CN117196027A - Training sample generation method and device based on knowledge graph - Google Patents
Training sample generation method and device based on knowledge graph Download PDFInfo
- Publication number
- CN117196027A CN117196027A CN202311472158.4A CN202311472158A CN117196027A CN 117196027 A CN117196027 A CN 117196027A CN 202311472158 A CN202311472158 A CN 202311472158A CN 117196027 A CN117196027 A CN 117196027A
- Authority
- CN
- China
- Prior art keywords
- sample
- formation
- relation
- execution unit
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 75
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 194
- 230000008569 process Effects 0.000 claims description 16
- 238000010276 construction Methods 0.000 claims description 11
- 230000007123 defense Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 11
- 238000004590 computer program Methods 0.000 description 8
- 238000012937 correction Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 101150049349 setA gene Proteins 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a training sample generation method and a training sample generation device based on a knowledge graph. Then randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a respective sub-map. In terms of form, any sample is a five-tuple consisting of a formation type, a first relationship side, an execution unit, a second relationship side, and a set of mounted devices. And finally, determining the scoring label of each sample, obtaining a training sample carrying the label, and generating a training sample set. According to the application, the knowledge graph is utilized to generate the sample data for training the formation capability assessment model, so that not only can a large-scale sample meeting the formation requirement be generated as required, but also the workload of sample data labeling can be effectively reduced, and the method is flexible and efficient.
Description
Technical Field
The application relates to the technical field of formation capability evaluation, in particular to a training sample generation method and device based on a knowledge graph.
Background
For the formation objective capability assessment problem, the conventional capability assessment method has limitations in weight determination, ambiguity assessment, randomness and the like. In recent years, by virtue of strong nonlinear fitting capability, capability assessment based on a deep neural network becomes a mainstream method, and is widely applied to the military and civil fields.
However, the method needs to rely on a large number of marked samples for model training, and the larger the sample size is, the higher the marking quality is, and the better the performance of the obtained evaluation model is. However, sample generation and annotation of formation objective capability assessment problems often requires more specialized domain knowledge. Thus, there is a need for a flexible, convenient, efficient and feasible training sample generation method for formation objective capability assessment.
Disclosure of Invention
The application provides a training sample generation method and device based on a knowledge graph, which can generate a large-scale sample meeting formation requirements as required, and is used for training formation target capability assessment models in various scenes so as to improve the performance of the assessment models.
An embodiment of the present application provides a training sample generating method based on a knowledge graph, where the method includes:
encoding the formation type, the execution unit and the carrying device group, and constructing a knowledge graph for expressing formation capacity according to the corresponding relation among the formation type, the execution unit and the carrying device group;
randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a sub-map respectively, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group;
and determining the scoring label of each sample, obtaining a training sample carrying the label, and generating a training sample set.
Optionally, the knowledge graph includes a vertex set and an edge set;
the encoding of the formation type, the execution unit and the onboard device group includes:
and respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and merging to obtain a vertex set.
Optionally, forming the vertex set includes:
determining the types and the quantity of the formation types, and sequentially encoding the formation types to obtain a formation type set，Wherein->Representing a set of formation types +.>The%>Vertex corresponding to->A seed formation type is used for the formation of the seed, and (2)>A number representing a formation type;
encoding each execution unit in turn to obtain an execution unit set，Wherein->Representing execution Unit set +.>The%>Vertex corresponding to->A seed Execution Unit (EU)>Representing the number of execution units;
coding the carrying device groups containing different device types and numbers in turn to obtain a carrying device group set，Wherein->Representing a set of mounted device groups->The%>Vertex corresponding to->Seed carrying device group,/->Representing the number of the carried equipment groups;
based on the above-mentioned formation type setExecution unit set->And a set of mounted device groups->Generating the vertex set ∈>，。
Optionally, the constructing a knowledge graph for expressing formation capability according to the correspondence between the formation type, the execution unit and the carrying device group includes:
constructing a first relation edge set according to the formation composition relation between the formation type and the execution unit;
constructing a second relation edge set according to the equipment carrying relation between the execution unit and the carrying equipment group;
combining the first relation edge set and the second relation edge set to obtain an edge set;
and constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
Optionally, the forming of the edge set includes;
constructing a first relation edge set according to the formation composition relation between the formation type and the execution units，Wherein->Vertex +.A vertex representing a formation type in the above vertex set>Top +.>Connection, i.e. first->The species formation type may include +.>A seed execution unit;
constructing a second relation edge set according to the device carrying relation between the execution unit and the carrying device group，Wherein->Representing execution unit vertices +.>Vertex of the mounted device group->Connection, i.e. first->The seed execution unit can be loaded with +.>A seed carrying device group;
based on the first relation edge setAnd a second set of relationship edges->Generating the above edge set->，。
Optionally, forming any of the samples includes:
selecting a target formation type from the set of formation types;
randomly selecting a plurality of first relation edges related to the target formation type from the first relation edge set, and determining corresponding target execution units according to the first relation edges;
for each selected target execution unit, selecting at least one second relation side related to the target execution unit from the second relation side set, and determining a corresponding target carrying device group according to the second relation side;
and constructing and forming the sample based on the target formation type, the first relation side, the target execution unit, the second relation side and the target carrying device group.
Optionally, determining the scoring label of each sample to obtain a training sample carrying the label includes:
determining attack scores, defense scores and reconnaissance scores of each sample by using an analytic hierarchy process to obtain a scoring result corresponding to each sample;
and taking the scoring result of each sample as a scoring label corresponding to the sample to obtain a training sample carrying the label.
Optionally, the step of determining the scoring label of each sample to obtain a training sample carrying the label includes:
scoring the attack capability, the defending capability and the reconnaissance capability of each sample by using a pre-training model to obtain a multi-dimensional initial score of each sample;
correcting the multi-dimensional initial score based on expert experience to obtain a final score of each sample;
and taking the final score of each sample as a scoring label of the sample to obtain a training sample carrying the label.
A second aspect of an embodiment of the present application provides a training sample generating device based on a knowledge-graph, where the device includes:
the knowledge graph construction module is used for encoding the formation type, the execution unit and the carrying device group and constructing a knowledge graph for expressing formation capacity according to the corresponding relation among the formation type, the execution unit and the carrying device group;
the sub-map extraction module is used for randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a sub-map respectively, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group;
the training sample generation module is used for determining the scoring label of each sample, obtaining a training sample carrying the label and generating a training sample set.
Optionally, the knowledge graph construction module includes:
the vertex set generation submodule is used for respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and combining the formation type set, the execution unit set and the carrying device group set to obtain a vertex set;
the first relation set generating sub-module is used for constructing a first relation edge set according to the formation composition relation between the formation type and the execution unit;
a second relation set generating sub-module, configured to construct a second relation edge set according to the device carrying relation between the execution unit and the carrying device group;
the edge set generating submodule is used for merging the first relation edge set and the second relation edge set to obtain an edge set;
and the construction submodule is used for constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
A third aspect of the embodiments of the present application provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the above-described method of the present application.
Compared with the prior art, the application has the following advantages:
according to the training sample generation method based on the knowledge graph, the formation type, the execution unit and the carrying device group are encoded, and the knowledge graph for expressing formation capacity is constructed according to the corresponding relation among the formation type, the execution unit and the carrying device group. Then randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a respective sub-map. In terms of form, any sample is a five-tuple consisting of a formation type, a first relationship side, an execution unit, a second relationship side, and a set of mounted devices. And finally, determining the scoring label of each sample, obtaining a training sample carrying the label, and generating a training sample set. According to the application, the array type, the execution unit and the carrying device which possibly exist in the formation are uniformly expressed, the knowledge graph about the formation is constructed, and the sample set is formed by randomly generating the sub-graph, so that the purpose of generating a large-scale sample meeting the formation requirement as required is achieved. In addition, a training sample carrying a label is formed by automatically calculating a sample label generation mode combined with expert correction, so that the workload of sample data labeling can be effectively reduced, and the method is flexible and efficient. The generated training samples can be used for training the formation target capacity assessment model in various scenes so as to improve the performance of the assessment model.
Drawings
FIG. 1 is a flowchart of a training sample generation method based on a knowledge-graph according to an embodiment of the present application;
FIG. 2 is a schematic diagram of formation types in a training sample generation method based on a knowledge-graph according to an embodiment of the present application;
fig. 3 is a schematic diagram of a knowledge graph constructed in a training sample generating method based on the knowledge graph according to an embodiment of the present application;
FIG. 4 is a schematic diagram of sub-patterns extracted in a training sample generation method based on a knowledge-graph according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a training sample generating device based on a knowledge-graph according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Referring to fig. 1, fig. 1 is a flowchart of a training sample generating method based on a knowledge-graph according to an embodiment of the application. As shown in fig. 1, the method comprises the steps of:
step S101: encoding the formation type, the execution unit and the carrying device group, and constructing a knowledge graph for expressing formation capability according to the corresponding relation among the formation type, the execution unit and the carrying device group.
In the present embodiment, for various formation scenes, there are three main types of subjects included in the scenes: the formation type, the execution units participating in the formation and the devices that the execution units can carry. For an aircraft formation scene, the main body comprises: formation type, aircraft group and onboard equipment group. The formation type refers to an arrangement mode of the aircraft, such as a longitudinal formation, a transverse formation, a ladder formation, a wedge formation and the like. The execution unit refers to a group of aircraft that can be used to compose a corresponding formation type, different groups of aircraft being composed of different models and different numbers of aircraft. The carrying device group refers to carrying devices which can be installed by the execution unit, and different carrying device groups can be composed of devices with different models and different numbers.
In the application, a knowledge graph for expressing formation capability is constructed, which mainly comprises two processes of constructing a vertex set and constructing an edge set.
Optionally, the process of encoding the formation type, the execution unit and the carrying device group to obtain the vertex set mainly includes: and respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and merging to obtain a vertex set.
In particular, the encoding may be performed as follows:
step S101-1: determining the types and the quantity of the formation types, and sequentially encoding the formation types to obtain a formation type set，Wherein->Representing a set of formation types +.>The%>Vertex corresponding to->A seed formation type is used for the formation of the seed, and (2)>Indicating the number of formation types.
For example, whenEqual to 4, setting +.>Representing a "wedge team", ">Representing "echelon", -a->Representing "team", "A>Representing a "cross team".
Step S101-2: encoding each execution unit in turn to obtain an execution unit set，Wherein->Representing execution Unit set +.>The%>Vertex corresponding to->A seed Execution Unit (EU)>Representing the number of execution units.
For example, in an aircraft formation scenario, whenEqual to 3, setting ++under the condition that 3 execution units for formation use are available, namely a type I early warning machine 1 frame, a type II fighter plane 2 frame and a type I unmanned plane 2 frame>Indicating the corresponding aircraft group of the I-type early warning machine 1 frame, < ->Representing the corresponding airplane group of type II fighter plane 2 frames->And the corresponding airplane group of the type I unmanned aerial vehicle 2 is shown.
Step S101-3: coding the carrying device groups containing different device types and numbers in turn to obtain a carrying device group set，Wherein->Representing a set of mounted device groups->The%>Vertex corresponding to->Seed carrying device group,/->Indicating the number of groups of devices to be mounted.
For example, whenEqual to 3, setting +.A.under the conditions that 3 kinds of carrying equipment groups needing to be installed are 3 kinds of I-type missiles 4, II-type missiles 4 and II-type bombs 2>4 corresponding carrying device groups of I-type missiles are indicated, namely ∈4>4 corresponding carrying device groups of II-type missiles are indicated, namely ∈4->The group of 2 corresponding mounted devices of the type II bomb is shown.
Step S101-4: based on the above-mentioned formation type setExecution unit set->And a set of mounted device groups->Generating the vertex set ∈>，。
Optionally, the constructing a knowledge graph for expressing formation capability according to the correspondence between the formation type, the execution unit and the carrying device group includes: constructing a first relation edge set according to the formation composition relation between the formation type and the execution unit; constructing a second relation edge set according to the equipment carrying relation between the execution unit and the carrying equipment group; combining the first relation edge set and the second relation edge set to obtain an edge set; and constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
In particular implementations, the encoding may be performed as follows to form the edge set:
step S101-5: constructing a first relation edge set according to the formation composition relation between the formation type and the execution units，Wherein->Vertex +.A vertex representing a formation type in the above vertex set>Top +.>Connection, i.e. first->The species formation type may include +.>An execution unit.
In this embodiment, the first relationship edge subset corresponding to each formation type may be obtained by connecting the formation type with the corresponding execution unit that may be included. After statistical merging, the connection relation between all formation types and all execution units, namely a first relation edge set, can be obtained. Thus first relation edge set->In effect characterizing the inclusion relationship between the formation type and the execution unit.
For example, formation type"Cross team" is indicated, provided that it can be represented by execution units +.>(representing the airplane group corresponding to the I-type early warning airplane 1 frame) and +.>(representing the corresponding airplane group of type II fighter plane 2 frames), the formation type vertexes are formed>Respectively with execution unit vertex->And->And (5) connection. Thereby formation type->The first relation side is included with +>And->。
Step S101-6: constructing a second relation edge set according to the device carrying relation between the execution unit and the carrying device group，Wherein->Representing execution unit vertices +.>Vertex of the mounted device group->Connection, i.e. first->The seed execution unit can be loaded with +.>A device group is mounted.
In the present embodiment, the second relation edge setIs associated with the first set of relationship edges +.>Is similar to the construction process except that the subject becomesExecution unit vertex->Vertex of the mounted device group->. Second relation edge set->The inclusion relationship between the execution units and the group of onboard devices is characterized in practice.
Step S101-7: based on the first relation edge setAnd a second set of relationship edges->Generating the above edge set->，。
The vertex set is obtained by the above waySum edge set->Then, a knowledge graph can be constructed>，. From the constructional element, the constructed knowledge graph can be actually regarded as five-tuple data consisting of a formation type set, a first relation edge set, an execution unit set, a second relation edge set and a carrying device group set. Reflecting the type of formation that can be carried out, the group of aircraft that can be laid out and the piggybacked in the scene of formation of aircraftAnd then generating a plurality of specific formation execution schemes according to the inclusion relation among the equipment groups. Thus can be treated by knowledge graph->The formation capability under this scenario is expressed.
Step S102: randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a sub-map, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group.
In this embodiment, after the knowledge graph for expressing the formation capability is constructed in the above manner, a sample set may be further constructed by randomly extracting sub-graphs from the knowledge graph，. Wherein,indicate->Strip sample->Representing the total number of samples.
In particular, any of the above samplesThe formation of (2) comprises:
step S102-1: a target formation type is selected from the set of formation types.
For example, one may from a set of formation typesRandom selection of->Species formation type (+)>) And in the sample->Add the team type vertex ++>And selecting a subsequent execution unit and a carried device group by taking the target formation type as a target formation type.
Step S102-2: and randomly selecting a plurality of first relation edges related to the target formation type from the first relation edge set, and determining a corresponding target execution unit according to the first relation edges.
In this embodiment, a plurality of first relationship edges related to the target formation type may be randomly selected from the first relationship edge set, where the first relationship edges related to the target formation type refer to vertices related to the selected formation typeAll first relation edges of the connection +.>And easily known, add>. By from->A plurality of strips are selected immediately to form a set +.>I.e. +.>And will aggregate->Add sample->。
On the basis, willThe first relation edge is respectively connected with the target execution unit vertex to form a set +.>Add sample->Wherein->。
Step S102-3: for each selected target execution unit, selecting at least one second relation side related to the target execution unit from the second relation side set, and determining a corresponding target carrying device group according to the second relation side.
In the present embodiment, for a collectionEach selected target execution unit vertex is randomly selected from a plurality of second relation edges connected with the selected target execution unit vertex, and then the second relation edges selected by the target execution unit vertices corresponding to the selected target execution unit vertices are combined to form a set->，And add sample->. It should be noted that, when selecting the second relationship edge connected to the vertex of each target execution unit, the number of the selected second relationship edges is at least 1, that is, it is required to ensure that each selected vertex of the target execution unit can extract a corresponding second relationship edge.
In addition, the collection can also be directly fromCorresponding second relation edge set +.>A plurality of strips are randomly selected to form a set +.>. At this time, a->。
On the basis, willA target mounted device group vertex set formed by all mounted device group vertices correspondingly connected with each second relation edge +.>Add sample->Wherein
。
Step S102-4: and constructing and forming the sample based on the target formation type, the first relation side, the target execution unit, the second relation side and the target carrying device group.
Through the selection process, a sample can be obtainedWhich is formally a knowledge graph +.>Is a subset of the set of (c).
It should be noted that the sub-pattern or sample selected in the above manner has random uncertainty. In each round of selection process, the number of execution units and the number of carrying device groups contained in each extracted sample are different due to different selection numbers of the first relation edge and the second relation edge, so that a large-scale sample can be generated as required.
Step S103: and determining the scoring label of each sample, obtaining a training sample carrying the label, and generating a training sample set.
In this embodiment, an analytic hierarchy process may be used to determine an attack score, a defense score, and a reconnaissance score of each sample, so as to obtain a score result corresponding to each sample. And then, taking the scoring result of each sample as a scoring label corresponding to the sample to obtain a training sample carrying the label.
The analytic hierarchy process can decompose the problem into different composition factors according to the nature of the problem and the total target to be achieved, and aggregate and combine the factors according to different levels and the mutual correlation influence and membership among the factors to form a multi-level analysis structure model. The decision problem is decomposed into different hierarchical structures according to the sequence of a total target, sub-targets of each layer and evaluation criteria until a specific spare power switching scheme, then the priority weight of each element of each layer to a certain element of the previous layer is obtained by a method for solving and judging matrix eigenvectors, and finally the final weight of each alternative scheme to the total target is integrated in a hierarchical mode through a weighted sum method. The higher the weight, the higher the score, indicating that the formation scheme is better.
Exemplary, for the sampleFirstly, grading each capability of the formation type, the contained execution units and the carried equipment group by using an analytic hierarchy process to obtain a grading result +.>And->. Wherein (1)>For sample->D is the total number of scored dimensions.
Optionally, the step of determining the scoring label of each sample to obtain the training sample carrying the label may further include: scoring the attack capability, the defending capability and the reconnaissance capability of each sample by using a pre-training model to obtain a multi-dimensional initial score of each sample; correcting the multi-dimensional initial score based on expert experience to obtain a final score of each sample; and taking the final score of each sample as a scoring label of the sample to obtain a training sample carrying the label.
In this embodiment, the scoring label of the sample may be determined by estimating the initial score and correcting the initial score by an expert to obtain the scoring result. The pre-training model can be a capability assessment model based on a deep neural network and is obtained through open-source data set training. Carrying out multidimensional scoring on the attack capability, the defending capability, the reconnaissance capability and the like of the sample by utilizing a pre-training model to obtain an initial score. Then, expert knowledge based on the +.>Is corrected to obtain the final score +.>Wherein->For sample->Final score of the d-th dimension of (c).
According to the embodiment of the application, the array type, the execution unit and the carrying device which possibly exist in the formation are uniformly expressed, the knowledge graph about the formation is constructed, and the sample set is formed in a mode of randomly generating the sub-graph, so that the purpose of generating a large-scale sample meeting the formation requirement as required is achieved. In addition, a training sample carrying a label is formed by automatically calculating a sample label generation mode combined with expert correction, so that the workload of sample data labeling can be effectively reduced, and the method is flexible and efficient. The generated training samples can be used for training the formation target capacity assessment model in various scenes so as to improve the performance of the assessment model.
The following describes the technical scheme of the application in detail in connection with an airplane formation scene:
1. and constructing an airplane formation knowledge graph.
The knowledge graph constructed for expressing the capability of aircraft formation can be expressed asWherein->For the vertex set of the knowledge graph, < > for>Is the edge set of the knowledge graph.
Vertices can be divided into three types by the different entities represented: formation type collectionExecution unit set->And a set of mounted device groups->The method comprises the following steps:
as shown in fig. 2, it is assumed that there are 4 formation types, which are encoded sequentially, for exampleRepresenting a "wedge team", ">Representing "echelon", -a->Representing "team", "A>Representing "horizontal team", a set of team types +.>。
In the same way, can obtainSetting +.>Is->Respectively corresponding to the I type early warning machine 1 frame and the II type early warning machine 1 frame>Respectively corresponding to a type I fighter plane 2 frame, a type I fighter plane 4 frame, a type I fighter plane 8 frame, a type II fighter plane 2 frame, a type II fighter plane 4 frame and a type II fighter plane 8 frame, and the two frames are respectively>Respectively corresponding to a type I bomber 2 frame, a type I bomber 4 frame, a type II bomber 2 frame and a type II bomber 4 frame, < ->Respectively corresponding to a type I unmanned aerial vehicle 2 frame, a type I unmanned aerial vehicle 4 frame, a type II unmanned aerial vehicle 2 frame and a type II unmanned aerial vehicle 4 frame.
Vertex set of carrying device group corresponding to all different device types and numbers>The type I missile corresponds to 4 type I missiles, 8 type I missiles, 16 type I missiles, 4 type II missiles, 8 type II missiles, 16 type II missiles, 4 type III missiles, 8 type III missiles and 16 type III missiles respectively>、Is->Corresponding to I, II type and III type electronic warfare devices respectively>Corresponding to 2I-type bombs, 4I-type bombs, 8I-type bombs, 2 II-type bombs, 4 II-type bombs and 8 II-type bombs respectively,corresponds to type I, type II, type III and type IV radars, respectively, < >>Is->Corresponding to type I and type II rocket guns respectively.
Then, according to the possible formation composition relation between the formation type and the execution unit, setting a first relation edge setAs shown in table 1 below:
TABLE 1 first relationship edge formation
At the same time, a second relation edge set is set according to the possible equipment carrying relation between the execution unit and the carrying equipment groupAs shown in table 2 below:
TABLE 2 second relationship edge formation
Thus, a complete knowledge graph can be constructedThe aircraft formation capacity is expressed as shown in fig. 3.
2. And extracting the sub-atlas to form a sample and a sample set.
Constructing a sample datasetWherein->For sample x->Is the total number of samples. One sample->The construction method of (2) is as follows:
randomly select the firstSpecies formation type (+)>) For example->I.e. in the sample->Add formation type vertex->Corresponding to the formation type "column".
From and to the vertexThe set of all first relation edges of the connection +.>A plurality of strips are randomly selected to form a set +.>Add sample->For example->Wherein->。/>
Will beThe set of all execution unit vertices connected by the first relationship edge>Add sample->Representing the respective aircraft type and quantity contained in the formation, wherein
。
Slave and aggregateA set of second relation edges corresponding to all the mounting relations connected with the vertices of each execution unit>A plurality of strips are randomly selected to form a set +.>Add sample->Such asWherein->。
Will beA set of vertices of all the groups connected by the second relationship side>Add sample->Representing a respective device carried by an execution unit comprised in a formation, wherein。
Thus, a sample can be obtainedIs a knowledge graph from the form>As shown in figure 4.
3. And marking the sample label to obtain a training sample.
For the sampleFirstly, grading three capacities of attack, defense and reconnaissance according to airplane formation types, contained airplane types, quantity and conditions of carried equipment by utilizing a pre-training model based on an analytic hierarchy process to obtain grading results. Wherein (1)>For sample->D dimension score of (e.g.)Representative attack capacity is 8.8 minutes, defense capacity is 6.4 minutes, and reconnaissance capacity is 5.6 minutes.
Then, expert based on knowledge and experience pairsIs corrected to obtain the final score +.>. Wherein (1)>For sample->Final scoring of the d-th dimension of (e.g.)Representing that the attack capacity after expert correction is 8.4 points, the defending capacity is 6.0 points and the reconnaissance capacity is 5.4 points.
And taking the final score of the sample as a score label of the sample, and obtaining the training sample carrying the label. On the basis, the training sample set can be formed by circularly extracting 1000 times.
Based on the same inventive concept, an embodiment of the application provides a training sample generating device based on a knowledge graph. Referring to fig. 5, fig. 5 is a schematic structural diagram of a training sample generating device based on a knowledge-graph according to an embodiment of the present application, including:
the knowledge graph construction module is used for encoding the formation type, the execution unit and the carrying device group and constructing a knowledge graph for expressing formation capacity according to the corresponding relation among the formation type, the execution unit and the carrying device group;
the sub-map extraction module is used for randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to a sub-map respectively, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group;
the training sample generation module is used for determining the scoring label of each sample, obtaining a training sample carrying the label and generating a training sample set.
Optionally, the knowledge graph construction module includes:
the vertex set generation submodule is used for respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and combining the formation type set, the execution unit set and the carrying device group set to obtain a vertex set;
the first relation set generating sub-module is used for constructing a first relation edge set according to the formation composition relation between the formation type and the execution unit;
a second relation set generating sub-module, configured to construct a second relation edge set according to the device carrying relation between the execution unit and the carrying device group;
the edge set generating submodule is used for merging the first relation edge set and the second relation edge set to obtain an edge set;
and the construction submodule is used for constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
Optionally, the sub-spectrum extraction module includes:
a target formation type selection sub-module for selecting a target formation type from the set of formation types;
the first relation edge selection submodule is used for randomly selecting a plurality of first relation edges related to the target formation type from the first relation edge set and determining a corresponding target execution unit according to the first relation edges;
the second relation edge selection sub-module is used for selecting at least one second relation edge related to the target execution unit from the second relation edge set for each selected target execution unit, and determining a corresponding target carrying device group according to the second relation edge;
the sample generation sub-module is used for constructing and forming the sample based on the target formation type, the first relation side, the target execution unit, the second relation side and the target carrying device group.
Optionally, the training sample generating module includes:
the first scoring module is used for determining attack scores, defense scores and reconnaissance scores of each sample by using an analytic hierarchy process to obtain a scoring result corresponding to each sample;
the first label marking module is used for taking the grading result of each sample as the grading label corresponding to the sample to obtain a training sample carrying the label.
Optionally, the step of determining the scoring label of each sample to obtain a training sample carrying the label includes:
the second scoring module is used for scoring the attack capability, the defending capability and the reconnaissance capability of each sample by utilizing the pre-training model to obtain a multi-dimensional initial score of each sample;
the grading correction sub-module is used for correcting the multi-dimensional initial grading based on expert experience to obtain the final grading of each sample;
and the second label marking module is used for taking the final score of each sample as the scoring label of the sample to obtain a training sample carrying the label.
For the device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments for relevant points.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described by differences from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other.
It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the application may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the application.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.
The method and the device for generating training samples based on the knowledge graph provided by the application are described in detail, and specific examples are applied to the description of the principle and the implementation mode of the application, and the description of the examples is only used for helping to understand the method and the core idea of the application; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.
Claims (10)
1. A training sample generation method based on a knowledge graph, the method comprising:
encoding formation types, execution units and carrying device groups, and constructing a knowledge graph for expressing formation capacity according to the corresponding relation among the formation types, the execution units and the carrying device groups;
randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to one sub-map respectively, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group;
and determining the scoring label of each sample, obtaining a training sample carrying the label, and generating a training sample set.
2. The method of claim 1, wherein the knowledge-graph comprises a set of vertices and a set of edges;
the encoding of the formation type, the execution unit and the piggyback device group comprises:
and respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and merging to obtain a vertex set.
3. The method of claim 2, wherein the forming of the set of vertices comprises:
determining the types and the quantity of the formation types, and sequentially encoding the formation types to obtain a formation type set，Wherein->Representing a set of formation types +.>The%>Vertex corresponding to->A seed formation type is used for the formation of the seed, and (2)>A number representing a formation type;
encoding each execution unit in turn to obtain an execution unit set，Wherein->Representing execution Unit set +.>The%>Vertex corresponding to->A seed Execution Unit (EU)>Representing the number of execution units;
coding the carrying device groups containing different device types and numbers in turn to obtain a carrying device group set，Wherein->Representing a set of mounted device groups->The first of (3)Vertex corresponding to->Seed carrying device group,/->Representing the number of the carried equipment groups;
based on the set of formation typesExecution unit set->And a set of mounted device groups->Generating said vertex set->，。
4. The method according to claim 2, wherein constructing a knowledge graph for expressing formation capabilities according to correspondence among the formation type, the execution unit, and the piggybacked device group includes:
constructing a first relation edge set according to formation composition relations between the formation types and the execution units;
constructing a second relation edge set according to the equipment carrying relation between the execution unit and the carrying equipment group;
combining the first relation edge set and the second relation edge set to obtain an edge set;
and constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
5. The method of claim 2, wherein the forming of the edge set comprises:
constructing a first relation edge set according to the formation composition relation between the formation type and the execution unit，Wherein->Representing formation type vertices +.>Top +.>Connection, i.e. first->The species formation type may include +.>A seed execution unit;
constructing a second relation edge set according to the device carrying relation between the execution unit and the carrying device group，Wherein->Representing execution unit vertices in the vertex setVertex of the mounted device group->Connection, i.e. first->The seed execution unit can be loaded with +.>A seed carrying device group;
based on the first relation edge setAnd a second set of relationship edges->Generating said edge set +.>，。
6. The method of claim 4, wherein forming any one of the samples comprises:
selecting a target formation type from the set of formation types;
randomly selecting a plurality of first relation edges related to the target formation type from the first relation edge set, and determining a corresponding target execution unit according to the first relation edges;
for each selected target execution unit, selecting at least one second relation edge related to the target execution unit from the second relation edge set, and determining a corresponding target carrying device group according to the second relation edge;
and constructing and forming the sample based on the target formation type, the first relation side, the target execution unit, the second relation side and the target carrying device group.
7. The method of claim 1, wherein determining the scoring tag for each sample results in a training sample carrying the tag, comprising:
determining attack scores, defense scores and reconnaissance scores of each sample by using an analytic hierarchy process to obtain a scoring result corresponding to each sample;
and taking the scoring result of each sample as a scoring label corresponding to the sample to obtain a training sample carrying the label.
8. The method of claim 1, wherein the step of determining the scoring tag for each sample to obtain training samples carrying tags comprises:
scoring the attack capability, the defending capability and the reconnaissance capability of each sample by using a pre-training model to obtain a multi-dimensional initial score of each sample;
correcting the multi-dimensional initial score based on expert experience to obtain a final score of each sample;
and taking the final score of each sample as a scoring label of the sample to obtain a training sample carrying the label.
9. A training sample generation device based on a knowledge-graph, the device comprising:
the knowledge graph construction module is used for encoding formation types, execution units and carrying device groups and constructing a knowledge graph for expressing formation capacity according to the corresponding relation among the formation types, the execution units and the carrying device groups;
the sub-map extraction module is used for randomly extracting a plurality of sub-maps from the knowledge maps to form a sample set; each sample in the sample set corresponds to one sub-map respectively, and any sample is a five-tuple formed by a formation type, a first relation side, an execution unit, a second relation side and a carrying device group;
the training sample generation module is used for determining the scoring label of each sample, obtaining a training sample carrying the label and generating a training sample set.
10. The apparatus of claim 9, wherein the knowledge-graph construction module comprises:
the vertex set generation submodule is used for respectively encoding the formation type, the execution unit and the carrying device group based on the type dimension and the number dimension to obtain a formation type set, an execution unit set and a carrying device group set, and combining the formation type set, the execution unit set and the carrying device group set to obtain a vertex set;
the first relation set generating sub-module is used for constructing a first relation edge set according to formation composition relations between the formation types and the execution units;
a second relation set generating sub-module, configured to construct a second relation edge set according to the device carrying relation between the execution unit and the carrying device group;
the edge set generation submodule is used for merging the first relation edge set and the second relation edge set to obtain an edge set;
and the construction submodule is used for constructing a knowledge graph for expressing the formation capability based on the vertex set and the edge set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311472158.4A CN117196027B (en) | 2023-11-07 | 2023-11-07 | Training sample generation method and device based on knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311472158.4A CN117196027B (en) | 2023-11-07 | 2023-11-07 | Training sample generation method and device based on knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117196027A true CN117196027A (en) | 2023-12-08 |
| CN117196027B CN117196027B (en) | 2024-02-02 |
Family
ID=89003816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311472158.4A Active CN117196027B (en) | 2023-11-07 | 2023-11-07 | Training sample generation method and device based on knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117196027B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021189971A1 (en) * | 2020-10-26 | 2021-09-30 | 平安科技（深圳）有限公司 | Medical plan recommendation system and method based on knowledge graph representation learning |
| CN115114421A (en) * | 2022-06-21 | 2022-09-27 | 青岛海信网络科技股份有限公司 | Question-answer model training method |
| CN115186819A (en) * | 2022-06-22 | 2022-10-14 | 电子科技大学 | Knowledge graph completion method combining sub-graph structure and relationship type |
| CN116502711A (en) * | 2023-06-27 | 2023-07-28 | 北京智谱华章科技有限公司 | Knowledge graph construction and dynamic expansion method, device, equipment and medium |
| CN116541535A (en) * | 2023-05-19 | 2023-08-04 | 北京理工大学 | Automatic knowledge graph construction method, system, equipment and medium |
| WO2023168601A1 (en) * | 2022-03-08 | 2023-09-14 | 京东方科技集团股份有限公司 | Method and apparatus for training natural language processing model, and storage medium |
-
2023
- 2023-11-07 CN CN202311472158.4A patent/CN117196027B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021189971A1 (en) * | 2020-10-26 | 2021-09-30 | 平安科技（深圳）有限公司 | Medical plan recommendation system and method based on knowledge graph representation learning |
| WO2023168601A1 (en) * | 2022-03-08 | 2023-09-14 | 京东方科技集团股份有限公司 | Method and apparatus for training natural language processing model, and storage medium |
| CN115114421A (en) * | 2022-06-21 | 2022-09-27 | 青岛海信网络科技股份有限公司 | Question-answer model training method |
| CN115186819A (en) * | 2022-06-22 | 2022-10-14 | 电子科技大学 | Knowledge graph completion method combining sub-graph structure and relationship type |
| CN116541535A (en) * | 2023-05-19 | 2023-08-04 | 北京理工大学 | Automatic knowledge graph construction method, system, equipment and medium |
| CN116502711A (en) * | 2023-06-27 | 2023-07-28 | 北京智谱华章科技有限公司 | Knowledge graph construction and dynamic expansion method, device, equipment and medium |
Non-Patent Citations (1)
| Title |
|---|
| 乔振浩;车万翔;刘挺;: "基于问题生成的知识图谱问答方法", 智能计算机与应用, no. 05 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117196027B (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ackermann et al. | A statistical approach to recognizing source classes for unassociated sources in the first Fermi-LAT catalog | |
| CN110061870B (en) | Node and edge-based combined efficiency evaluation method in tactical internet | |
| CN103984746B (en) | Based on the SAR image recognition methodss that semisupervised classification and region distance are estimated | |
| CN106033443A (en) | Method and device for expansion query in vehicle retrieval | |
| Mazeika et al. | The trojan detection challenge | |
| CN105160580A (en) | Symbol network structure balance of multi-objective particle swarm optimization based on decomposition | |
| CN117196027B (en) | Training sample generation method and device based on knowledge graph | |
| Tecott Metz et al. | The case for campaign analysis: A method for studying military operations | |
| CN108322478B (en) | Attack and defense game-based website defense strategy selection method | |
| US20120084249A1 (en) | Method for pollen-based geolocation | |
| CN110852574B (en) | Target threat assessment method and medium based on improved grey target theory | |
| CN117611018A (en) | Effectiveness deduction evaluation method based on unmanned cluster scale and index dynamic adjustment | |
| CN112818809A (en) | Method, device and storage medium for detecting image information | |
| CN107688582B (en) | Resource recommendation model obtaining method and device | |
| CN115269571A (en) | Data quality evaluation method based on data processing | |
| Verma et al. | Using AI/ML to predict perpetrators for terrorist incidents | |
| CN109145518B (en) | Method for constructing reliability decision graph model of large-scale complex equipment | |
| May et al. | An Automated RV Evaluation Procedure1 | |
| Zhang | Data Driven Traffic Accidents Visualization and Root Cause Quantification with Explainable Artificial Intelligence for improved Road Safety and Urban Planning | |
| CN118133958B (en) | Military combat command system and method based on augmented reality and knowledge graph | |
| CN108764578B (en) | Wind power plant macroscopic intelligent site selection method combining Monte Carlo simulation and analytic hierarchy process | |
| Mu et al. | A memetic algorithm using local structural information for detecting community structure in complex networks | |
| Bo et al. | Research on assessment of anti-terrorist combat capability based on cloud theory | |
| CN111488893B (en) | Image classification method and device | |
| Wan et al. | Link Analysis of Conflict through Signed Spectral Embedding: The Russo-Ukrainian War |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |