Intelligence co-processing platform and potential relation prediction method based on knowledge graph

Abstract

The invention discloses an information co-processing platform based on a knowledge graph and a potential relationship prediction method. The information co-processing platform comprises a data acquisition processing layer, a knowledge extraction intelligent analysis layer and a knowledge application decision-making layer, the data acquisition processing layer comprises a data acquisition module, a data format conversion module and a knowledge base management module; the knowledge extraction intelligent analysis layer comprises a keyword extraction module, a keyword layering module, a semantic fusion module and a knowledge ontology dynamic evolution module; and the knowledge application decision-making layer comprises an evolution state judgment module, a military strategy graph generation module, an evaluation result acquisition module and a military strategy guidance information updating module. According to the method, a large amount of dispersed and isolated intelligence is integrated from actual military field business data by taking a real-time military strategy as a guide, association of military intelligence information is realized from a semantic level, powerful support is provided for comprehensive analysis, research, judgment and display of the intelligence information, analysis efficiency is improved, and effective cooperation of military strategy and intelligence data processing is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of military intelligence data processing, and in particular to an intelligence collaborative processing platform and a potential relationship prediction method based on a knowledge graph. Background Art

[0002] Relying on the progress of information technology, existing battlefield intelligence information has shown explosive growth, complex intelligence sources, and diverse representation forms. Traditional search engines and query methods cannot meet the needs of capturing and understanding battlefield information and a large amount of military knowledge.

[0003] The invention with the publication number CN115878812A discloses a method for analyzing strategic intent based on open source intelligence, including obtaining open source intelligence text and preprocessing; extracting multiple event elements, and extracting the core words of the event; extracting the event dependency subgraph containing meta-event pairs; obtaining the annotation labels corresponding to each meta-event in the meta-event pair; constructing a preliminary event graph; obtaining an optimized event graph; predicting subsequent events for events in the optimized event graph through a strategic event prediction model, and obtaining the causes of events in the optimized event graph. The method provided in this application mines information from unstructured text, integrates relevant data and resources, and expands the open source intelligence text data set, extracts event elements and event relationships in the open source intelligence text data set, analyzes the enemy's strategic intent through the event graph, realizes the transformation of traditional intelligence work to informatization and intelligence, and provides auxiliary decision-making capabilities for intelligence analysis work.

[0004] The invention with publication number CN115878811A discloses a method of intelligent analysis and deduction of military intelligence based on event-based graphs, proposes an idea of ​​intelligence analysis and deduction centered on events, and elaborates on the specific process from the automatic construction of event-based graphs to event-centered visual reasoning. The invention uses event-driven intelligence analysis technology to conduct comprehensive monitoring and visual analysis of battlefield intelligence data, which can provide managers with a visual, manageable and controllable military command and decision-making platform, helping users to explore the value of military intelligence data and improve the command and decision-making efficiency of combat personnel.

[0005] The invention with the publication number CN114896387A discloses a method, device and computer-readable storage medium for visualizing military intelligence analysis. After acquiring and preprocessing event texts in various fields, the event texts are divided into event classes according to the event ontology library and event elements of the event texts are extracted. The main content of the event texts and event class relationships are extracted according to the event elements. The event evolution path is predicted and derived through the Bayesian network classification model, and the event class relationships and event evolution path are visualized. Thus, the overall context of the intelligence event can be displayed in the form of a graph as much as possible, and the relevant impacts brought by the event are predicted through the machine learning classification model, and customized field event content is pushed to the corresponding users.

[0006] However, the aforementioned invention requires the construction of a causal map, and optimization based on the causal map to analyze the enemy's strategic intentions. It can only provide partial information as a reference for military strategy. The information provided depends on the construction of the causal map, ignoring the real-time changes of military strategy, and cannot achieve effective coordination between military strategy and intelligence data processing, making it difficult to explore potential relationships in military intelligence.

[0007] The invention with the publication number CN115408532A discloses a method for constructing a knowledge graph of weapons and equipment for open source intelligence. The method obtains military text data from open source resources, pre-processes the military text data, obtains standardized military text data, annotates the standardized military text data, obtains a training set (a dataset of weapon and equipment entity recognition to be trained, a dataset of weapon and equipment attribute extraction to be trained, and a dataset of weapon and equipment entity link to be trained), improves the model by training the model, and identifies and extracts attributes of weapons and equipment based on the improved model, and finally constructs a knowledge graph, solving the problem of low utilization rate of military open source information, time-consuming and labor-intensive query, and inconvenience caused by the scattered, uneven quality, and large amount of data of the current open source military information. The invention can only be used to construct a knowledge graph of weapons and equipment, and cannot achieve effective coordination of military strategy and intelligence data processing, nor does it involve potential relationships in military intelligence. Summary of the invention

[0008] The purpose of the present invention is to provide an intelligence collaborative processing platform and a potential relationship prediction method based on a knowledge graph. Starting from actual military business data and guided by real-time military strategies, the present invention studies military knowledge graphs for the field of military intelligence information, integrates a large amount of scattered and isolated intelligence, and realizes the association of military intelligence information from a semantic level. It provides strong support for the comprehensive analysis, judgment and display of intelligence information, improves analysis efficiency, realizes effective coordination of military strategies and intelligence data processing, and effectively predicts potential relationships in military intelligence under the real-time guidance of military strategies.

[0009] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0010] An intelligence collaborative processing platform based on knowledge graph, the intelligence collaborative processing platform includes a data acquisition processing layer, a knowledge extraction intelligent analysis layer and a knowledge application decision layer;

[0011] The data acquisition processing layer includes a data acquisition module, a data format conversion module and a knowledge base management module;

[0012] The data acquisition module acquires military intelligence business data from multiple data sources, cleans and classifies them, and stores them in corresponding databases. Military intelligence business data include various military intelligence results data analyzed and studied in recent years and intelligence information reported by departments at all levels with real-time access; the data format conversion module is used to uniformly represent the multi-source heterogeneous military intelligence business data collected by the data acquisition module; the knowledge base management module is used to generate a semantic network knowledge base based on the military intelligence business data after format conversion, dynamically expand the semantic network knowledge base, and integrate duplicate content and conflicting content in the knowledge ontology construction; the semantic network knowledge base includes four categories: people, organizations, resources, and military scenario events; a class domain knowledge ontology, each domain knowledge entity has corresponding attribute information, and each domain knowledge entity includes an active state attribute;

[0013] The knowledge extraction intelligent analysis layer includes a keyword extraction module, a keyword stratification module, a semantic fusion module and a knowledge ontology dynamic evolution module;

[0014] The keyword extraction module is used to obtain several words or phrases with high relevance to the entity as basic keywords;

[0015] The keyword stratification module constructs a core keyword set for the current period based on the evolution trend of historical military strategy related information and the military strategy guidance information for the current period; then, according to the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, the remaining basic keywords are stratified and classified to construct a spatial distribution model between keywords and entities, and between keywords;

[0016] The semantic fusion module takes the association between keywords and knowledge ontologies in various fields as the main feature, establishes the feature vector of knowledge ontologies in various fields, and calculates the similarity between knowledge ontologies in various fields, and obtains the relationship between knowledge ontologies in various fields according to the general threshold; starting from the keyword frequency and distance characteristics, it mines the hierarchical and sequential relationships between knowledge entities in various fields, constructs a knowledge ontology evolution semantic network, and performs semantic fusion on multi-source heterogeneous data in the semantic network knowledge base;

[0017] The knowledge ontology dynamic evolution module starts from the dynamic evolution of the knowledge ontologies of each field and between each other, performs upper and lower semantic reasoning on the integrated semantic network knowledge base, finds the link between the two knowledge ontologies, and then extends the link to discover the potential semantics or broader relationships on the link, and mines the evolution trend of the multi-relationships between people, organizations, resource behavior sequences and military scenario events to build a multi-knowledge semantic network; introduces the time slice method to cumulatively expand the network topology decoupling, sets a position label for each node and link edge in each time slice, indicates whether there is a state transfer of the network node and edge in the corresponding time slice, discovers and infers the knowledge group, and updates the active state attributes of each field knowledge entity;

[0018] The knowledge application decision layer includes an evolution state determination module, a military strategy map generation module, an evaluation result acquisition module and a military strategy guidance information update module;

[0019] The evolution state identification module identifies the evolution state and evolution trend of military strategy related information from the aspects of characters, organizations, resources and military scenario events according to the discovery and reasoning results of the knowledge group, and updates the active state attributes of each domain knowledge entity at the same time; the military strategy map generation module selects several key focus characters, key focus organizations, key focus resources and key focus events of the current period according to the updated active state attributes to generate a military strategy map; the evaluation result collection module collects the evaluation results of military experts on military strategy related information; the military strategy guidance information update module dynamically updates the military strategy guidance information of the next period in combination with the evolution state and evaluation results of the military strategy related information.

[0020] Furthermore, the basic structure of the data in the semantic network knowledge base is: SKB = (C, E, A, V), where C = {c1, c2, ..., c n} represents a category set, E = {e1, e2, ..., e m} represents an entity set, A = {a1, a2, ..., a k} represents a set of entity attributes, where a i ∈A represents the basic attributes, Indicates the value corresponding to the attribute; in the category set {C}, there is a parent-child relationship between categories, that is, c j ∈c S , indicating category c j It is category c S A subclass of .

[0021] Furthermore, the knowledge base management module includes an entity construction component, a semantic network knowledge base generation component and a conflict fusion component;

[0022] The entity construction component is used to study the characteristics of some unstructured texts, manually construct an initial military domain knowledge entity attribute library, and then expand the military domain knowledge entity attribute library based on word similarity, and at the same time, use a dictionary method combined with part-of-speech attribute discrimination to annotate entities and attributes of unstructured texts;

[0023] The semantic network knowledge base generation component uses the keyword spatial position co-occurrence feature as a training set, uses a deep learning model to train the annotated entity and attribute data, uses the trained deep learning model to extract new domain knowledge data, evaluates the effectiveness of the extracted entities and attributes, dynamically expands the evaluated effective entities and attributes to the military domain knowledge entity attribute library, and then identifies the domain knowledge entities and attributes in the massive text based on the reinforcement machine learning method;

[0024] The conflict fusion component analyzes term conflicts, semantic conflicts and predicate conflicts, and adopts a hybrid method of logic tree fusion, frequency fusion and syntax fusion to select and fuse duplicate content and conflicting content in knowledge ontology construction.

[0025] Further, the conflict fusion component includes a terminology library, a predicate library, an ontology library, a terminology conflict processing unit, a semantic conflict processing unit, a predicate conflict processing unit and a fusion unit;

[0026] The terminology library and predicate library respectively extract relevant knowledge ontology from the military domain knowledge entity attribute library, respectively construct terminology set, predicate set and semantic set and transmit them to the ontology library; the ontology library analyzes the terminology set, predicate set and semantic set to obtain the existing terminology conflict, predicate conflict and semantic conflict; the terminology conflict processing unit calls the logic tree fusion model, the frequency fusion model and the syntax fusion model to process the terminology conflict, the semantic conflict processing unit calls the syntax fusion model to process the semantic conflict, and the predicate conflict processing unit calls the frequency fusion model and the syntax fusion model to process the predicate; the fusion unit integrates the processing results of the terminology conflict processing unit, the semantic conflict processing unit and the predicate conflict processing unit to generate a fused knowledge item;

[0027] The logic tree fusion model fuses conflicting items by determining the logic tree relationship between the two conflicting items; the frequency fusion model uses high-frequency co-occurring knowledge items as the highest score output; the syntactic fusion model uses the syntactic dependency tree discriminant to determine the semantic space dependency relationship between entities, attribute words and attribute value words, and makes choices among the conflicting items.

[0028] Furthermore, the keyword extraction module obtains a word set D-terms by segmenting the document set D of each domain knowledge entity content, maps the document set D to a graph G, where vertices are words and edges represent the co-occurrence frequency of two words, and then calculates the network node centrality between nodes, sorts the network node centrality of each word, merges some words into phrases through the n-gram algorithm, and outputs several results with the highest ranking as basic keywords related to the domain knowledge entity.

[0029] Furthermore, the keyword stratification module selects entities that are active in the current period based on the evolution trend of historical military strategy-related information, and selects a first keyword related to the entity from the basic keywords for each entity-keyword association relationship; obtains and analyzes the military strategy guidance information of the current period, and selects a second keyword related to the entity from the basic keywords according to the entity corresponding to the military strategy guidance information; and combines the first keyword and the second keyword to form a core keyword set for the current period.

[0030] Furthermore, the semantic fusion module uses an entity similarity calculation model and a category network fusion model to fuse the semantic knowledge network;

[0031] The entity similarity calculation model uses the following formula to merge entities between the semantic network knowledge bases SKB1 and SKB2:

[0032] f:(C i , E i , A i , V i )×(C j , E j , A j , V j ) → (C, E, A, V)

[0033] If the entity names in the two knowledge bases are the same, it is considered that the two knowledge bases describe the same content; if the entity name in SKB1 is i The same entity name was not found in SKB2 j , but find e with the same domain knowledge description or a similarity greater than the first preset similarity threshold j , then the entities in the two knowledge bases are considered to be the same; if the entity name e in SKB1 i In SKB2, only e with similarity less than the first preset similarity threshold can be found. j , but the similarity of other attributes and attribute values ​​is greater than the second preset similarity threshold, then the entities in the two knowledge bases are considered to be the same; when the two entities are judged to be the same, let e = e i Or e=e j , C=(Ci ∪C j ), A=(A i ∪A j ),in

[0034] When two entities are judged to be the same or two semantic network category nodes are merged, the category network fusion model is used to execute C=(C i ∪C j ); When the entity has only one parent class, determine whether the parent classes between the two knowledge bases are the same:

[0035] If c i With c j If the category words are the same, then the parent categories are the same;

[0036] If c i With c j If the category words are inconsistent, but the domain knowledge descriptions are the same or the similarity is greater than the third preset threshold, the categories in the two knowledge bases are considered to be the same; if the category words and domain content descriptions are inconsistent, then it is judged that the similarity of other attributes of the category and the attribute values ​​is greater than the fourth preset similarity threshold, then the categories in the two knowledge bases are considered to be the same; when an entity has multiple parent classes in different knowledge bases, and Then C i ∩C j As the common parent class of the entity, other parent classes call the deep inheritance algorithm and the fusion algorithm.

[0037] Furthermore, the knowledge ontology dynamic evolution module includes a unit knowledge reasoning component, a multi-relation reasoning component and a dynamic knowledge reasoning component;

[0038] The unit knowledge reasoning component loads the knowledge network data and establishes a data structure of neighbor nodes and label information corresponding to each node; for each node in the knowledge network, according to the label attributes of the node and its neighbor nodes, and the specified label information ratio adjustable parameter p, the probability value of the neighbor node of the node being wandered to is calculated, and a number of random selections are made from the neighbor nodes of the node, and the probability of each neighbor node being selected meets the calculated probability value; then, according to the obtained probability value and wandering parameters, wandering is started to obtain several wandering paths; according to the wandering path, the Word2vec method is called for training to obtain word vectors, and multi-label classification tasks are performed on the knowledge network nodes;

[0039] The multivariate relational reasoning component constructs possible candidate LE patterns based on the combination of all edge types on the knowledge network; traverses the entire graph to find all corresponding instances of different candidate LE patterns, calculates the weights and values ​​of different candidate LE patterns, and solves the credible LE pattern through a greedy algorithm; uses the selected credible LE pattern to match in the knowledge network to obtain the reasoning result;

[0040] The dynamic knowledge reasoning component introduces time slices to cumulatively expand the decoupling of network topology. By setting a position label for each node and linked edge in each time slice, it indicates whether there is a state transfer of network nodes and edges in the corresponding time slice, and discovers and infers knowledge groups.

[0041] The present invention also discloses a potential relationship prediction method based on knowledge graph, and the potential relationship prediction method is executed based on the intelligence collaborative processing platform based on knowledge graph as described above;

[0042] The potential relationship prediction method comprises the following steps:

[0043] Obtain military intelligence business data from multiple data sources, clean and classify them, and store them in the corresponding database. Military intelligence business data includes various military intelligence results data analyzed and studied in recent years and intelligence information reported by departments at all levels in real time;

[0044] The multi-source heterogeneous military intelligence business data collected by the data collection module are uniformly represented; a semantic network knowledge base is generated based on the military intelligence business data after format conversion, and the semantic network knowledge base is dynamically expanded, while the duplicate and conflicting contents in the knowledge ontology construction are integrated; the semantic network knowledge base includes four aspects: people, organizations, resources, and military scenario events; in the domain knowledge ontology, each domain knowledge entity has corresponding attribute information, and each domain knowledge entity includes an active state attribute;

[0045] Acquire several words or phrases that are highly relevant to the entity as basic keywords;

[0046] Based on the evolution trend of historical military strategy-related information and the military strategy guidance information of the current period, a core keyword set of the current period is constructed; then, according to the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, the remaining basic keywords are layered and classified, and a spatial distribution model between keywords and entities, and between keywords and keywords is constructed;

[0047] The relationship between keywords and knowledge ontologies in various fields is used as the main feature to establish the feature vector of knowledge ontologies in various fields, and the similarity between knowledge ontologies in various fields is calculated. The relationship between knowledge ontologies in various fields is obtained according to the common threshold. Starting from the keyword frequency and distance characteristics, the hierarchical and sequential relationships between knowledge entities in various fields are mined, and a knowledge ontology evolution semantic network is constructed to semantically fuse multi-source heterogeneous data in the semantic network knowledge base.

[0048] Starting from the dynamic evolution of knowledge ontologies in various fields and between each other, the upper and lower layers of semantic reasoning are performed on the integrated semantic network knowledge base. After finding the link between two knowledge ontologies, the link is extended to discover the potential semantics or broader relationships on the link, and the evolution trend of the multi-relationships between people, organizations, resource behavior sequences and military scenario events is mined to construct a multi-knowledge semantic network; the time slice method is introduced to cumulatively expand the network topology decoupling, and position labels are set for each node and link edge in each time slice to indicate whether there is a state transfer of network nodes and edges in the corresponding time slice, so as to discover and reason about knowledge groups and update the active state attributes of each domain knowledge entity;

[0049] Based on the findings and reasoning results of the knowledge group, the evolution status and evolution trend of military strategy-related information are judged from the aspects of people, organizations, resources and military scenario events, and the active state attributes of each domain knowledge entity are updated; according to the updated active state attributes, several key focus people, key focus organizations, key focus resources and key focus events in the current period are selected to generate a military strategy map;

[0050] Collect the evaluation results of military experts on military strategy-related information; dynamically update the military strategy guidance information for the next cycle based on the evolution status of military strategy-related information and the evaluation results.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] First, the knowledge graph-based intelligence collaborative processing platform and potential relationship prediction method of the present invention start from actual military business data, take real-time military strategy as the guide, study the military knowledge graph for the field of military intelligence information, integrate a large amount of scattered and isolated intelligence, and realize the association of military intelligence information from the semantic level, so as to provide strong support for the comprehensive analysis, judgment and display of intelligence information, improve analysis efficiency, realize the effective coordination of military strategy and intelligence data processing, and effectively predict the potential relationships in military intelligence under the real-time guidance of military strategy.

[0053] Second, the knowledge graph-based intelligence collaborative processing platform and potential relationship prediction method of the present invention can take military strategy as the center, and for data from different sources and in different formats, realize the correlation analysis and convergence collaboration of military intelligence field data oriented to military strategy. Through the visualization processing of military strategy-related information, it provides an effective basis for military field personnel to adjust their strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the structure of the intelligence collaborative processing platform based on knowledge graph of the present invention;

[0055] Figure 2 Schematic diagram for entity recognition and attribute extraction;

[0056] Figure 3 Provides conflict resolution schematics;

[0057] Figure 4 Schematic diagram of the spatial distribution model structure between keywords and entities, and between keywords. DETAILED DESCRIPTION

[0058] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings.

[0059] See also Figure 1 , this embodiment discloses an intelligence collaborative processing platform based on knowledge graph, the intelligence collaborative processing platform includes a data acquisition processing layer, a knowledge extraction intelligent analysis layer and a knowledge application decision layer;

[0060] The data acquisition processing layer includes a data acquisition module, a data format conversion module and a knowledge base management module;

[0061] The data acquisition module acquires military intelligence business data from multiple data sources, cleans and classifies them, and stores them in corresponding databases. Military intelligence business data include various military intelligence results data analyzed and studied in recent years and intelligence information reported by departments at all levels with real-time access; the data format conversion module is used to uniformly represent the multi-source heterogeneous military intelligence business data collected by the data acquisition module; the knowledge base management module is used to generate a semantic network knowledge base based on the military intelligence business data after format conversion, dynamically expand the semantic network knowledge base, and integrate duplicate content and conflicting content in the knowledge ontology construction; the semantic network knowledge base includes four categories: people, organizations, resources, and military scenario events; a class domain knowledge ontology, each domain knowledge entity has corresponding attribute information, and each domain knowledge entity includes an active state attribute;

[0062] The knowledge extraction intelligent analysis layer includes a keyword extraction module, a keyword stratification module, a semantic fusion module and a knowledge ontology dynamic evolution module;

[0063] The keyword extraction module is used to obtain several words or phrases with high relevance to the entity as basic keywords;

[0064] The keyword stratification module constructs a core keyword set for the current period based on the evolution trend of historical military strategy related information and the military strategy guidance information for the current period; then, according to the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, the remaining basic keywords are stratified and classified to construct a spatial distribution model between keywords and entities, and between keywords;

[0065] The semantic fusion module takes the association between keywords and knowledge ontologies in various fields as the main feature, establishes the feature vector of knowledge ontologies in various fields, and calculates the similarity between knowledge ontologies in various fields, and obtains the relationship between knowledge ontologies in various fields according to the general threshold; starting from the keyword frequency and distance characteristics, it mines the hierarchical and sequential relationships between knowledge entities in various fields, constructs a knowledge ontology evolution semantic network, and performs semantic fusion on multi-source heterogeneous data in the semantic network knowledge base;

[0066] The knowledge ontology dynamic evolution module starts from the dynamic evolution of the knowledge ontologies of each field and between each other, performs upper and lower semantic reasoning on the integrated semantic network knowledge base, finds the link between the two knowledge ontologies, and then extends the link to discover the potential semantics or broader relationships on the link, and mines the evolution trend of the multi-relationships between people, organizations, resource behavior sequences and military scenario events to build a multi-knowledge semantic network; introduces the time slice method to cumulatively expand the network topology decoupling, sets a position label for each node and link edge in each time slice, indicates whether there is a state transfer of the network node and edge in the corresponding time slice, discovers and infers the knowledge group, and updates the active state attributes of each field knowledge entity;

[0067] The knowledge application decision layer includes an evolution state determination module, a military strategy map generation module, an evaluation result acquisition module and a military strategy guidance information update module;

[0068] The evolution state identification module identifies the evolution state and evolution trend of military strategy related information from the aspects of characters, organizations, resources and military scenario events according to the discovery and reasoning results of the knowledge group, and updates the active state attributes of each domain knowledge entity at the same time; the military strategy map generation module selects several key focus characters, key focus organizations, key focus resources and key focus events of the current period according to the updated active state attributes to generate a military strategy map; the evaluation result collection module collects the evaluation results of military experts on military strategy related information; the military strategy guidance information update module dynamically updates the military strategy guidance information of the next period in combination with the evolution state and evaluation results of the military strategy related information.

[0069] 1. Data Collection and Processing Layer

[0070] The data collection and processing layer is used to collect, preprocess and analyze the basic data of multi-source heterogeneous data. The basic data sources mainly include the massive military intelligence business data accessed in real time, various intelligence results analyzed and studied in recent years, and intelligence information reported by troops at all levels. On this basis, the basic theory of domain knowledge ontology model is studied, various relationships and relationship definitions are constructed, and various data resources are organized, managed and stored around the domain knowledge ontology model. The data collection and processing layer includes data collection module, data format conversion module and knowledge base management module.

[0071] In order to collect data effectively, the present invention provides the construction and display functions of the basic database. The basic database supports batch entry of existing basic data, integration of basic data and trend intelligence data on the same platform, uploading, storage, and classification management functions of basic data, display and viewing functions of basic data messages, and full-text retrieval of basic database data, etc.

[0072] In order to solve the problem of inconsistent knowledge description in the fusion of multi-source semantic network knowledge base, the present invention adopts a unified multi-source semantic knowledge base fusion framework. The fused semantic network knowledge base is mainly composed of four parts: knowledge ontology / entity (entity), category (category), attribute (attribute) and attribute value (attributevalue). The entity as the leaf node of the entire knowledge base semantic network is an instance (instance). The category hierarchy system consists of categories, sub (parent) categories, and entities under categories. Entities or categories can belong to multiple categories or parent categories respectively. As shown in Table 1:

[0073] Table 1 Schematic diagram of the time series and spatial content model of knowledge ontology

[0074]

[0075]

[0076] An entity is the smallest unit of any knowledge base. It is the name of a thing and can be classified into categories. It is a specific instance of a category.

[0077] A category will also be given a simple document explanation in the semantic web knowledge base, and the parent category, subcategory, and entity of the category will be given.

[0078] Attributes are used to describe entities and categories. For each entity or category, the attribute is a term, category, or domain knowledge description obtained from the semantic network knowledge base.

[0079] Attribute value is the specific description of the attribute in the entity.

[0080] Based on the above four tuples, this project gives the basic structure of the semantic knowledge base (SKB):

[0081] SKB=(C,E,A,V)

[0082] Where C = {c1, c2, ..., c n} represents a category set, E = {e1, e2, ..., e m} represents an entity set, A = {a1, a2, ..., a k} represents a set of entity attributes, where a i ∈A represents the basic attributes, Indicates the value corresponding to the attribute. In the category set {C}, there is a parent-child relationship between categories, that is, c j ∈c S , indicating category c j It is category c S Subcategory of (Chile-Category).

[0083] Entity recognition is to classify entities in text into a certain semantic type. The present invention mainly adopts the following three methods for entity recognition: dictionary-based, statistical-based and rule-based methods. The dictionary-based method mainly searches for named entities in the vocabulary through string matching. The rule-based method mainly adds lexical rules, grammatical rules, and semantic rules to the entity recognition process, and then recognizes various types of named entities through rule matching. The statistical-based method uses manual annotation or original corpus for training. First, a language model is established, and then the model parameters are estimated on the training data, which is conducive to transplantation to different languages ​​and new fields. The statistical-based method mainly uses some statistical models such as hidden Markov models, maximum entropy models, support vector machines, conditional random fields (Conditional Random Fields, CRFs), deep learning, etc. The task of attribute extraction is to construct an attribute table for each entity semantic class and extract attribute values. The main method of attribute extraction is pattern matching. See the schematic diagram of entity recognition and attribute extraction for details. Figure 2 In the present invention, the semantic network knowledge base includes four types of domain knowledge ontologies: people, organizations, resources, and military scenario events. Each domain class is further divided into subclasses. In practical applications, the subclasses need to be further divided. The four types of domain knowledge ontologies can be used to display the four types of information in the final military strategy map. The domains are closely related to each other, and at the same time, they have positive significance for the selection of core keywords.

[0084] The knowledge base management module includes entity construction component, semantic network knowledge base generation component and conflict fusion component;

[0085] The entity construction component is used to study the characteristics of some unstructured texts, manually construct the initial military domain knowledge entity attribute library, and then use the deep learning LSTM sequence mining algorithm to establish word similarity to expand the military domain knowledge entity attribute library. At the same time, the unstructured text that needs to be processed is annotated with entities and attributes using a dictionary method combined with part-of-speech attribute identification.

[0086] In terms of feature extraction of entities and attributes, the semantic network knowledge base generation component uses the spatial co-occurrence features of keywords as training sets, uses deep learning models to train annotated entity and attribute data, selects effective parameter models to extract new domain knowledge data, evaluates the effectiveness of extracted entities and attributes, and dynamically expands the evaluated effective entities and attributes to the military domain knowledge entity attribute library. On this basis, the corpus is continuously increased, the domain knowledge entity attribute library is continuously and dynamically expanded, and then the domain knowledge entities and attributes in massive texts are identified using reinforcement machine learning methods.

[0087] In the process of constructing entity attribute sets, since the credibility and expression of each knowledge item corpus may conflict, the conflict fusion component analyzes term conflicts, semantic conflicts and predicate conflicts, and adopts a hybrid method of logic tree fusion, frequency fusion and syntax fusion to select and fuse the repeated and conflicting contents in the knowledge ontology construction. Figure 3 As shown, the conflict fusion component includes a term base, a predicate base, an ontology base, a term conflict processing unit, a semantic conflict processing unit, a predicate conflict processing unit and a fusion unit.

[0088] The terminology library and predicate library extract relevant knowledge ontology from the military knowledge entity attribute library, respectively, and construct term sets, predicate sets and semantic sets respectively and transfer them to the ontology library; the ontology library analyzes the term sets, predicate sets and semantic sets to obtain the existing term conflicts, predicate conflicts and semantic conflicts; the term conflict processing unit calls the logic tree fusion model, frequency fusion model and syntax fusion model to process the term conflicts, the semantic conflict processing unit calls the syntax fusion model to process the semantic conflicts, and the predicate conflict processing unit calls the frequency fusion model and syntax fusion model to process the predicates; the fusion unit integrates the processing results of the term conflict processing unit, the semantic conflict processing unit and the predicate conflict processing unit to generate fused knowledge items. The logic tree fusion model fuses the conflicting items by judging the logic tree relationship between the two conflicting items; for example, for the attribute attribute1 under the entity Entity0 extracted between the knowledge items, two value1 and value2 are extracted. Without considering the time state migration between value1 and value2, the term logic conflict between value1 and value2 occurs. This project first judges the logic tree relationship between value1 and value2. If value1 contains value2, value1 is treated as the secondary attribute attribute12 of attribute attribute1 of entity Entity0; otherwise, the same treatment is applied. The frequency fusion model uses high-frequency co-occurring knowledge items as the highest score output; the syntactic fusion model uses the syntactic dependency tree discriminant to discriminate the semantic space dependency relationship between entities, attribute words, and attribute value words, and selects and discards multiple conflicting items to achieve knowledge fusion and distinguish the true from the false.

[0089] 2. Knowledge Extraction and Intelligent Analysis Layer

[0090] The knowledge extraction intelligent analysis layer provides the system with a target-centric multi-mode data processing method and capability. Combined with the system's data storage model, it provides offline computing, real-time computing, and graph computing frameworks for the system's various data analysis application needs; it implements deep learning-based knowledge ontology and event extraction, network structure evolution, event association analysis and other processing services, builds a domain knowledge ontology library, and multi-domain knowledge ontology network relationships; it realizes the discovery of behavior patterns of domain knowledge ontology, builds relationships between multi-domain knowledge ontologies, and realizes reasoning applications in the dynamic evolution of domain knowledge ontologies. The knowledge extraction intelligent analysis layer includes a keyword extraction module, a keyword stratification module, a semantic fusion module, and a knowledge ontology dynamic evolution module.

[0091] (2.1) Keyword extraction module

[0092] The keyword extraction module is mainly used to obtain several words or phrases with high relevance to the entity. First, the word set D-terms is obtained by segmenting the document set D of each domain knowledge entity content. The document D is mapped to a graph G, where the vertices are words and the edges represent the co-occurrence frequency of two words. The BC (network node centrality) value between the nodes is then calculated, and then the BC value of each word is sorted, and some words are merged into phrases through the n-gram algorithm. Finally, the highest ranked results are output as keywords related to the domain knowledge entity.

[0093] (2.2) Keyword stratification module

[0094] In order to further display the hierarchical relationship of the extracted keywords, according to the regularity of keyword appearance in the military domain knowledge entity document set, the present invention divides the keywords into core keywords, important keywords and general keywords, and then constructs the relationship between domain knowledge entities and keywords. The spatial distribution model between keywords and entities, and keywords and keywords is as follows: Figure 4 shown.

[0095] In the present invention, the guidance of military strategy on the convergence of intelligence data is mainly achieved through the selection of core keywords. Specifically, the keyword stratification module selects the first keyword from the basic keywords based on the evolution trend of historical military strategy-related information and the correlation between keywords; obtains and analyzes the military strategy guidance information of the current period, and selects the second keyword from the basic keywords according to the military strategy guidance information; and combines the first keyword and the second keyword to form a core keyword set for the current period. It should be noted that the number of the first keyword and the second keyword selected here is usually multiple.

[0096] As a common keyword, the first keyword is greatly affected by the evolution trend and real-time evolution status of historical military strategy related information, is not subjectively affected by military experts, and has higher objectivity. In order to ensure that military personnel can be provided with as complete intelligence information as possible, after calibrating the first keyword, LSTM deep sequence neural network training can also be used to obtain a wider set of core keywords. As a characteristic keyword, the second keyword has a better correlation with the military strategy guidance information of the current cycle, so that the subsequent knowledge reasoning process can be more targeted at real-time military strategy changes and realize the synergistic relationship between military strategy and data processing. As a preferred example, the selection of the first keyword can be achieved by setting keyword selection rules or introducing machine learning algorithms, first analyzing the evolution trend of historical military strategy related information, screening important figures, organizations, resources or military scenario events, and then combining the correlation between the screened important figures, organizations, resources or military scenario events and keywords to select the first keyword. As for the selection of the second keyword, the viewing tendency data of military personnel for military strategy related information, the evaluation results of military personnel for military strategy related information, etc. can be collected, and the keywords related to the tendency data or evaluation results are analyzed, and some second keywords are selected or eliminated by adjusting the priority. Therefore, the core keyword obtained by combining the first keyword and the second keyword is not only objective and extensive, but also real-time and targeted under strategy guidance, so that the subsequent knowledge reasoning process can also smoothly coordinate with the changing process of military strategy.

[0097] Based on the determination of core keywords, the keywords extracted by the keyword extraction module are divided into two categories: important and general keywords related to domain knowledge through the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, and a spatial distribution model between keywords and entities, and between keywords is constructed.

[0098] (2.3) Semantic Fusion Module

[0099] The semantic fusion module includes a domain knowledge entity association relationship extraction component, a causal relationship discrimination component, and a semantic fusion component for multi-source heterogeneous data.

[0100] The domain knowledge entity association relationship extraction component takes the association relationship between keywords and domain knowledge entities as the main feature, establishes the feature vector of domain knowledge entities, calculates the similarity between domain knowledge entities, and obtains the relationship between domain knowledge entities based on the common threshold. Considering the sparsity of domain knowledge entities, the hierarchical training strategy HAM (hierarchical abstract machines) is adopted to obtain models with different precisions. The top-level model is used to eliminate a large amount of irrelevant data, and the bottom-level model uses a multi-classification method to accurately identify the association relationship between domain knowledge entities.

[0101] The causal relationship identification component is used to identify the causal relationship between domain knowledge entities. Specifically, the identification of the preceding relationship between domain knowledge entities that have established an association relationship can be treated as a two-class classification problem, that is, it is converted into a feature selection and classification algorithm selection problem. The causal relationship identification component mainly starts from the keyword frequency and distance features. Considering that there may be a correlation between the above feature vectors, for example, SVM can be used to identify the preceding relationship between domain knowledge entities.

[0102] Under the premise of giving the definitions of entities and categories, the entity similarity calculation model and category network fusion model, the semantic fusion components of multi-source heterogeneous data, fuse multiple semantic knowledge networks.

[0103] The entity similarity calculation model uses the following formula to merge entities between the semantic network knowledge bases SKB1 and SKB2:

[0104] f:(C i , E i , A i , V i )×(C j , E j , A j , V j ) → (C, E, A, V)

[0105] If the entity names in the two knowledge bases are the same, it is considered that the two knowledge bases describe the same content; if the entity name in SKB1 is i The same entity name was not found in SKB2 j , but find e with the same domain knowledge description or a similarity greater than the first preset similarity threshold j , then the entities in the two knowledge bases are considered to be the same; if the entity name e in SKB1 i In SKB2, only e with similarity less than the first preset similarity threshold can be found. j, but the similarity of other attributes and attribute values ​​is greater than the second preset similarity threshold, then the entities in the two knowledge bases are considered to be the same; when the two entities are judged to be the same, let e = e i Or e=e j , C=(C i ∪C j ), A=(A i ∪A j ),in

[0106] When two entities are judged to be the same or two semantic network category nodes are merged, the category network fusion model is used to execute C=(C i ∪C j ); When the entity has only one parent class, determine whether the parent classes between the two knowledge bases are the same:

[0107] If c i With c j If the category words are the same, then the parent categories are the same;

[0108] If c i With c j If the category words are inconsistent, but the domain knowledge descriptions are the same or the similarity is greater than the third preset threshold, the categories in the two knowledge bases are considered to be the same; if the category words and domain content descriptions are inconsistent, then it is judged that the similarity of other attributes of the category and the attribute values ​​is greater than the fourth preset similarity threshold, then the categories in the two knowledge bases are considered to be the same; when an entity has multiple parent classes in different knowledge bases, and Then C i ∩C j As the common parent class of the entity, other parent classes call the deep inheritance algorithm and the fusion algorithm.

[0109] (2.4) Knowledge ontology dynamic evolution module

[0110] The reasoning technology of semantic network knowledge, starting from the self-generation and dynamic evolution of knowledge ontology, can be divided into the reasoning of upper and lower semantics of unit knowledge nodes, the relationship reasoning of multiple knowledge nodes and the reasoning of knowledge changes in the dynamic evolution of the entire network.

[0111] In terms of unit knowledge reasoning, the present invention improves the existing random walk algorithm to construct a semantic network reasoning algorithm for military intelligence content. Assuming that there are N nodes in the knowledge network data to be processed, the current node in the walk process is C, and a node is selected from the neighbor nodes of C as the next node in the walk path. Assuming that the C node has E neighbor nodes, it can be expressed as:

[0112] neighbors(C)={n1,n2,n3,…,n E},0≤E <N

[0113] At the same time, the neighbor nodes in neighbors(C) that have the same label as node C are represented as:

[0114] common(C)={m1,m2,m3,…,m k},0≤k≤E

[0115] Obviously, common(C) belongs to the subset of neighbors(C). Suppose node D is selected as the next node of node C to be walked, where D belongs to the set of neighbors(C). The probability that node D belongs to the set of common(C) satisfies

[0116] P(D∈common(C))=p,D∈neighbors(C)

[0117] Among them, probability p is the walking parameter set before the node walk begins. Calculate the new variable of node C:

[0118]

[0119] Assign a probability of being visited to each neighbor node of node C. Pass this set of probability values ​​to AliasMethod. The overall calculation process of unit knowledge reasoning is given below:

[0120] Step 1: Load the knowledge network data and build the data structure of neighbor nodes and label information corresponding to each node;

[0121] Step 2: For each node in the knowledge network, based on the label attributes of the node and its neighboring nodes, and the specified label information ratio adjustable parameter p, calculate the probability value of the neighboring node of the node being wandered to, and use AliasMethod to implement random selection from the neighboring nodes of the node several times. The probability of each neighboring node being selected meets the calculated probability value.

[0122] Step 3: Based on the probability value obtained in the previous step and other walking parameters, such as walking length and the number of times walking from each node, start walking and obtain several walking paths;

[0123] Step 4: According to the walking path, call the Word2vec method for training to obtain the word vector;

[0124] Step 5: Perform multi-label classification tasks on the knowledge network nodes to test the classification effect of the algorithm.

[0125] Multivariate relational reasoning can be specifically described as performing link extension (LE) after finding the link between two knowledge ontologies to discover the potential semantics or broader relationships on the link. The present invention uses a greedy algorithm in reinforcement learning to achieve fast convergence reasoning in multivariate relations. The specific steps are as follows:

[0126] 1) Construct possible LE models. For all edge types on the knowledge network GT / S, any combination of three types of edges can constitute a candidate LE model.

[0127] 2) Reinforcement learning is used to implement mode selection. The entire graph is traversed to find all instances corresponding to different modes, the weights and values ​​of different modes are calculated, and a credible LE mode is solved through a greedy algorithm.

[0128] 3) Use the selected pattern to match in the network to obtain the inference result.

[0129] In terms of the reasoning of knowledge changes in the dynamic evolution of the entire network, the high-speed change of knowledge must be solved for knowledge discovery in the field of military intelligence. The present invention uses the Streaming stream processing framework of the Spark platform to realize dynamic reasoning and knowledge discovery for knowledge networks. In order to reduce the algorithmic pressure of sparse data streams on the platform to increase the platform, the network topology decoupling is cumulatively expanded by introducing time slices. Position labels are set in each time slice by the edge of each node and link to indicate whether the network nodes and edges exist or the state is transferred in the corresponding time slice, so as to realize the discovery and reasoning of knowledge groups. The knowledge ontology dynamic evolution module also updates the active state attributes of each domain knowledge entity. The update rules of the active state of different domain knowledge entities can be formulated in a customized manner. For example, for some weapons, as long as they are still under research or in existence, they will always be in an active state. For some military scenario events such as conferences, they can be adjusted to an inactive state at the end of the meeting. Preferably, different domain knowledge entities have different active values, which are used to divide key focus events and non-key focus events. For example, the activity value can be adjusted based on the evaluation results of military scenario events by military experts. The activity value of some resource entities is related to the attribute value. The activity value of some people or organizations is related to the complexity of their association relationships or the activity value of associated nodes, and so on.

[0130] The present invention constructs a knowledge graph based on the extracted entities, attributes, relationships and other knowledge. For example, a semantic network based on a graph data structure composed of nodes (Point) ("entities") and edges (Edge) ("relationships") is constructed, and a relationship network is obtained by connecting all different types of information (people, organizations, weapons, countries, etc.), which can analyze and solve problems from the perspective of "relationships". In order to facilitate the operation of staff, the knowledge extraction intelligent analysis layer also provides some plug-ins, which are used to support the storage, query and front-end display functions of the knowledge graph, support the automatic update of the knowledge graph, and support the manual editing function of the knowledge graph.

[0131] 3. Knowledge Application Decision-making Layer

[0132] The knowledge application decision-making layer provides business support for the application of knowledge in the field of military intelligence. On the one hand, it provides military personnel with business applications related to knowledge in the field of military intelligence, such as providing business service capabilities such as military strategy map display, intelligent search, ontology target profiling, relationship network analysis, time series analysis, knowledge recommendation, etc., providing historical event display functions, and providing data services and computing service support for upper-level applications, etc. On the other hand, the knowledge application decision-making layer also needs to judge the evolution status and evolution trend of military strategy-related information, collect the real-time military strategy intentions of military personnel, and dynamically update the next cycle of military strategy guidance information based on the evolution status and evaluation results of military strategy-related information. Use military strategy guidance information to automatically optimize core keywords and guide the knowledge extraction intelligent analysis layer to optimize the knowledge reasoning process so that the knowledge reasoning results are more in line with real-time military strategy needs.

[0133] The knowledge application decision layer includes an evolution state discrimination module, a military strategy map generation module, an evaluation result collection module, and a military strategy guidance information update module. The evolution state discrimination module discriminates the evolution state and evolution trend of military strategy related information from the aspects of people, organizations, resources, and military scenario events based on the discovery and reasoning results of the knowledge group; the military strategy map generation module generates a military strategy map based on the output results of the evolution state discrimination module; the evaluation result collection module collects the evaluation results of military experts on military strategy related information; the military strategy guidance information update module dynamically updates the military strategy guidance information of the next cycle in combination with the evolution state and evaluation results of military strategy related information.

[0134] The knowledge application decision-making layer also includes query modules such as database module, entity and relationship extraction module, event subject library, and organization library.

[0135] Aiming at the application of knowledge in the field of military intelligence, the present invention takes military strategy as the core of guidance, and conducts research from three aspects: knowledge query, knowledge recommendation and visualization. Taking knowledge query as an example, users' queries on knowledge are often expressed as a set of keywords, and the answer is often a set of connected nodes, which are represented in the knowledge network as subgraphs that match the search content. Due to the semantic completeness of keyword matching, a huge knowledge network has a variety of subgraphs that meet the conditions, or have different topological structures, or have different keyword distributions. In the query search process, the content relevance of the nodes and edges is first calculated, and the search results are generated using a heuristic search algorithm. The top-k subgraphs are sorted in combination with the sorting algorithm to give the final top-k search results. In view of the particularity of military strategy, the knowledge application decision layer provides query modules such as a database, a univariate relationship query component, a multivariate relationship query component, an event subject library and an organization library. Preferably, the entity and relationship extraction module also includes components such as a univariate relationship query component and a multivariate relationship query component, which are used to provide functions such as univariate relationship query and multivariate relationship query.

[0136] The database module includes a database management component, a database query component and a database display component; the database management component is used to store the data cleaned and classified by the data acquisition module into two types of databases, namely the basic database and the trend database, and add entry time, publication date and keyword attributes to each incoming message; the database query component provides data query functions, including a publication date time range plug-in, a keyword query plug-in and a database type, and searches for messages in the selected database according to the publication date range and keywords entered by the user, and calls the database display component to push the search results to the large display screen.

[0137] The entity and relationship extraction module includes an entity recognition component, an entity graph component and a relationship recognition component; the entity recognition component queries relevant entities based on the input keywords, and pushes the relevant attributes of the queried entity to the display screen; the entity graph component includes a first entity name plug-in, and the entity graph component queries the relevant nodes of the entity in the knowledge graph based on the input single entity name, obtains and displays its associated attributes and relationship query results; the relationship recognition component includes a relationship selection plug-in and multiple second entity name plug-ins, and the relationship recognition component infers and displays the knowledge graph of the specified relationships between the input multiple entities based on the input multiple relationship types, through multiple layers of penetration between the multiple entities; the specified relationships include associated relationships, associated organizations, and associated events between entities.

[0138] For example, by inputting a single entity into the univariate relationship query component, you can query the relevant nodes of the entity in the knowledge graph and obtain query results such as its associated attributes and relationships. For example, univariate relationship (person-person relationship) query, univariate relationship (organization-person relationship) query, and univariate relationship (person-organization relationship) query. By inputting multiple entities into the multi-relationship query component, through multi-layer penetration between multi-entities, you can obtain query results such as the associated relationships, associated organizations, and associated events between entities. For example, multi-relationship (person-person relationship) query, multi-relationship (person-organization relationship) query, and so on.

[0139] The following is an example of the extraction results of entities, attributes, etc.

[0140] Entity extraction types: people, organizations, military units, countries, cities, regions, weapons, etc.

[0141] Attribute extraction type:

[0142] Attribute extraction types for people: Chinese name, English name, nationality, date of birth, place of birth, gender, position, education, etc.; Attribute extraction types for organizations: nature, establishment time, and establishment place.

[0143] Relation extraction type:

[0144] Relationship between characters:

[0145] <character> colleague<character>, <character> relative<character>, <character> classmate<character>, <character> fellow villager<character>, <character> superior<character>, <character> accomplice<character>, <character> cooperation<character>, <character> support<character>, <character> funding<character>.

[0146] Relationship between people and organizations:

[0147] <character> works for <organization>, <character> is the leader of <organization>, <character> is a core member of <organization>, <character> is a member of <organization>, <character> is a member of <organization>, <character> supports <organization>, <character> funds <organization>, <character> founded <organization>, <character> cooperates with <organization>.

[0148] Relationship between organizations:

[0149] <Organization> cooperates with <Organization>, <Organization> funds <Organization>, <Organization> is affiliated with <Organization>, <Organization> is in competition with <Organization>, <Organization> is hostile to <Organization>, <Organization> supports <Organization>.

[0150] The relationship between the character and the country:

[0151] <person> was born in <country>, <person> is a national of <country>, <person> lives in <country>, <person> studies in <country>, <person> works in <country>, <person> serves <country>, <person> visits <country>.

[0152] Relationship between characters and military units:

[0153] <Character> is affiliated with <military unit>, <Character> serves in <military unit>, <Character> visits <military unit>.

[0154] The relationship between characters and cities:

[0155] <Character> was born in <city>, <Character> lives in <city>, <Character> works in <city>, <Character> visits <city>.

[0156] The relationship between weapons and the state:

[0157] <country> develops <weapons>, <country> possesses <weapons>, <country> purchases <weapons>, <country> seizes <weapons>, <country> intercepts <weapons>, <country> uses <weapons>, <country> destroys <weapons>.

[0158] Existing multivariate relationship queries often calculate the maximum weight subgraph of the keyword set input by the user in the entire domain knowledge network. Taking two keywords as an example, it is to calculate the shortest path or the most meaningful path between two nodes. There are two situations in the figure: one is that although a certain node is connected to many nodes, the node has a very small correlation with its closest node; the other is that although the distance between the two nodes is very long, the node and the node content on the path have a very high similarity, and the comprehensive calculation of these nodes will have a very high correlation. Therefore, after calculating the shortest path between the two nodes, the present invention uses the Steiner node representation method to determine the potential hub node on the shortest path (the HITS algorithm can be used to calculate the hub value of each node). If there is a hub node, the second shortest path is found. If the next shortest path does not exist, the previous shortest path containing the hub node is used. This method can relatively make up for the shortcomings of directly finding the shortest path between two nodes.

[0159] The event topic library realizes the analysis of key event topics from the dimensions of event basic information, event introduction, development history, related messages, relationship maps, event hot words, etc., and supports the addition and configuration management of key events, realizes the summary and analysis of different event topics generated based on the development and evolution of events, realizes the aggregation and timeline analysis of related messages based on event topics, and realizes the viewing of event topic related messages. Specifically, the event topic library includes an event analysis component, an event addition component, an event association component, a related message aggregation component, and a related message query component; the event analysis component analyzes each key event from multiple dimensions including event basic information, event introduction, development history, related messages, relationship maps, and event hot words; the event addition component is used to manually add key events; the event association component is used to summarize and analyze different event topics generated by the development and evolution of events; the related message aggregation component is used to aggregate messages related to event topics according to the timeline; the related message query component is used to provide a related message viewing function.

[0160] The organization library manages key organizations from the perspectives of basic organization information, organization introduction, organizational experience, organizational structure, and relationship map, and supports the addition and configuration management of key focus organizations. Specifically, the organization library includes an organization basic information management component, an organization experience management component, an associated event management component, an associated message management component, and a relationship map generation component; the organization basic information management component is used to manage the basic information of all organizational entities in the organization library; the organization experience management component is used to manage the change process of the organization's related personnel structure and resource information; the associated event management component is used to add all event types related to organizational entities in real time based on the results of knowledge reasoning; the associated message management component is used to manage all messages related to the organization; and the relationship map generation component is used to organize and generate a binary relationship knowledge map related to the organization.

[0161] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0162] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0163] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0164] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0165] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0166] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An intelligence collaborative processing platform based on knowledge graph, characterized in that: The intelligence collaborative processing platform includes a data acquisition and processing layer, a knowledge extraction and intelligent analysis layer, and a knowledge application and decision-making layer; The data acquisition processing layer includes a data acquisition module, a data format conversion module and a knowledge base management module; The data acquisition module acquires military intelligence business data from multiple data sources, cleans and classifies them, and stores them in corresponding databases. Military intelligence business data include various military intelligence results data analyzed and studied in recent years and intelligence information reported by departments at all levels with real-time access; the data format conversion module is used to uniformly represent the multi-source heterogeneous military intelligence business data collected by the data acquisition module; the knowledge base management module is used to generate a semantic network knowledge base based on the military intelligence business data after format conversion, dynamically expand the semantic network knowledge base, and integrate duplicate content and conflicting content in the knowledge ontology construction; the semantic network knowledge base includes four categories: people, organizations, resources, and military scenario events; a class domain knowledge ontology, each domain knowledge entity has corresponding attribute information, and each domain knowledge entity includes an active state attribute; The knowledge extraction intelligent analysis layer includes a keyword extraction module, a keyword stratification module, a semantic fusion module and a knowledge ontology dynamic evolution module; The keyword extraction module is used to obtain several words or phrases with high relevance to the entity as basic keywords; The keyword stratification module constructs a core keyword set for the current period based on the evolution trend of historical military strategy related information and the military strategy guidance information for the current period; then, according to the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, the remaining basic keywords are stratified and classified to construct a spatial distribution model between keywords and entities, and between keywords; The semantic fusion module takes the association between keywords and knowledge ontologies in various fields as the main feature, establishes the feature vector of knowledge ontologies in various fields, and calculates the similarity between knowledge ontologies in various fields, and obtains the relationship between knowledge ontologies in various fields according to the general threshold; starting from the keyword frequency and distance characteristics, it mines the hierarchical and sequential relationships between knowledge entities in various fields, constructs a knowledge ontology evolution semantic network, and performs semantic fusion on multi-source heterogeneous data in the semantic network knowledge base; The knowledge ontology dynamic evolution module starts from the dynamic evolution of the knowledge ontologies of each field and between each other, performs upper and lower semantic reasoning on the integrated semantic network knowledge base, finds the link between the two knowledge ontologies, and then extends the link to discover the potential semantics or broader relationships on the link, and mines the evolution trend of the multi-relationships between people, organizations, resource behavior sequences and military scenario events to build a multi-knowledge semantic network; introduces the time slice method to cumulatively expand the network topology decoupling, sets a position label for each node and link edge in each time slice, indicates whether there is a state transfer of the network node and edge in the corresponding time slice, discovers and infers the knowledge group, and updates the active state attributes of each field knowledge entity; The knowledge application decision layer includes an evolution state determination module, a military strategy map generation module, an evaluation result acquisition module and a military strategy guidance information update module; The evolution state identification module identifies the evolution state and evolution trend of military strategy related information from the aspects of characters, organizations, resources and military scenario events according to the discovery and reasoning results of the knowledge group, and updates the active state attributes of each domain knowledge entity at the same time; the military strategy map generation module selects several key focus characters, key focus organizations, key focus resources and key focus events of the current period according to the updated active state attributes to generate a military strategy map; the evaluation result collection module collects the evaluation results of military experts on military strategy related information; the military strategy guidance information update module dynamically updates the military strategy guidance information of the next period in combination with the evolution state and evaluation results of the military strategy related information.

2. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The basic structure of the data in the semantic network knowledge base is: SKB = (C, E, A, V), where C = {c1, c2, ..., c n } represents a category set, E = {e1, e2, ..., e m } represents an entity set, A = {a1, a2, ..., a k } represents an entity attribute set where a i ∈A represents the basic attributes, Indicates the value corresponding to the attribute; in the category set {C}, there is a parent-child relationship between categories, that is, c j ∈c s , indicating category c j It is category c S A subclass of .

3. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The knowledge base management module includes an entity construction component, a semantic network knowledge base generation component and a conflict fusion component; The entity construction component is used to study the characteristics of some unstructured texts, manually construct an initial military domain knowledge entity attribute library, and then expand the military domain knowledge entity attribute library based on word similarity, and at the same time, use a dictionary method combined with part-of-speech attribute discrimination to annotate entities and attributes of unstructured texts; The semantic network knowledge base generation component uses the keyword spatial position co-occurrence feature as a training set, uses a deep learning model to train the annotated entity and attribute data, uses the trained deep learning model to extract new domain knowledge data, evaluates the effectiveness of the extracted entities and attributes, dynamically expands the evaluated effective entities and attributes to the military domain knowledge entity attribute library, and then identifies the domain knowledge entities and attributes in the massive text based on the reinforcement machine learning method; The conflict fusion component analyzes term conflicts, semantic conflicts and predicate conflicts, and adopts a hybrid method of logic tree fusion, frequency fusion and syntax fusion to select and fuse duplicate content and conflicting content in knowledge ontology construction.

4. The intelligence collaborative processing platform based on knowledge graph according to claim 3 is characterized in that: The conflict fusion component includes a term library, a predicate library, an ontology library, a term conflict processing unit, a semantic conflict processing unit, a predicate conflict processing unit and a fusion unit; The terminology library and predicate library respectively extract relevant knowledge ontology from the military domain knowledge entity attribute library, respectively construct terminology set, predicate set and semantic set and transmit them to the ontology library; the ontology library analyzes the terminology set, predicate set and semantic set to obtain the existing terminology conflict, predicate conflict and semantic conflict; the terminology conflict processing unit calls the logic tree fusion model, the frequency fusion model and the syntax fusion model to process the terminology conflict, the semantic conflict processing unit calls the syntax fusion model to process the semantic conflict, and the predicate conflict processing unit calls the frequency fusion model and the syntax fusion model to process the predicate; The fusion unit integrates the processing results of the term conflict processing unit, the semantic conflict processing unit, and the predicate conflict processing unit to generate a fusion knowledge item; The logic tree fusion model fuses the conflicting items by determining the logic tree relationship between the two conflicting items; The frequency fusion model uses high-frequency co-occurring knowledge items as the highest score output; the syntactic fusion model uses the syntactic dependency tree discriminant to discriminate the semantic space dependency relationship between entities, attribute words and attribute value words, and makes choices for multiple conflicting items.

5. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The keyword extraction module obtains a word set D-terms by segmenting the document set D of each domain knowledge entity content, maps the document set D to a graph G, where the vertices are words and the edges represent the co-occurrence frequency of two words, then calculates the network node centrality between the nodes, sorts the network node centrality of each word, merges some words into phrases through the n-gram algorithm, and outputs several results with the highest ranking as basic keywords related to the domain knowledge entity.

6. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The keyword stratification module selects entities that are active in the current period based on the evolution trend of historical military strategy-related information, and selects the first keyword related to the entity from the basic keywords for each entity-keyword association relationship; obtains and analyzes the military strategy guidance information of the current period, and selects the second keyword related to the entity from the basic keywords according to the entity corresponding to the military strategy guidance information; and combines the first keyword and the second keyword to form a core keyword set for the current period.

7. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The semantic fusion module uses an entity similarity calculation model and a category network fusion model to fuse the semantic knowledge network; The entity similarity calculation model uses the following formula to merge entities between the semantic network knowledge bases SKB1 and SKB2: f:(C i ,E i ,A i ,V i )×(C j ,E j ,A j ,V j )→(C,E,A,V) If the entity names in the two knowledge bases are the same, it is considered that the two knowledge bases describe the same content; if the entity name in SKB1 is i The same entity name was not found in SKB2 j , but find e with the same domain knowledge description or a similarity greater than the first preset similarity threshold j , then the entities in the two knowledge bases are considered to be the same; if the entity name e in SKB1 i In SKB2, only e with similarity less than the first preset similarity threshold can be found. j , but the similarity of other attributes and attribute values ​​is greater than the second preset similarity threshold, then the entities in the two knowledge bases are considered to be the same; when the two entities are judged to be the same, let e = e i Or e=e j , C=(C i ∪C j ), A=(A i ∪A j ),in When two entities are judged to be the same or two semantic network category nodes are merged, the category network fusion model is used to execute C=(C i ∪C j ); When the entity has only one parent class, determine whether the parent classes between the two knowledge bases are the same: If c i With c j If the category words are the same, then the parent categories are the same; If c i With c j If the category words are inconsistent, but the domain knowledge descriptions are the same or the similarity is greater than the third preset threshold, the categories in the two knowledge bases are considered to be the same; if the category words and domain content descriptions are inconsistent, then it is judged that the similarity of other attributes of the category and the attribute values ​​is greater than the fourth preset similarity threshold, then the categories in the two knowledge bases are considered to be the same; when an entity has multiple parent classes in different knowledge bases, and Then C i ∩C j As the common parent class of the entity, other parent classes call the deep inheritance algorithm and the fusion algorithm.

8. The intelligence collaborative processing platform based on knowledge graph according to claim 1 is characterized in that: The knowledge ontology dynamic evolution module includes a unit knowledge reasoning component, a multi-relation reasoning component and a dynamic knowledge reasoning component; The unit knowledge reasoning component loads the knowledge network data and establishes a data structure of neighbor nodes and label information corresponding to each node; for each node in the knowledge network, according to the label attributes of the node and its neighbor nodes, and the specified label information ratio adjustable parameter p, the probability value of the neighbor node of the node being wandered to is calculated, and a number of random selections are made from the neighbor nodes of the node, and the probability of each neighbor node being selected meets the calculated probability value; then, according to the obtained probability value and wandering parameters, wandering is started to obtain a number of wandering paths; According to the walking path, the Word2vec method is called for training to obtain word vectors and perform multi-label classification tasks on the knowledge network nodes; The multivariate relational reasoning component constructs possible candidate LE patterns based on the combination of all edge types on the knowledge network; traverses the entire graph to find all corresponding instances of different candidate LE patterns, calculates the weights and values ​​of different candidate LE patterns, and solves the credible LE pattern through a greedy algorithm; uses the selected credible LE pattern to match in the knowledge network to obtain the reasoning result; The dynamic knowledge reasoning component introduces time slices to cumulatively expand the decoupling of network topology. By setting a position label for each node and linked edge in each time slice, it indicates whether there is a state transfer of network nodes and edges in the corresponding time slice, and discovers and infers knowledge groups.

9. A potential relationship prediction method based on knowledge graph, characterized in that: The potential relationship prediction method is performed based on the knowledge graph-based intelligence collaborative processing platform as claimed in any one of claims 1 to 8; The potential relationship prediction method comprises the following steps: Obtain military intelligence business data from multiple data sources, clean and classify them, and store them in the corresponding database. Military intelligence business data includes various military intelligence results data analyzed and studied in recent years and intelligence information reported by departments at all levels in real time; The multi-source heterogeneous military intelligence business data collected by the data collection module are uniformly represented; a semantic network knowledge base is generated based on the military intelligence business data after format conversion, and the semantic network knowledge base is dynamically expanded, while the duplicate and conflicting contents in the knowledge ontology construction are integrated; the semantic network knowledge base includes four aspects: people, organizations, resources, and military scenario events; in the domain knowledge ontology, each domain knowledge entity has corresponding attribute information, and each domain knowledge entity includes an active state attribute; Acquire several words or phrases that are highly relevant to the entity as basic keywords; Based on the evolution trend of historical military strategy-related information and the military strategy guidance information of the current period, a core keyword set of the current period is constructed; then, according to the dependency relationship of syntactic analysis and the co-occurrence frequency of keywords, the remaining basic keywords are layered and classified, and a spatial distribution model between keywords and entities, and between keywords and keywords is constructed; The relationship between keywords and knowledge ontologies in various fields is used as the main feature to establish the feature vector of knowledge ontologies in various fields, and the similarity between knowledge ontologies in various fields is calculated. The relationship between knowledge ontologies in various fields is obtained according to the common threshold. Starting from the keyword frequency and distance characteristics, the hierarchical and sequential relationships between knowledge entities in various fields are mined, and a knowledge ontology evolution semantic network is constructed to semantically fuse multi-source heterogeneous data in the semantic network knowledge base. Starting from the dynamic evolution of knowledge ontologies in various fields and between each other, the upper and lower layers of semantic reasoning are performed on the integrated semantic network knowledge base. After finding the link between two knowledge ontologies, the link is extended to discover the potential semantics or broader relationships on the link, and the evolution trend of the multi-relationships between people, organizations, resource behavior sequences and military scenario events is mined to construct a multi-knowledge semantic network; the time slice method is introduced to cumulatively expand the network topology decoupling, and position labels are set for each node and link edge in each time slice to indicate whether there is a state transfer of network nodes and edges in the corresponding time slice, so as to discover and reason about knowledge groups and update the active state attributes of each domain knowledge entity; Based on the findings and reasoning results of the knowledge group, the evolution status and evolution trend of military strategy-related information are judged from the aspects of people, organizations, resources and military scenario events, and the active state attributes of each domain knowledge entity are updated; according to the updated active state attributes, several key focus people, key focus organizations, key focus resources and key focus events in the current period are selected to generate a military strategy map; Collect the evaluation results of military experts on military strategy-related information; dynamically update the military strategy guidance information for the next cycle based on the evolution status of military strategy-related information and the evaluation results.

Images