Open source information weapon equipment map construction method and system based on joint relation extraction

Abstract

The invention discloses an open source information weapon equipment atlas construction method and system based on joint relation extraction. A joint relation extraction model oriented to the field of military weapon equipment is constructed by fusing a Chine-RoBERTa-wwm pre-training model and a global pointer network. The process of constructing the weaponry atlas through the atlas system comprises the steps of data preparation, atlas construction, concept construction, data access, knowledge extraction, knowledge fusion, atlas visualization and atlas analysis. In the weapon equipment field, the entity and relation triple data in the unstructured text can be effectively extracted, the entity relation extraction accuracy in the professional field is improved, military personnel can conveniently and rapidly construct a knowledge graph when facing a large amount of weapon equipment data, and access and query are conducted; and the efficiency of acquiring knowledge for mass data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of graph construction, and in particular to a method and system for constructing an open source information weapon equipment graph based on joint relationship extraction. Background Art

[0002] In the era of big data, how to enable computers to autonomously identify, analyze and understand massive amounts of complex data and extract effective information from them is one of the core goals of the information age. Knowledge is contained in data, and knowledge graphs are a type of knowledge description framework generated in the big data environment. They represent knowledge in the form of graph structures, normalizing and scaling massive amounts of fragmented and unrelated information. The knowledge network it constructs displays data in a structured and related manner, which speeds up the speed at which people can more accurately obtain valuable information in the information society.

[0003] Knowledge graphs not only demonstrate their unique advantages in tasks such as semantic search, intelligent question answering, and recommendation systems, but also empower various industries and fields in the development of social informatization, such as the financial services industry, smart judicial field, medical field, and military field. At present, knowledge graphs have become a hot research direction in military development. The construction of a knowledge graph system in the field of military weapons and equipment has laid a solid foundation for building an informationized army and promoting intelligent warfare, and plays an important role in military operations and training.

[0004] However, the current professional data in the field of weapons and equipment is increasing in size and from a wide range of sources. There are problems such as inconsistent data standards, redundant information, and poor correlation, which prevents it from achieving its optimal performance. The entity relationship network constructed by the knowledge graph makes it particularly good at processing the above-mentioned multi-source heterogeneous complex data. By structuring and associating data, it can enhance the data's expressiveness and modeling flexibility, making it easier for relevant personnel and machines to understand, process data, and acquire knowledge.

[0005] A core link in the process of building a knowledge graph is entity relationship extraction, which aims to extract the semantic relationship between entities in the text content, so as to organize various entities and concepts. There are two main methods for entity relationship extraction: pipeline method and joint learning method. Based on the pipeline method, the entity relationship extraction task is divided into two subtasks: entity recognition and relationship extraction, and they are executed successively. This extraction method lacks sufficient information association between entities and relationships, and there will be problems of error propagation and repeated processing of data information. The joint learning strategy can effectively solve the above problems by promoting information transmission and interaction between entities and relationships, thereby enhancing the relevance of information data and reducing the appearance of redundant information.

[0006] In summary, in today's military information and intelligent background, how to extract effective data information of weapons and equipment from massive data and establish knowledge graphs in professional fields is of great significance. The present invention provides a method and system for constructing an open source information weapons and equipment graph based on joint relationship extraction. The system can not only intuitively and three-dimensionally display the existing weapons and equipment information, and uniformly manage and analyze the information such as production units, design units, countries of origin, service objects, technical specifications, etc. related to weapons and equipment, but also flexibly integrate and associate weapons and equipment data of different types and uses, so that military personnel can quickly build knowledge graphs, access and query when facing a large amount of weapons and equipment data, and improve the efficiency of acquiring knowledge when facing massive data. Summary of the invention

[0007] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a method and system for constructing an open source information weapons and equipment graph based on joint relationship extraction, so as to improve the accuracy of entity relationship extraction in professional fields and facilitate military personnel to quickly construct a knowledge graph when faced with a large amount of weapons and equipment data.

[0008] In order to solve the above technical problems, the present invention discloses a method and system for constructing an open source information weapon equipment map based on joint relationship extraction, and the steps are as follows:

[0009] Step 1. Obtain open source text data about weapons and equipment on the Internet, and pre-process the original data of weapons and equipment to form basic data of weapons and equipment, and save it as a text file;

[0010] Step 2. Use the graph building function to create an empty weapons and equipment knowledge graph;

[0011] Step 3. Use the concept construction function to sort out the entities and relationships in the basic data of weapons and equipment, define the subject and relationship of the concept layer, realize the ontology construction of the corresponding field, and associate the empty knowledge graph of weapons and equipment;

[0012] Step 4. Import the unstructured weapons and equipment basic data text file through data access, use the knowledge extraction function, call the joint relationship extraction model to extract entity relationship triples from the imported data, form the full amount of weapons and equipment triple data, and generate the corresponding basic weapons and equipment knowledge graph;

[0013] Step 5. Use the knowledge fusion function to fuse and filter the generated weapons and equipment knowledge graph to form standardized weapons and equipment triple data, and output the standardized weapons and equipment professional field knowledge graph;

[0014] Step 6. Visualize the knowledge graph of the standardized weapons and equipment professional field, use the graph analysis function to check the generation of the knowledge graph of the weapons and equipment professional field, and conduct query and analysis.

[0015] Furthermore, in step 1, the preprocessing is to clean and manually check invalid data such as lack of subject, repeated text, and reference errors;

[0016] The creation of the new map described in step 2 mainly includes the definition of the map name and the description of the map.

[0017] Furthermore, the concept construction described in step 3 is to obtain the weapon equipment basic data, and then get the weapon equipment entity and relationship list by combing the data. The defined entities include weapon equipment, country, military organization, weapon equipment type, unit organization, date and time, structure, military action, event activity, name, and the defined relationships include country of production, production unit, design unit, structure type, commissioning, participation in activities, equipment type, assembly, service time, service, occurrence time, derivative type, ownership, and alias code. In the process of defining the entities and relationships of the weapon equipment concept layer, the subject and the object need to be associated through relationships to form a triple relationship network. After completion, the definition content is associated with the created weapon equipment knowledge graph.

[0018] Furthermore, the data access described in step 4 is configured through the graph construction function, the data source type is selected as file, the data format is selected as unstructured, the data access task title is set, and the access graph is specified as the newly created weapons and equipment knowledge graph mentioned above.

[0019] Furthermore, the joint relationship extraction model in step 4 is pre-processed by obtaining open source text corpus training data from the Internet such as Weapon Encyclopedia and China.com, and manually annotating the corpus data using the Label Studio annotation platform based on the defined entity relationships to construct a weapons and equipment data set for model training. The joint relationship extraction includes:

[0020] Step 4-1: Perform deep feature extraction on the input sentence, encode the sentence into a semantic representation vector, and generate adversarial samples;

[0021] Step 4-2: Predict potential relationships using the high-dimensional semantic vector representation of the original sentence and the introduced adversarial sample semantic representation vector;

[0022] Step 4-3: Based on the predicted set of potential relationships, use the global pointer network entity recognition to obtain all entities corresponding to each relationship, and combine it with the global corresponding module to jointly determine the correct entity relationship triples.

[0023] Furthermore, in step 4, the weapons and equipment knowledge graph automatically obtains the triple data of all weapons and equipment to form a basic weapons and equipment knowledge graph.

[0024] Furthermore, in step 5, knowledge fusion is performed on the weapons and equipment knowledge graph. The fusion methods are mainly entity merging, attribute alignment, and relationship merging. After completion, a standardized weapons and equipment professional field knowledge graph is output.

[0025] Furthermore, the graph analysis described in step 6 can visualize and check the standardized graph information finally generated, and check the relationship network of a subject through query. The inspection content includes whether the subject is extracted accurately, whether the relationship between the subject and the object conforms to the relationship definition, and perform statistical analysis of the graph information through path analysis and relationship statistics.

[0026] The present invention discloses a method and system for constructing an open source information weapon equipment map based on joint relationship extraction. Aiming at the field of military weapons and equipment, the invention provides a path method for non-technical personnel to quickly construct a map, and helps to quickly mine effective information from massive data. An end-to-end extraction framework is used to identify unstructured text data in the field of military weapons and equipment. The method is used to improve the accuracy of entity relationship extraction in a specific field, and effectively solve the problem of overlapping triples in complex text contexts in the field. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the existing technical solutions and methods, the embodiments of the present invention will be described in more detail below in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the implementation steps and technologies of the present invention. The above-mentioned and other purposes, features and advantages of the present invention will become clearer. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 It is a workflow diagram of a method and system for constructing an open source information weapon equipment map based on joint relationship extraction of the present invention;

[0029] Figure 2 It is the graph construction process and steps in the embodiment of the present invention;

[0030] Figure 3 Schematic diagram of the weapon equipment joint relationship extraction model structure in an embodiment of the present invention;

[0031] Figure 4 1 is a schematic diagram comparing the masking strategies of BERT and Chinese-RoBERTa-wwm-ext in an embodiment of the present invention;

[0032] Figure 5 It is a schematic diagram of the global pointer network entity identification structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0034] In the era of big data, knowledge is contained in data, and knowledge graphs can structure and associate massive amounts of fragmented and unrelated information data through knowledge networks, accelerating the speed at which people can more accurately obtain valuable information in the information society. The present invention is aimed at the field of military weapons and equipment, and the knowledge graph construction system constructed can facilitate relevant personnel to simplify the construction and management process of knowledge graphs, such as Figure 1 As shown in the figure, for the complex professional terms and overlapping entity relationships in the field of weapons and equipment, the joint relationship extraction model is used to increase the accuracy of triple extraction, which can effectively improve the extraction efficiency and accuracy, thereby quickly constructing a graph.

[0035] The present invention discloses a method and system for constructing an open source information weapon equipment map based on joint relationship extraction. For specific map construction steps, see Figure 2 , including the following steps:

[0036] Step 1. Obtain open source text data about weapons and equipment on the Internet, and pre-process the original data of weapons and equipment to form basic data of weapons and equipment, and save it as a text file;

[0037] Step 2. Use the graph building function to create an empty weapons and equipment knowledge graph;

[0038] Step 3. Use the concept construction function to sort out the entities and relationships in the basic data of weapons and equipment, define the subject and relationship of the concept layer, realize the ontology construction of the corresponding field, and associate the empty knowledge graph of weapons and equipment;

[0039] Step 4. Import the unstructured weapons and equipment basic data text file through data access, use the knowledge extraction function, call the joint relationship extraction model to extract entity relationship triples from the imported data, form the full amount of weapons and equipment triple data, and generate the corresponding basic weapons and equipment knowledge graph;

[0040] Step 5. Use the knowledge fusion function to fuse and filter the generated weapons and equipment knowledge graph to form standardized weapons and equipment triple data, and output the standardized weapons and equipment professional field knowledge graph;

[0041] Step 6. Visualize the knowledge graph of the standardized weapons and equipment professional field, use the graph analysis function to check the generation of the knowledge graph of the weapons and equipment professional field, and conduct query and analysis.

[0042] In this embodiment, in step 1, the preprocessing is to clean and manually check invalid data such as missing subjects, repeated texts, and reference errors;

[0043] The creation of the new graph described in step 2 mainly includes the definition of the graph name and the description of the graph, which is mainly used to create a knowledge graph base for storing the professional field of weapons and equipment.

[0044] In this embodiment, the concept construction described in step 3 is to define entities and relationships through this function after data access is performed on the basic data of weapons and equipment. Select a new entity on the concept layer definition canvas, enter the entity type name and icon, and save it. After completing the creation of at least two entities, the corresponding entities are connected and the relationship type name is defined. The defined entities include weapons and equipment, countries, military organizations, weapons and equipment types, units and institutions, date and time, structures, military operations, events and activities, and names. The defined relationships include country of production, production units, design units, structural types, commissioning, participation in activities, equipment types, assembly, service time, service, occurrence time, derivatives, attribution, and alias codes. Through concept construction, the subject and the object are associated with a relationship and saved, and at the same time associated with the weapons and equipment knowledge graph.

[0045] In this embodiment, the data access and knowledge extraction described in step 4 is a process of performing access data extraction in the form of a task. The first step of setting the task is completed by setting the access data source type to "file", the extracted data source format to "unstructured format", setting the task title to "weapons and equipment data extraction task" and the target graph to "weapons and equipment knowledge graph". The second step is to select the local path of the file, select the text file with the entity file name as the weapon and equipment professional field data to upload, select the extraction method as interface in the transmission mode, select the extraction model as the pre-trained joint relationship extraction model, and save it after completing the setting. Run the set weapon and equipment data extraction task in the data task extraction window, and the system automatically performs the data extraction work.

[0046] In this embodiment, the construction of the joint relationship extraction model described in step 4 is achieved by pre-processing the training data of open source text corpora such as Weapon Encyclopedia and China.com on the Internet, and annotating the entity relationships of the corpus data using the LabelStudio annotation platform based on the defined entity relationships. All triple sequences, entity types, and the starting and ending positions of the entities in the sentence sequence can be obtained, thereby forming a weapons and equipment relationship extraction data set that can be used for model training. The annotated military weapons and equipment data set is then divided into training set, validation set, and test set in a ratio of 8:1:1, and input into the constructed relationship extraction model. For details on the structure of the joint relationship extraction model for weapons and equipment, see Figure 3 , including the following steps:

[0047] Step 4-1 performs deep feature extraction on the input sentence, encodes the sentence into a semantic representation vector, and generates adversarial samples;

[0048] The present invention uses the Chinese-RoBERTa-wwm-ext Chinese pre-trained large model as the encoder of the weapon equipment joint relationship extraction model, and combines it with the Fast Gradient Method (FGM) adversarial training, and uses the rich semantic information learned by the pre-trained model on a large-scale corpus to map the input text sequence to a high-dimensional vector space and convert it into a high-dimensional feature vector representation as the basic input of the subsequent relationship prediction task module. In the pre-training stage, the encoder adopts a full-word masking strategy to change the masking unit from a single character to a whole word, so that the model can more effectively grasp the contextual semantic information and connections of the Chinese text content, see Figure 4 . FGM adversarial training introduces adversarial samples based on the original input samples, thereby improving the model's resistance to small perturbations in the data. The specific steps are as follows: the input sentence sequence passes through the encoder's 12-layer Transformer containing a multi-head self-attention mechanism and a feedforward neural network to obtain the corresponding high-dimensional vector representation h, and adversarial samples are generated by adding perturbations. The perturbation Δx is generated through the gradient information of the input data, and the perturbation Δx is added to the input h to generate adversarial samples h', i.e. h+Δx. These adversarial samples are used together with the original data as inputs for subsequent relationship prediction tasks to enhance the generalization and robustness of the model. The perturbation is calculated as follows:

[0049]

[0050] Where h is the high-dimensional feature representation of the original input sentence, Δx is the perturbation added to the input, ε is a preset perturbation coefficient used to control the amplitude of the perturbation, and θ is the model parameter. is the gradient of the loss function L with respect to the input h, is the L2 norm of the gradient.

[0051] Step 4-2 Potential relationship prediction

[0052] In the professional field of military weapons and equipment, the acquired text data usually contains a large number of dense and complex weapon entities and relationships. In order to efficiently and accurately extract matching relationships from complex texts, we can first predict the possible relationship set in the text sequence, and then identify the corresponding entity pairs based on the relationship set.

[0053] The prediction of potential relationships is regarded as a multi-label classification problem. Given the embedding representation h of a sentence with n tokens, an average pooling operation is performed on the embedding representation of the sentence, and then an activation function is used to convert the weighted pooled vector into the probability value of the existence of each relationship. A threshold is preset. If Pr is greater than the threshold, the relationship is considered to exist, otherwise it does not exist.

[0054] h avg =Avgpool(h)∈R (d×1)

[0055] P r =σ(W r h avg +b r )

[0056] Where h∈R (n×d) is the high-dimensional semantic feature vector of the input sentence, h avg is the vector representation after the average pooling operation. r ∈R (d×1) is a trainable weight parameter, b r is the bias parameter, σ is the Sigmoid activation function, P r is the probability of the existence of this relationship.

[0057] Step 4-3 Global pointer network entity identification and global correspondence module

[0058] The high-confidence relationship set R obtained through potential relationship prediction is used to identify entities under limited relationships in turn. In order to solve the problem of overlapping and nested entities in the field of weapons and equipment, the global pointer network annotation method is used in combination with the semantic vector representation of sentences for joint training, so as to obtain entity recognition based on specific relationship constraints. The identification of the starting position and the end position of the entity can be divided into entity extraction and entity type classification subtasks, respectively, and RoPE relative position encoding is introduced. The final entity scoring score is obtained by adding the extraction and classification scores. If the score exceeds the threshold, the corresponding label is set to 1 in the matrix, otherwise it is removed, such as Figure 5 The specific calculation steps are as follows:

[0059] q i =w q· (h i +r i )+b q

[0060] k i =w k· (h i +r i )+b k

[0061] q j =w q· (h j +r j )+b q

[0062] k j =w k· (h j +r j )+b k

[0063] In the formula, h i Represents the vector representation of the i-th token of the input sentence, r i ∈R (d×1) is the embedding representation of the i-th relation in the relation set obtained by the latent relation prediction module, h i +r i It represents the high-dimensional feature representation of the sentence that incorporates a specific relation vector, q i and k i They represent the query vector and key vector corresponding to the i-th token of the input sentence, respectively, and w q and w k is a trainable weight parameter, b q and b k is the bias parameter.

[0064]

[0065]

[0066]

[0067] In the formula, u i and u j is a relative position vector, α represents one of the starting position type and the ending position type,

[0068] and They represent the entity extraction score and entity type classification score respectively.

[0069] According to the entity recognition under the specific relationship, all the head entities and tail entities that may belong to the relationship under each potential relationship type are obtained, and then a comprehensive judgment is made with the global correspondence module to determine the correct entity pair under the relationship type, so as to realize the filtering and screening of entity relationship triples. The global correspondence matrix obtains the score matrix of the association between all subject and object entities in the sentence through the semantic representation vector obtained in the encoding layer. Each element in the matrix represents the confidence of the starting position of the corresponding subject and object. The higher the confidence value, the higher the possibility that the subject and object correspond to a triple. For example, in the corpus sentence "Wing Loong UAV is developed by AVIC.", "Wing Loong UAV" and "AVIC" are the correct subject and object entity pairs corresponding to the same relationship triple. Then, in the global correspondence matrix, the corresponding association confidence value of "Wing Loong UAV" and "AVIC" will be higher.

[0070] The global correspondence matrix M is an n×n matrix. The value of each element in the matrix is ​​the association confidence of the entity pair in the corresponding sentence. By setting the association threshold, if the value of the association confidence is greater than the association threshold, it is retained, that is, the corresponding label is set to 1 in the matrix, otherwise it is removed. The value of each element in the matrix is ​​calculated as follows:

[0071]

[0072] In the formula, is the vector representation of the jth object word in the input sentence, W g is the trainable weight matrix, σ is the sigmoid activation function, b g is the bias parameter.

[0073] The total loss during the training process is obtained by adding together the loss calculations of potential relationship prediction, global pointer network entity recognition, and global correspondence by setting different weights. By using the trained model, the unstructured text corpus in the field of weapons and equipment is jointly extracted to obtain the implied relationships.

[0074] In step 4, the triple data of all weapons and equipment are extracted through the joint relationship extraction model, and a comparison check is performed after the data extraction task is completed to confirm that the extraction of all triple data is completed.

[0075] In this embodiment, in step 5, knowledge fusion is performed on the entire amount of weapons and equipment triple data. The fusion methods are mainly entity merging, attribute alignment, and relationship merging to generate a standardized weapons and equipment knowledge graph and output weapons and equipment professional field graph information.

[0076] In step 6, the generated graph information is checked, and the relationship network of a subject is checked through query. The check content includes whether the subject is extracted accurately, whether the relationship between the subject and the object conforms to the relationship definition, and whether the graph is generated accurately through path analysis and relationship statistics.

[0077] The present invention provides a method and system for constructing an open source information weapon equipment map based on joint relationship extraction. The above description of the disclosed embodiments enables professionals and technicians in the field to implement or use the present invention. Various modifications to these embodiments will be obvious to professionals and technicians in the field, and the general principles defined in the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in the present invention, but will conform to the widest range consistent with the principles and novel features disclosed in the present invention.

Claims

1. A method for constructing an open source information weapons and equipment map based on joint relationship extraction, characterized in that: The method comprises: Step 1. Obtain open source text data about weapons and equipment on the Internet, and pre-process the original data of weapons and equipment to form basic data of weapons and equipment, and save it as a text file; Step 2. Use the graph building function to create an empty weapons and equipment knowledge graph; Step 3. Use the concept construction function to sort out the entities and relationships in the basic data of weapons and equipment, define the subject and relationship of the concept layer, realize the ontology construction of the corresponding field, and associate the empty knowledge graph of weapons and equipment; Step 4. Import the unstructured weapons and equipment basic data text file through data access, use the knowledge extraction function, call the joint relationship extraction model to extract entity relationship triples from the imported data, form the full amount of weapons and equipment triple data, and generate the corresponding basic weapons and equipment knowledge graph; Step 5. Use the knowledge fusion function to fuse and filter the generated weapons and equipment knowledge graph to form standardized weapons and equipment triple data, and output the standardized weapons and equipment professional field knowledge graph; Step 6. Visualize the knowledge graph of the standardized weapons and equipment professional field, use the graph analysis function to check the generation of the knowledge graph of the weapons and equipment professional field, and conduct query and analysis.

2. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 1 is characterized in that: In step 1, the preprocessing is to clean and manually check invalid data such as missing subjects, repeated texts, and reference errors; The creation of the new map described in step 2 mainly includes the definition of the map name and the description of the map.

3. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 2 is characterized in that: The concept construction described in step 3 is to obtain the basic data of weapons and equipment, and then obtain a list of weapons and equipment entities and relationships by combing the data. The defined entities include weapons and equipment, countries, military organizations, weapons and equipment types, units and institutions, dates and times, structures, military operations, events and activities, and names. The defined relationships include country of origin, production units, design units, structural types, commissioning, participation in activities, equipment types, assembly, service time, service, occurrence time, derivative types, ownership, and alias codes. In the process of defining entities and relationships at the conceptual layer of weapons and equipment, the subject and object need to be associated through relationships to form a triple relationship network. After completion, the definition content will be associated with the created weapons and equipment knowledge graph.

4. According to claim 3, a method and system for constructing an open source information weapon equipment map based on joint relationship extraction is characterized in that: The data access described in step 4 is configured through the graph building function, selecting the data source type and data format, setting the data access task title and specifying the access graph.

5. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 4 is characterized in that: The joint relationship extraction model in step 4 is obtained by constructing training set samples by manually annotating unstructured open source information text data obtained on the network for training. The joint relationship extraction includes: Step 4-1: Perform deep feature extraction on the input sentence, encode the sentence into a semantic representation vector, and generate adversarial samples; Step 4-2: Predict potential relationships using the high-dimensional semantic vector representation of the original sentence and the introduced adversarial sample semantic representation vector; Step 4-3: Based on the predicted set of potential relationships, use the global pointer network entity recognition to obtain all entities corresponding to each relationship, and combine it with the global corresponding module to jointly determine the correct entity relationship triples.

6. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 5 is characterized in that: In step 4, the weapons and equipment knowledge graph automatically obtains the triple data of all weapons and equipment to form the basic weapons and equipment knowledge graph.

7. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 6 is characterized in that: In step 5, the weapons and equipment knowledge graph is fused by entity merging, attribute alignment, and relationship merging. After completion, a standardized weapons and equipment professional field knowledge graph is output.

8. The method for constructing an open source information weapons and equipment map based on joint relationship extraction according to claim 7 is characterized in that: The graph analysis described in step 6 can visualize and check the standardized graph information finally generated, and check the relationship network of a subject through query. The inspection content includes whether the subject is extracted accurately, whether the relationship between the subject and the object conforms to the relationship definition, and perform statistical analysis of the graph information through path analysis and relationship statistics.

Images