment entity linking data set into a corresponding model for training processing to obtain a trained entity linking model; performing weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set;

[0113] determining a module for:

[0114] based on the entity linking model, performing entity linking on the weapon equipment entity data set and existing entities in a weapon equipment knowledge graph to obtain a target candidate entity data set, determining a target attribute data set of the weapon equipment according to the target candidate entity data set and the weapon equipment attribute data set, storing the target attribute data set into a database to obtain the weapon equipment knowledge graph.

[0115] optionally, the computing module is further configured to:

[0116] obtaining a semantic representation data set of the standardized military text data input by a Bert pre-training language model, outputting a probability maximum entity label sequence through a CRF layer, and determining a weapon equipment entity data set based on the entity label sequence; performing weapon equipment attribute extraction operations on the standardized military text data based on the relation extraction model to obtain a weapon equipment attribute data set.

[0117] further, the computing module is further configured to:

[0118] convert the to-be-trained weapon equipment attribute extraction data set into an attribute vector sequence through Bert Embedding, input the attribute vector sequence into DGCNN for encoding to obtain an encoded attribute vector sequence;

[0119] input the encoded vector sequence into Self-Attention, splice the output result with prior features to obtain a spliced attribute vector sequence, input the spliced attribute vector sequence into CNN and Dense layer, predict the head and tail positions of the weapon equipment to obtain an intermediate relation extraction model;

[0120] input any of the weapon equipment into the intermediate relation extraction model for training to obtain the trained relation extraction model.

[0121] further, the computing module is further configured to:

[0122] Sampling any of the weapon equipment, inputting the coded attribute vector sequence corresponding to the sub-attribute vector sequence of any of the weapon equipment into the LSTM to obtain the attribute encoding vector sequence of any of the weapon equipment;

[0123] Inputting the coded attribute vector sequence into another layer of Self-Attention, splicing the output result with the attribute encoding vector sequence of any of the weapon equipment to obtain a target attribute vector sequence;

[0124] Inputting the target attribute vector sequence into CNN and Dense to predict the head and tail positions of the weapon equipment, thereby obtaining the trained relationship extraction model.

[0125] Optionally, the determining module is further configured to:

[0126] According to the standardized military text data, candidate entity link data sets are obtained based on Baidu Encyclopedia and Wikipedia crawling, and the candidate entity link data sets are obtained.

[0127] The candidate entity link data sets and their corresponding entity references are spliced to obtain spliced candidate entity link data sets, and the candidate entities and the entity references are separated by a separator.

[0128] The spliced candidate entity link data sets are input into each BERT pre-training model, and the spliced candidate entity link data sets are input into a two-class Dence layer after passing through a Dropout layer, and the output results are spliced and input into a Dence layer to obtain multiple spliced candidate entity link data sets.

[0129] The multiple spliced candidate entity link data sets are input into Softmax, and the candidate entity data set with the highest score is taken as a target candidate entity data set.

[0130] Optionally, the determining module is further configured to:

[0131] Based on the entity name of the target candidate entity data set, a search is performed in a knowledge graph database, and a first query result is returned.

[0132] If the first query result is NIL, the target candidate entity data set and the corresponding target attribute data set are stored in the knowledge graph database.

[0133] Optionally, the determining module is further configured to:

[0134] The attribute name based on the target attribute data set is looked up in the knowledge graph database, and a second query result is returned, if the second query result is null, the attribute is stored in the attribute database of the corresponding entity of the knowledge graph, otherwise, the attribute is combined with the attribute of the corresponding entity and stored in the attribute database of the corresponding entity of the knowledge graph.

[0135] Corresponding to the weapon equipment knowledge graph construction method for open source intelligence, an embodiment of the present application provides a weapon equipment knowledge graph construction device for open source intelligence, the device comprises:

[0136] The acquisition unit 601 is configured to:

[0137] Acquire military text data, and preprocess the military text data to obtain standardized military text data.

[0138] The labeling unit 602 is configured to:

[0139] According to the standardized military text data, determine the military text data to be trained, and perform labeling processing on the military text data to be trained to obtain a weapon equipment entity recognition data set to be trained, a weapon equipment attribute extraction data set to be trained and a weapon equipment entity linking data set to be trained.

[0140] The calculation unit 603 is configured to:

[0141] Input the weapon equipment entity recognition data set to be trained into a corresponding model for training processing to obtain a trained entity recognition model, input the weapon equipment attribute extraction data set to be trained into a corresponding model for training processing to obtain a trained relation extraction model, and input the weapon equipment entity linking data set to be trained into a corresponding model for training processing to obtain a trained entity linking model; perform weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set.

[0142] The determination unit 604 is configured to:

[0143] Based on the entity linking model, perform entity linking on the weapon equipment entity data set and the existing entities in the weapon equipment knowledge graph to obtain a target candidate entity data set, determine a target attribute data set of the weapon equipment based on the target candidate entity data set and the weapon equipment attribute data set, store the target attribute data set in a database, and obtain the weapon equipment knowledge graph.

[0144] Corresponding to the weapon equipment knowledge graph construction method for open source intelligence described above, the embodiment of the present application provides a storage medium having computer instructions stored thereon, characterized in that: when the computer instructions are executed, the steps of the weapon equipment knowledge graph construction method for open source intelligence described above are performed.

[0145] The electronic device of the above embodiment is used to implement the corresponding method in the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0146] Figure 7 A specific electronic device hardware structure schematic diagram provided by the embodiment is shown, which can include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040 and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030 and the communication interface 1040 are connected to each other through the bus 1050 for internal communication.

[0147] The processor 1010 can be implemented in the form of a general CPU (Central Processing Unit, central processor), a microprocessor, an ASIC (Application Specific Integrated Circuit, application specific integrated circuit) or one or more integrated circuits, etc., for executing related programs to implement the technical solutions provided by the embodiments of the present application.

[0148] The memory 1020 can be implemented in the form of a ROM (Read Only Memory, read-only memory), a RAM (Random Access Memory, random access memory), a static storage device, a dynamic storage device, etc. The memory 1020 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are saved in the memory 1020 and called and executed by the processor 1010.

[0149] The input / output interface 1030 is used to connect input / output modules to realize information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0150] The communication interface 1040 is configured to connect a communication module (not shown in the figure) to realize the communication interaction between the device and other devices. The communication module can realize the communication through a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0151] The bus 1050 includes a path for transmitting information between various components (for example, the processor 1010, the memory 1020, the input / output interface 1030 and the communication interface 1040) of the device.

[0152] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device can also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device can also only contain the components necessary for the implementation of the embodiments of the present specification, and does not have to contain all the components shown in the figure.

[0153] The computer readable medium of the embodiment includes permanent and non-permanent, removable and non-removable media, which can realize information storage by any method or technology. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0154] Those skilled in the art should understand that the above discussion of any embodiment is only exemplary and is not intended to limit the scope of the present disclosure (including the claims); under the idea of the present disclosure, the above embodiments or technical features in different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of one or more embodiments of the present specification as described above. In order to be brief, they are not provided in detail.

[0155] Additionally, to simplify the description and discussion, and so as not to obscure one or more embodiments of the description, well-known power supply / ground connections to integrated circuit (IC) chips and other components can or can not be shown in the provided figures. Furthermore, devices can be shown in block diagram form in order to avoid obscuring one or more embodiments of the description, and this also acknowledges the fact that the details in regard to the implementation of such block device are highly dependent on the platform within which the one or more embodiments of the description are being implemented (i.e., such details should be well within the purview of one of ordinary skill in the art). Where specific details are set forth in order to describe an illustrative embodiment of the disclosure, it will be apparent to one of ordinary skill in the art that the one or more embodiments of the description can be practiced without, or with variation of, these specific details. Thus, the description is to be considered as illustrative and not restrictive, and the scope of the one or more embodiments of the description is to be determined not with the assistance of the foregoing description alone, but rather in light of the appended claims in conjunction with recognizing the one or more embodiments of the description can over come a variety of non-anticipated

[0156] While the present disclosure has been described in connection with specific embodiments thereof, many alternatives, modifications and variations will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) can use the embodiments discussed.

[0157] The one or more embodiments of the description are intended to cover all such alternatives, modifications and variations as can come within the scope of the appended claims. Accordingly, any and all such modifications, variations or equivalents that fall within the scope of the one or more embodiments of the description should be considered within the scope of the disclosure.

Claims

1. An open source intelligence-oriented weapon equipment knowledge graph construction method, characterized in that, The method comprises: acquiring military text data, preprocessing the military text data to obtain standardized military text data; determining to-be-trained military text data according to the standardized military text data, and performing label processing on the to-be-trained military text data to obtain a to-be-trained weapon equipment entity recognition data set, a to-be-trained weapon equipment attribute extraction data set, and a to-be-trained weapon equipment entity linking data set; inputting the to-be-trained weapon equipment entity recognition data set into a corresponding model for training processing to obtain a trained entity recognition model, inputting the to-be-trained weapon equipment attribute extraction data set into a corresponding model for training processing to obtain a trained relation extraction model, and inputting the to-be-trained weapon equipment entity linking data set into a corresponding model for training processing to obtain a trained entity linking model; performing weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set; based on the entity linking model, performing entity linking on the weapon equipment entity data set and existing entities in a weapon equipment knowledge graph to obtain a target candidate entity data set, determining a target attribute data set of a weapon equipment based on the target candidate entity data set and the weapon equipment attribute data set, storing the target attribute data set in a database, and obtaining a weapon equipment knowledge graph, wherein the step of inputting the to-be-trained weapon equipment attribute extraction data set into a corresponding model for training processing to obtain a trained relation extraction model comprises: converting the to-be-trained weapon equipment attribute extraction data set into an attribute vector sequence through Bert Embedding, inputting the attribute vector sequence into DGCNN for encoding to obtain an encoded attribute vector sequence, inputting the encoded attribute vector sequence into Self-Attention, splicing an output result and prior features to obtain a spliced attribute vector sequence, and inputting the spliced attribute vector sequence into CNN and Dense layers to predict a head and tail position of a weapon equipment and obtain an intermediate relation extraction model; training any of the weapon equipment based on the intermediate relation extraction model to obtain the trained relation extraction model. the step of performing weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity data set and a corresponding weapon equipment attribute data set comprises:

2. The open source intelligence-oriented weapon equipment knowledge graph construction method according to claim 1, wherein, obtaining a semantic representation data set of the standardized military text data through a Bert pre-training language model, outputting a probability maximum entity label sequence through a CRF layer, determining a weapon equipment entity data set based on the entity label sequence, and performing weapon equipment attribute extraction operations on the standardized military text data based on the relation extraction model to obtain a weapon equipment attribute data set. ​ 3. The open source intelligence-oriented weapon equipment knowledge graph construction method of claim 1, wherein, The step of training the relationship extraction model according to any of the weapon equipment based on the intermediate relationship extraction model includes: Sampling any of the weapon equipment, inputting the coded attribute vector sequence corresponding to the sub-attribute vector sequence of any of the weapon equipment into the LSTM, and obtaining the attribute encoding vector sequence of any of the weapon equipment; Input the coded attribute vector sequence into another layer of Self-Attention, splice the output result with the attribute encoding vector sequence of any of the weapon equipment, and obtain the target attribute vector sequence; Input the target attribute vector sequence into CNN and Dense, predict the head and tail positions of the weapon equipment, and obtain the trained relationship extraction model.

4. The open source intelligence-oriented weapon equipment knowledge graph construction method according to claim 1, wherein, The step of performing entity linking on the weapon equipment entity data set and the existing entity in the weapon equipment knowledge graph based on the entity linking model to obtain a target candidate entity data set includes: According to the standardized military text data, candidate entity linking data sets are obtained by crawling Baidu Encyclopedia and Wikipedia, and the candidate entity linking data sets are obtained. Splice the candidate entity linking data set and its corresponding entity referent to obtain a spliced candidate entity linking data set, and separate the candidate entity and the entity referent with a separator. Input the spliced candidate entity linking data set into each BERT pre-training model, and input the spliced candidate entity linking data set into a two-class Dence layer after passing through a Dropout layer, splice the output result, and input it into a Dence layer to obtain a plurality of spliced candidate entity linking data sets. Input the plurality of spliced candidate entity linking data sets into Softmax, and take the candidate entity data set with the highest score as the target candidate entity data set.

5. The open source intelligence-oriented weapon equipment knowledge graph construction method according to claim 1, wherein, The method further includes: Based on the entity name of the target candidate entity data set, search in the knowledge graph database and return a first query result; If the first query result is NIL, the target candidate entity data set and the corresponding target attribute data set are stored in the knowledge graph database.

6. The open source intelligence-oriented weapon equipment knowledge graph construction method according to claim 1, wherein, The method further includes: Based on the attribute name of the target attribute data set, search in the knowledge graph database and return a second query result, if the second query result is NIL, store the attribute in the attribute database of the corresponding entity of the knowledge graph; otherwise, merge the attribute with the attribute of the corresponding entity, and store it in the attribute database of the corresponding entity of the knowledge graph.

7. An open source intelligence-oriented weapon equipment knowledge graph construction system, characterized in that, The system includes: An acquisition module for acquiring military text data, preprocessing the military text data, and obtaining standardized military text data; An annotation module for determining the military text data to be trained according to the standardized military text data, and performing annotation processing on the military text data to be trained to obtain weapon equipment entity recognition data sets, weapon equipment attribute extraction data sets, and weapon equipment entity linking data sets to be trained. The computing module is configured to input the to-be-trained weapon equipment entity recognition dataset into a corresponding model for training processing to obtain a trained entity recognition model, input the to-be-trained weapon equipment attribute extraction dataset into a corresponding model for training processing to obtain a trained relation extraction model, and input the to-be-trained weapon equipment entity linking dataset into a corresponding model for training processing to obtain a trained entity linking model; and perform weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity dataset and a corresponding weapon equipment attribute dataset. The determining module is configured to perform entity linking on the weapon equipment entity dataset and an existing entity in a weapon equipment knowledge graph based on the entity linking model to obtain a target candidate entity dataset, determine a target attribute dataset of the weapon equipment based on the target candidate entity dataset and the weapon equipment attribute dataset, store the target attribute dataset in a database, and obtain the weapon equipment knowledge graph. The step of inputting the to-be-trained weapon equipment attribute extraction dataset into a corresponding model for training processing to obtain a trained relation extraction model includes the following steps: The to-be-trained weapon equipment attribute extraction dataset is converted into an attribute vector sequence through Bert Embedding, and the attribute vector sequence is input into DGCNN for encoding to obtain an encoded attribute vector sequence. The encoded vector sequence is input into Self-Attention, the output result is spliced with prior features to obtain a spliced attribute vector sequence, the spliced attribute vector sequence is input into CNN and Dense layers, the beginning and end positions of the weapon equipment are predicted, and an intermediate relation extraction model is obtained. Any weapon equipment is input into the intermediate relation extraction model for training to obtain the trained relation extraction model.

8. An open source intelligence-oriented weapon equipment knowledge graph construction device, characterized in that, The apparatus includes: An obtaining unit is configured to obtain military text data, pre-process the military text data, and obtain standardized military text data. An annotating unit is configured to determine to-be-trained military text data based on the standardized military text data, and perform annotation processing on the to-be-trained military text data to obtain a to-be-trained weapon equipment entity recognition dataset, a to-be-trained weapon equipment attribute extraction dataset, and a to-be-trained weapon equipment entity linking dataset. The computing unit is configured to input the to-be-trained weapon equipment entity recognition dataset into a corresponding model for training processing to obtain a trained entity recognition model, input the to-be-trained weapon equipment attribute extraction dataset into a corresponding model for training processing to obtain a trained relation extraction model, and input the to-be-trained weapon equipment entity linking dataset into a corresponding model for training processing to obtain a trained entity linking model; and perform weapon equipment recognition and weapon equipment attribute extraction operations on the standardized military text data based on the entity recognition model and the relation extraction model to obtain a weapon equipment entity dataset and a corresponding weapon equipment attribute dataset. The determining unit is configured to perform entity linking on the weapon equipment entity dataset and an existing entity in a weapon equipment knowledge graph based on the entity linking model to obtain a target candidate entity dataset, determine a target attribute dataset of the weapon equipment based on the target candidate entity dataset and the weapon equipment attribute dataset, store the target attribute dataset in a database, and obtain the weapon equipment knowledge graph. The step of inputting the to-be-trained weapon equipment attribute extraction dataset into a corresponding model for training processing to obtain a trained relation extraction model includes the following steps: The to-be-trained weapon equipment attribute extraction dataset is converted into an attribute vector sequence through Bert Embedding, and the attribute vector sequence is input into DGCNN for encoding to obtain an encoded attribute vector sequence. The encoded vector sequence is input into Self-Attention, and the output result is spliced with prior features to obtain a spliced attribute vector sequence, which is input into CNN and Dense layers to predict the head and tail positions of the weapon equipment and obtain an intermediate relation extraction model. Any of the weapon equipment is input into the intermediate relation extraction model for training to obtain the trained relation extraction model.

9. A storage medium having stored thereon computer instructions, characterized in that: The computer instructions perform the steps of the open source intelligence-oriented weapon equipment knowledge graph construction method in any one of claims 1 to 6.

Images