CN116561385B - Knowledge representation-based plan quick matching recommendation method - Google Patents
Knowledge representation-based plan quick matching recommendation method Download PDFInfo
- Publication number
- CN116561385B CN116561385B CN202310836284.7A CN202310836284A CN116561385B CN 116561385 B CN116561385 B CN 116561385B CN 202310836284 A CN202310836284 A CN 202310836284A CN 116561385 B CN116561385 B CN 116561385B
- Authority
- CN
- China
- Prior art keywords
- scheme
- node
- scene
- matching
- neighbors
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000004364 calculation method Methods 0.000 claims abstract description 44
- 238000010586 diagram Methods 0.000 claims abstract description 15
- 238000005096 rolling process Methods 0.000 claims abstract description 12
- 238000004220 aggregation Methods 0.000 claims description 2
- 230000002776 aggregation Effects 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 abstract description 9
- 238000013528 artificial neural network Methods 0.000 abstract description 6
- 238000012163 sequencing technique Methods 0.000 abstract description 4
- 230000004044 response Effects 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/80—Information retrieval; Database structures therefor; File system structures therefor of semi-structured data, e.g. markup language structured data such as SGML, XML or HTML
- G06F16/83—Querying
- G06F16/835—Query processing
- G06F16/8373—Query execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a knowledge representation-based plan quick matching recommendation method, and belongs to the technical field of matching calculation. The method is used for rapidly matching and recommending the emergency scheme according to the scene situation under the scenes such as emergency disasters or accidents. According to the method, on the basis of defining task scene characteristics and scheme characteristics, an emergency scheme recommendation knowledge representation diagram is constructed, a graph convolution network training scene and scheme embedding characterization are adopted, a scene and scheme matching calculation model is obtained, and matching calculation and sequencing recommendation of an emergency scheme are provided for the task scene. The scene-scheme matching calculation model training method based on the graph rolling network can ensure the scheme matching accuracy and the recommendation efficiency; compared with a scheme recommendation method of a neural network, the emergency scheme recommendation method based on the knowledge representation has better interpretability, and supports the expansion of the knowledge representation diagram, so that the updating of a matching calculation model of a scene and a scheme is supported.
Description
Technical Field
The invention belongs to the technical field of matching calculation, and particularly relates to a knowledge representation-based rapid plan matching recommendation method.
Background
The emergency plan is a formulated emergency management, command and rescue plan when facing sudden events such as natural disasters, accident disasters, public safety events and the like. Currently, emergency response mainly adopts a pre-established emergency plan, and when an emergency occurs, the emergency response is properly adjusted according to an emergency type, a hazard degree, a site environment and other condition matching schemes, and response preparation time is shortened as much as possible by selecting a proper emergency plan. Especially in emergency situations, the scheme matching method and the recommendation efficiency are important to improve the emergency response speed.
At present, a recommendation system in the commercial field mainly adopts an artificial intelligence algorithm based on a neural network, is mainly applied to the fields of electronic commerce and personalized information, and needs a large number of data samples for training, however, due to the fact that the data samples related to emergency plan recommendation are fewer, data accumulation is seriously lacking; and because the algorithm has poor interpretability, reasonable basis cannot be given to recommended scheme suggestion, so that auxiliary decision making is not convincing. Therefore, the artificial intelligence method based on the neural network in the current commercial field is difficult to support the emergency plan matching recommendation application.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a knowledge representation-based plan quick matching recommendation method.
The method comprises the following steps:
step S1, determining a task scene, task scene characteristics, a scheme and scheme characteristics, wherein the task scene is,,/>Representing task scenario->Is +.>,/>,Representation scheme->M represents a task scene feature number, and n represents a scheme feature number;
s2, constructing an undirected knowledge representation diagram based on the task scene, the task scene characteristics, the scheme and the scheme characteristics, wherein the undirected knowledge representation diagram comprises nodesThe undirected edges of the undirected knowledge representation represent the relation between neighboring nodes +.>,/>Representing the ith edge relationship;
and S3, training a graph convolution network in a matching calculation model by utilizing the undirected knowledge representation graph and optimizing the task scene and the embedded representation of the scheme, and matching a corresponding scheme for a new scene by utilizing the trained matching calculation model.
According to the method of the present invention, in said step S2, said side relationship comprises:
task scenarioIs->J task scene features of (2)>Side relation between;
Scheme for the production of a semiconductor deviceIs->Is>Side relation between->;
Task scenarioScheme->Side relation between->。
According to the method of the present invention, in the step S3, when training the graph rolling network in the matching calculation model, a matching recommendation relationship between the task scene and the solution is used as a training label, where:
when the task sceneScheme->When there is history recommendation relationship, the undirected knowledge representation graph relationshipLabel->;
Otherwise the first set of parameters is selected,,/>。
according to the method of the present invention, in said step S3, when training said graph rolling network in said matching calculation model, the number of positive and negative samples in the training data is the same, wherein:
the positive sample is characterized as;
The negative sample is characterized as。
According to the method of the present invention, in said step S3, in training said graph rolling network in said matching calculation model, said embedded token is expressed as:
wherein ,representing the current node +_>Representing an embedded representation of the current node,representing an aggregate function +.>Representing the current nodeAll of the neighboring nodes are connected to each other,representing the number of all neighbor nodes of the current node, +.>Any neighbor node representing the current node, +.>Representing embedded characterization of any neighbor node of the current node;
the neighbor node is a node within a b-layer adjacent to the current node, wherein:
wherein ,representing a distance function.
According to the method of the invention, in said step S3:
the current node is a task sceneAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
According to the method of the invention, in said step S3:
the current node is a schemeAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
According to the method of the invention, in said step S3:
the current node is a task sceneOr a schemeCorresponding featuresAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
According to the method of the invention, in said step S3:
calculating a matching score according to the embedded representation of the task scene and the embedded representation of the scheme, wherein the matching score=emmbedding #)×embedding(/>);
Taking cross entropy loss of the matching score and the training label as a loss function, loss=entry (score, label), and entry () represents cross entropy;
and reversely deriving the loss function loss, iteratively updating parameters of the matched calculation model until cut-off, and selecting the model with the minimum loss as the optimal matched calculation model.
In summary, the technical scheme provided by the invention is used for rapidly matching and recommending the emergency scheme according to the scene situation under the scenes such as emergency disasters or accidents.
According to the method, on the basis of defining task scene characteristics and scheme characteristics, an emergency scheme recommendation knowledge representation diagram is constructed, a graph convolution network training scene and scheme embedding characterization are adopted, a scene-scheme matching calculation model is obtained, and matching calculation and sequencing recommendation of an emergency scheme are provided for the task scene. The scene-scheme matching calculation model training method based on the graph rolling network can ensure the scheme matching accuracy and the recommendation efficiency; the emergency scheme recommendation based on the knowledge representation has better interpretability compared with a scheme recommendation method of a neural network, and supports the expansion of the knowledge representation diagram, thereby supporting the updating of a scene-scheme matching calculation model.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings which are required in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are some embodiments of the invention and that other drawings may be obtained from these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flow chart of a knowledge representation-based fast matching recommendation method for a plan according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The knowledge representation-based scheme quick matching recommendation method provided by the invention comprises the following steps (shown in figure 1):
step S1, determining a task scene, task scene characteristics, a scheme and scheme characteristics, wherein the task scene is,,/>Representing task scenario->Is +.>,/>,Representation scheme->M represents a task scene feature number, and n represents a scheme feature number;
s2, constructing an undirected knowledge representation diagram based on the task scene, the task scene characteristics, the scheme and the scheme characteristics, wherein the undirected knowledge representation diagram comprises nodesThe undirected edges of the undirected knowledge representation represent the relation between neighboring nodes +.>,/>Representing the ith edge relationship;
and S3, training a graph convolution network in a matching calculation model by utilizing the undirected knowledge representation graph and optimizing the task scene and the embedded representation of the scheme, and matching a corresponding scheme for a new scene by utilizing the trained matching calculation model.
Specifically, the method comprises the steps of firstly extracting task scene characteristics and plan characteristics, establishing a task scene and plan knowledge representation diagram, adopting a graph convolution network training scene and a plan embedding representation to obtain a task scene-plan matching calculation model for emergency plan sequencing recommendation, and recording a history plan recommendation result for enriching the knowledge representation diagram.
Specifically, in step S1, task scenario features and scenario features are defined: the task scene is,,/>Representing task scenario->Is the jth feature of (2). For example, the characteristics of the accident disaster rescue task scene include: accident area, number of accident objects, accident object location, degree of emergency, field environment, etc. The scheme is->,,/>Representation scheme->Is the jth feature of (2). For example, the features of an accident disaster relief scheme include: rescue areas, rescue objects, the number of rescue objects, the degree of emergency, the number of rescue forces, the number of rescue supplies, and the like. m represents a task scene feature number, and n represents a scheme feature number.
In some embodiments, in the step S2, the side relationship includes:
task scenarioIs->J task scene features of (2)>Side relation between;
Scheme for the production of a semiconductor deviceIs->Is>Side relation between->;
Task scenarioScheme->Side relation between->。
Specifically, a undirected knowledge representation is constructed according to the task scene and the scheme characteristics: nodes included in the knowledge representation. Undirected edges in the knowledge representation represent the relation between neighboring nodes +.>The side relationships include three classes: 1) Task scenario feature->Representing the relation between task scenario and features, task scenario +.>Has the characteristics->The method comprises the steps of carrying out a first treatment on the surface of the 2) Scheme-feature->Representing the relation between the scheme and the features, scheme->Has the characteristics->The method comprises the steps of carrying out a first treatment on the surface of the 3) Scene-scheme->Representing the selection relation between the scene and the scheme, in the scene +.>Lower selection scheme->。
In some embodiments, in the step S3, in training the graph rolling network in the matching calculation model, a matching recommendation relationship between the task scene and the solution is used as a training label, where:
when the task sceneScheme->When there is history recommendation relationship, the undirected knowledge representation graph relationshipLabel->;
Otherwise the first set of parameters is selected,,/>。
specifically, according to the knowledge representation graph structure, training a graph convolution network, optimizing embedded characterization of task scenes and scheme nodes, and meeting the matching relation of scenes and schemes to obtain a scene-scheme matching calculation model. Taking the matching recommendation relationship between the scene and the scheme as a training label: when the task sceneScheme->When there is history recommendation relationship, it is expressed as +.>Then->The method comprises the steps of carrying out a first treatment on the surface of the Otherwise, the scene has no recommendation relation with the scheme in history, and the scene is expressed as +.>Then->。
In some embodiments, in the step S3, in training the graph rolling network in the matching calculation model, the number of positive and negative samples in training data is the same, wherein:
the positive sample is characterized as;
The negative sample is characterized as。
Specifically, the positive sample and the negative sample are balanced: considering that the scene-scheme matching relation in the existing data is usually less, the number of positive samples and the number of negative samples in the training data are less, and the negative samples are sampled, so that the number of the positive samples and the number of the negative samples are similar. The positive sample consists of scene nodes and scheme nodes with matching relations:
the negative sample consists of scene nodes and scheme node combinations which have no recommendation relationship in history:
in some embodiments, in the step S3, in training the graph rolling network in the matching calculation model, the embedded token is expressed as:
wherein ,representing the current node of the network,/>representing an embedded representation of the current node,representing an aggregate function +.>Representing all the neighboring nodes of the current node,representing the number of all neighbor nodes of the current node, +.>Any neighbor node representing the current node, +.>Representing embedded characterization of any neighbor node of the current node;
the neighbor node is a node within a b-layer adjacent to the current node, wherein:
wherein ,representing a distance function.
Specifically, node embedding characterization is calculated according to the knowledge representation, and node embedding characterization is obtained through aggregation of neighbor node calculation:
considering the b-layer neighbor node, the b-layer neighbor node can be calculated through the distance between the nodes in the knowledge representation graphThe first can be calculated by layert layer (t is more than or equal to 0)<b) And the neighbor nodes of the nodes obtain the node of the t+1st layer.
In some embodiments, in said step S3:
the current node is a task sceneAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
In some embodiments, in said step S3:
the current node is a schemeAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
In some embodiments, in said step S3:
the current node is a task sceneOr a schemeCorresponding featuresAnd b=2:
wherein ,,/>representing the current node +.>Layer 1 neighbor node of->,/>Representing the current node +.>Is a layer 2 neighbor node->。
Specifically, considering layer 2 neighbor nodes in general, the computation of neighbor can be divided into:
for scene nodes: scheme with layer 1 neighbor nodes as scene feature nodes or recommended nodesThe 2 nd layer neighbor node is a scheme/scene node adjacent to the scene feature node or a feature node adjacent to the recommended scheme node,/>。
Aiming at scheme nodes: layer 1 neighbor nodes are scheme feature nodes or matched scenesThe layer 2 neighbor node is a scheme/scene node adjacent to the scheme feature node or a feature node adjacent to the matched scene node,/>。
For characteristic nodes: layer 1 neighbor nodes are scenes or schemesLayer 2 neighbor node is a neighbor node of scene or scheme +.>,/>。
In some embodiments, in said step S3:
calculating a matching score according to the embedded representation of the task scene and the embedded representation of the scheme, wherein the matching score=emmbedding #)×embedding(/>);
Taking cross entropy loss of the matching score and the training label as a loss function, loss=entry (score, label), and entry () represents cross entropy;
and reversely deriving the loss function loss, iteratively updating parameters of the matched calculation model until cut-off, and selecting the model with the minimum loss as the optimal matched calculation model.
Specifically, a matching score is calculated from the calculated scheme embedded representation and the scene embedded representation,. Cross entropy of score and label less = enter (score, label) was used as the loss function. And (3) carrying out inverse derivation on the loss function, namely backward (), and carrying out iterative updating on the model parameters. Repeating the steps until the termination condition is met (for example, the specified iteration times are reached), and selecting the model with the smallest loss/highest precision as the optimal model.
Specifically, new scene embedding characterization calculation: for the new scene C, generating features of the new scene, such as C= { accident area, accident object number, accident object position and emergency … } according to the scene description, establishing an adjacent relation with feature nodes according to the features of the new scene, and calculating embedded representation empedding (C) of the new scene by using the graph rolling network model obtained in the step three.
Scheme matching calculation: for all schemes in the scheme library, calculating the matching degree of the new scene C and Pi。
The proposal recommends: ranking score (C, pi) by score, ranking higher scoring schemes by rankThe scene is recommended as a recommendation protocol.
Knowledge representation graph update: and adding the new scene of the selected plan into the history recommendation relationship, and updating the graph structure. For scene featuresAdding an edge +_in the knowledge representation graph>The method comprises the steps of carrying out a first treatment on the surface of the For recommended plans->Adding an edge +_in the knowledge representation graph>。
Scene-scheme matching calculation model update: when the change of the knowledge representation exceeds a specified threshold, e.g. when the number of edges changes by more than a certain proportionAnd updating the scene-scheme matching calculation model by utilizing the step three.
In summary, the technical scheme provided by the invention is used for rapidly matching and recommending the emergency scheme according to the scene situation under the scenes such as emergency disasters or accidents. According to the method, on the basis of defining task scene characteristics and scheme characteristics, an emergency scheme recommendation knowledge representation diagram is constructed, a graph convolution network training scene and scheme embedding characterization are adopted, a scene-scheme matching calculation model is obtained, and matching calculation and sequencing recommendation of an emergency scheme are provided for the task scene. The scene-scheme matching calculation model training method based on the graph rolling network can ensure the scheme matching accuracy and the recommendation efficiency; the emergency scheme recommendation based on the knowledge representation has better interpretability compared with a scheme recommendation method of a neural network, and supports the expansion of the knowledge representation diagram, thereby supporting the updating of a scene-scheme matching calculation model.
Compared with the plan recommendation method based on the neural network, the technical scheme provided by the invention can verify the rationality of plan recommendation through the connection relation between the nodes in the knowledge representation diagram, and has better interpretability. According to the scheme, the cold start problem can be better processed, a new scene which does not appear in the training process can be obtained, and the plan can be recommended according to scene characteristics. The method can support expansion of knowledge representation graphs, and the knowledge representation graphs are continuously enriched along with gradual collection of data, so that a scene-scheme matching calculation model is further optimized. Compared with a recommendation method based on a knowledge graph alone, the scene-plan matching calculation model training method based on the graph rolling network can further ensure the accuracy rate and the recommendation efficiency of plan matching.
Claims (6)
1. A knowledge representation-based rapid matching recommendation method for a plan, the method comprising:
step S1, determining a task scene, task scene characteristics, a scheme and scheme characteristics, wherein the task scene is C i , Representing task scenario C i The j-th feature of the scheme is P i ,/>Representation scheme P i Is the first of (2)j features, m represents a task scene feature number, and n represents a scheme feature number;
the task scene refers to an accident disaster rescue task scene, and the task scene characteristics of the accident disaster rescue task scene comprise an accident area, the number of accident objects, the positions of the accident objects, the emergency degree and the site environment;
the scheme refers to an accident disaster rescue scheme, and the scheme characteristics of the accident disaster rescue scheme comprise rescue areas, rescue objects, the number of the rescue objects, the degree of emergency, the number of rescue forces and the number of rescue supplies;
s2, constructing an undirected knowledge representation diagram based on the task scene, the task scene characteristics, the scheme and the scheme characteristics, wherein the undirected knowledge representation diagram comprises nodesThe undirected edges of the undirected knowledge representation represent the relationship e= { E between neighboring nodes i },e i Representing the ith edge relationship;
s3, training a graph convolution network in a matching calculation model by utilizing the undirected knowledge representation graph and optimizing the task scene and the embedded representation of the scheme, and matching a corresponding scheme for a new scene by utilizing the trained matching calculation model;
in the step S3, in training the graph rolling network in the matching calculation model, the embedded token is expressed as:
wherein node represents the current node, embedding (node) represents the embedded representation of the current node, aggregation () represents the aggregate function, neighbors (node) represents all neighbor nodes of the current node, | neighbors (node) | represents all neighbor node numbers of the current node, node 'represents any neighbor node of the current node, and ebedding (node') represents the embedded representation of any neighbor node of the current node;
the neighbor node is a node within a b-layer adjacent to the current node, wherein:
neighbors(node)={node′|dist(node,node′)≤b}
wherein dist () represents a distance function;
in the step S3:
the current node is task scene C i And b=2:
wherein ,P i ∈neighbors(C i ,1),neighbors(C i 1) represents the current node C i Is a layer 1 neighbor node of (c),neighbors(C i 2) represents the current node C i Is a layer 2 neighbor node->
In the step S3:
the current node is scheme P i And b=2:
wherein ,C i ∈neighbors(P i ,1),neighbors(P i 1) represents the current node P i Is a layer 1 neighbor node of (c),neighbors(P i 2) represents the current node P i Is a layer 2 neighbor node->
2. The knowledge representation-based rapid matching recommendation method for a protocol of claim 1, wherein in said step S2, said side relationship comprises:
task scenario C i And the task scene C i J-th task scenario feature of (2)Side relation between->
Scheme P i With the scheme P i Features of the j-th aspect of (2)Side relation between->
Task scenario C i Scheme P i Side relation e between i =(C i ,P i )。
3. The knowledge representation-based rapid matching recommendation method for a protocol according to claim 2, wherein in the step S3, when training the graph convolution network in the matching calculation model, a matching recommendation relationship between the task scene and the scheme is used as a training label, wherein:
when task scene C i Scheme P i When there is a history recommendation relationship, the relationship E (C) of the undirected knowledge representation i ,P i ) =1, tag label (C i ,P i )=1;
Otherwise, E (C i ,P i )=0,label(C i ,P i )=0。
4. A knowledge representation based protocol fast matching recommendation method according to claim 3, wherein in said step S3, in training said graph-convolution network in said matching calculation model, the number of positive and negative samples in the training data is the same, wherein:
the positive samples are characterized as pos_train_data= { (C) i ,P i )|label(C i ,P i )==1};
The negative samples are characterized as neg_train_data= { (C) i ,P i )|label(C i ,P i )==0}。
5. The knowledge representation-based rapid matching recommendation method for a protocol of claim 4, wherein in said step S3:
the current node is task scene C i Or scheme P i In the case of the corresponding feature fi and b=2:
neighbors(node)=neighbors(f i ,1)+neighbors(f i ,2)
={C i |E(C i ,f i )=1}∪{P i |E(f i ,P i )=1}+neighbors(C i ,1)∪neighbors(P i ,1)
wherein ,Ci ,P i ∈neighbors(f i ,1),neighbors(f i 1) represents the current node f i Layer 1 neighbor nodes of (f) i ,1)={C i |E(C i ,f i )=1}∪{P i |E(f i ,P i )=1},neighbors(f i 2) represents the current node f i Layer 2 neighbor nodes of (f) i ,2)=neighbors(C i ,1)∪neighbors(P i ,1)。
6. The knowledge representation-based rapid matching recommendation method for a protocol of claim 5, wherein in said step S3:
calculating a matching score from the embedded representation of the task scenario and the embedded representation of the solution, wherein the matching score = enabling (C i )×embedding(P i );
Taking cross entropy loss of the matching score and the training label as a loss function, loss=entry (score, label), and entry () represents cross entropy;
and reversely deriving the loss function loss, iteratively updating parameters of the matched calculation model until cut-off, and selecting the model with the minimum loss as the optimal matched calculation model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310836284.7A CN116561385B (en) | 2023-07-10 | 2023-07-10 | Knowledge representation-based plan quick matching recommendation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310836284.7A CN116561385B (en) | 2023-07-10 | 2023-07-10 | Knowledge representation-based plan quick matching recommendation method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116561385A CN116561385A (en) | 2023-08-08 |
CN116561385B true CN116561385B (en) | 2023-10-13 |
Family
ID=87500474
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310836284.7A Active CN116561385B (en) | 2023-07-10 | 2023-07-10 | Knowledge representation-based plan quick matching recommendation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116561385B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114117032A (en) * | 2021-12-08 | 2022-03-01 | 中国民航科学技术研究院 | Method and device for generating plan based on real-time emergency data and electronic equipment |
CN114491082A (en) * | 2022-03-31 | 2022-05-13 | 南京众智维信息科技有限公司 | Plan matching method based on network security emergency response knowledge graph feature extraction |
WO2023123624A1 (en) * | 2021-12-31 | 2023-07-06 | 中国科学院深圳先进技术研究院 | Method and system for predicting influenza outbreak trend in city, and terminal and storage medium |
-
2023
- 2023-07-10 CN CN202310836284.7A patent/CN116561385B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114117032A (en) * | 2021-12-08 | 2022-03-01 | 中国民航科学技术研究院 | Method and device for generating plan based on real-time emergency data and electronic equipment |
WO2023123624A1 (en) * | 2021-12-31 | 2023-07-06 | 中国科学院深圳先进技术研究院 | Method and system for predicting influenza outbreak trend in city, and terminal and storage medium |
CN114491082A (en) * | 2022-03-31 | 2022-05-13 | 南京众智维信息科技有限公司 | Plan matching method based on network security emergency response knowledge graph feature extraction |
Non-Patent Citations (3)
Title |
---|
KEVIN Cheng 等."An Assisted-Teaching Language Platform Based On Internet".《Proceedings of 2009 IEEE International Symposium on IT in Medicine & Education(ITME2009)》.2009,全文. * |
侯位昭 等."融合知识图谱及贝叶斯网络的智能推荐方法".《中国电子科学研究院学报》.2020,第15卷(第5期),全文. * |
王骏东 等."基于知识图谱的配电网故障辅助决策研究".《电网技术》.2021,第45卷(第6期),全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN116561385A (en) | 2023-08-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107529651B (en) | Urban traffic passenger flow prediction method and equipment based on deep learning | |
CN109117987B (en) | Personalized traffic accident risk prediction recommendation method based on deep learning | |
CN114120102A (en) | Boundary-optimized remote sensing image semantic segmentation method, device, equipment and medium | |
CN108876796A (en) | A kind of lane segmentation system and method based on full convolutional neural networks and condition random field | |
CN108961429B (en) | Automatic segmentation and splicing method for cultural relic fragment model | |
CN109753959B (en) | Road traffic sign detection method based on self-adaptive multi-scale feature fusion | |
CN109214422B (en) | Parking data repairing method, device, equipment and storage medium based on DCGAN | |
CN112685504B (en) | Production process-oriented distributed migration chart learning method | |
CN112396587B (en) | Method for detecting congestion degree in bus compartment based on collaborative training and density map | |
CN111814658B (en) | Scene semantic structure diagram retrieval method based on semantics | |
CN104268140A (en) | Image retrieval method based on weight learning hypergraphs and multivariate information combination | |
CN114358389A (en) | Short-term power load prediction method combining VMD decomposition and time convolution network | |
CN114565628B (en) | Image segmentation method and system based on boundary perception attention | |
CN115761735A (en) | Semi-supervised semantic segmentation method based on self-adaptive pseudo label correction | |
CN116662468A (en) | Urban functional area identification method and system based on geographic object space mode characteristics | |
CN111858565A (en) | Multi-dimensional high-precision track intelligent prediction method based on segment clustering | |
CN116561385B (en) | Knowledge representation-based plan quick matching recommendation method | |
CN114037014A (en) | Reference network clustering method based on graph self-encoder | |
Verma et al. | A systematic review on the advancement in the study of fuzzy variational problems | |
Blier-Wong et al. | Rethinking representations in P&C actuarial science with deep neural networks | |
CN110109005B (en) | Analog circuit fault testing method based on sequential testing | |
Zhang et al. | Automatic identification of building structure types using unmanned aerial vehicle oblique images and deep learning considering facade prior knowledge | |
CN115762147A (en) | Traffic flow prediction method based on adaptive graph attention neural network | |
CN116030255A (en) | System and method for three-dimensional point cloud semantic segmentation | |
CN111324641B (en) | Personnel estimation method and device, computer-readable storage medium and terminal equipment |
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 |