CN113626612A - Prediction method and system based on knowledge graph reasoning - Google Patents
Prediction method and system based on knowledge graph reasoning Download PDFInfo
- Publication number
- CN113626612A CN113626612A CN202110931960.XA CN202110931960A CN113626612A CN 113626612 A CN113626612 A CN 113626612A CN 202110931960 A CN202110931960 A CN 202110931960A CN 113626612 A CN113626612 A CN 113626612A
- Authority
- CN
- China
- Prior art keywords
- graph
- layer
- candidate
- query
- knowledge
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000003190 augmentative effect Effects 0.000 claims abstract description 98
- 230000014509 gene expression Effects 0.000 claims abstract description 69
- 230000004044 response Effects 0.000 claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims description 33
- 238000003860 storage Methods 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000002776 aggregation Effects 0.000 claims description 7
- 238000004220 aggregation Methods 0.000 claims description 7
- 230000003416 augmentation Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 description 26
- 230000006870 function Effects 0.000 description 23
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 239000000969 carrier Substances 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000001939 inductive effect Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 241001235534 Graphis <ascomycete fungus> Species 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Animal Behavior & Ethology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A prediction method and system based on knowledge-graph reasoning are provided, the method comprising: augmenting a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph, the lengths of which do not exceed L, are augmented to a length L in the augmented knowledge graph; obtaining an expression of a relationship directed graph about a query based on the augmented knowledge graph, wherein the query comprises a query entity, a query relationship and a response entity, and the relationship directed graph about the query comprises candidate relationship directed graphs for different candidate response entities; establishing a loss function for the knowledge-graph based on the expression of the relational directed graph; training a graph network model of the knowledge-graph based on the loss function; predicting responsive entities in the query using the trained graph network model.
Description
Technical Field
The present application relates to knowledge graph embedding technology in the field of artificial intelligence, and more particularly, to a prediction method and system based on knowledge graph reasoning.
Background
With the rapid development of information network technology, various information network data contents present an explosive growth situation. Such content is generally characterized by large scale, heterogeneous diversity and loose organization structure, and provides challenges for people to effectively acquire information and knowledge. Knowledge map (KG) is a Knowledge base of semantic network, and can describe Knowledge resources and carriers thereof by using visualization technology, and mine, analyze, construct, draw and display Knowledge and mutual relations among Knowledge resources and carriers.
The knowledge graph is a special graph structure and comprises entities (entity) and relations (relation), wherein the entities are used as nodes, and the relations are used as directed edges. In a knowledge graph, each edge may be represented as a triple (h, r, t) in the form of (head entity, relationship, tail entity) to indicate that two entities h (i.e., head entity) and t (i.e., tail entity) are connected by a relationship r, e.g., (new york, isLocatedIn, USA) may represent that new york is located in USA.
Knowledge Graph Reasoning (KGR) aims to deduce unknown facts from existing facts (facts) using the information already in the Graph. For a predicted fact, it is usually necessary to make inferences using semantic or structural information around the nodes and relationships it contains. The reasoning tasks in the graph can be divided into two types, namely, transformation reasoning (transforming reasoning), namely performing factual reasoning in the existing node range, and inductive reasoning (inductive reasoning), namely performing factual reasoning on unknown nodes in the training phase.
The prior art typically employs a relational path-based approach to reasoning, however, paths are limited in capturing complex topologies in a knowledge graph. In addition, the knowledge graph embedding model in the prior art generally only utilizes semantic information, and each node needs to learn one embedding (embedding) independently, and for a new node in a prediction stage, because the training stage does not learn the corresponding embedding, induction and deduction in such a scene cannot be performed, and meanwhile, an inference structure cannot be explained. In addition, although research shows that generalized derivation can be performed by using rules based on relationships (prediction of different nodes can be extended because the rules are not limited by specific nodes), the graph structure information is not used, the important properties of the original graph are lost, and the conversion reasoning task is general. Therefore, there is a need for an efficient prediction method based on knowledge-graph reasoning that can fully preserve the structural information of the knowledge-graph.
Disclosure of Invention
According to an embodiment of the present invention, there is provided a prediction method based on knowledge-graph reasoning, the method including: augmenting a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes connected by the edges and are directed edges, and relationship paths in the knowledge graph, the lengths of which do not exceed L, are augmented to a length L in the augmented knowledge graph; obtaining an expression of a relational directed graph about a query based on the augmented knowledge graph, wherein the query comprises a query entity, a query relation and a response entity, and the relational directed graph about the query comprises candidate relational directed graphs for different candidate response entities; establishing a loss function for a graph network model of the knowledge-graph based on the expression of the relational directed graph; training a graph network model of the knowledge-graph based on the loss function; predicting response entities in the query using the trained graph network model; the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a convergence point located at the L th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the convergence point, and any path from the source node to the convergence point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
Optionally, the step of augmenting the knowledge-graph may comprise: adding a self-directed relationship to each node in the knowledge graph, and adding a corresponding edge presenting an opposite relationship to each edge in the knowledge graph, so that a relationship path of which the length does not exceed the length L in the knowledge graph is extended to the length L in the augmented knowledge graph.
Optionally, the step of obtaining the expression of the relational directed graph with respect to the query may comprise: obtaining at least one candidate triplet for the query by setting an entity in the augmented knowledge graph as a candidate answer entity for the query; obtaining a representation of a candidate relationship directed graph for each candidate triplet based on the augmented knowledge-graph.
Optionally, the step of obtaining the representation of the candidate relationship directed graph with respect to each candidate triplet may comprise: setting a query entity in the candidate triple as a source node of the candidate relationship directed graph and setting a candidate response entity in the candidate triple as a convergence point of the candidate relationship directed graph; extracting 1-order neighbor nodes of the source nodes, L-order neighbor nodes and corresponding edges layer by layer from the 1 st layer until the L-th layer based on the augmented knowledge graph; if the extracted neighbor nodes in the L-th layer do not comprise the rendezvous point, setting the expression of the candidate relation directed graph related to the candidate triple as a preset vector; if the extracted neighbor nodes in the L-th layer comprise the convergent point, extracting nodes and edges of each layer in the candidate relationship directed graph layer by layer from the convergent point of the L-th layer until the layer 1 based on the 1 st order neighbor node to the L-order neighbor node and the corresponding edge of the extracted source node; establishing a candidate relation directed graph of the candidate triple based on the nodes and edges of each layer in the extracted candidate relation directed graph, and calculating the expression of the nodes of each layer from the 1 st layer to the L th layer of the candidate relation directed graph through a graph neural network; and determining the expression of the aggregation point of the L-th layer obtained by calculation as a final expression of the candidate relation directed graph related to the candidate triple.
Optionally, the step of obtaining the expression of the relational directed graph with respect to the query may comprise: setting the query entity in the query as a source node of a relationship directed graph; extracting neighbor nodes and corresponding edges of source nodes on each layer from the layer 1 on the basis of the augmented knowledge graph layer by layer, and calculating the expression of the neighbor nodes on each layer by layer through a graph neural network on the basis of the extracted neighbor nodes and corresponding edges of each layer until the layer L; determining neighbor nodes of the L-th layer as candidate response entities of the query, and respectively determining expressions of the neighbor nodes of the L-th layer obtained by calculation as expressions of a candidate relationship directed graph for the corresponding candidate response entities of the query; and setting the expression of the relation directed graph aiming at the nodes except the nodes of the L-th layer in the augmented knowledge graph as a preset vector.
According to another embodiment of the present invention, there is provided a prediction system based on knowledge-graph reasoning, the system including: an augmenting device configured to augment a knowledge graph to obtain an augmented knowledge graph spectrum, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph whose length does not exceed L are augmented to a length L in the augmented knowledge graph; a processing device configured to obtain, based on the augmented knowledge-graph, a representation of a relational directed graph with respect to a query, wherein the query comprises a query entity, a query relation, and a response entity, and the relational directed graph with respect to the query comprises candidate relational directed graphs for different candidate response entities; a training device configured to establish a loss function for the graph network model of the knowledge-graph based on the expression of the relational directed graph and train the graph network model of the knowledge-graph based on the loss function; the predicting device is configured to predict a response entity in the query by using the trained graph network model, wherein the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a gathering point located at the L th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the gathering point, and any path from the source node to the gathering point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
Optionally, the augmentation apparatus may be configured to: adding a self-directed relation to each node in the knowledge graph, and adding a corresponding side presenting an opposite relation to each side in the knowledge graph, so that a relation path of which the length does not exceed the length L in the knowledge graph is extended to the length L in the augmented knowledge graph, thereby obtaining the augmented knowledge graph.
Optionally, the processing apparatus may be configured to obtain the representation of the relational directed graph with respect to the query by: obtaining at least one candidate triplet for the query by setting an entity in the augmented knowledge graph as a candidate answer entity for the query; obtaining a representation of a candidate relationship directed graph for each candidate triplet based on the augmented knowledge-graph.
Optionally, the processing apparatus may be configured to obtain the representation of the candidate relationship directed graph with respect to each candidate triplet by: setting a query entity in the candidate triple as a source node of the candidate relationship directed graph and setting a candidate response entity in the candidate triple as an aggregation point of the candidate relationship directed graph; extracting 1-order neighbor nodes of the source nodes, L-order neighbor nodes and corresponding edges layer by layer from the 1 st layer until the L-th layer based on the augmented knowledge graph; if the extracted neighbor nodes in the L-th layer do not comprise the convergent point, setting the expression of the candidate relation directed graph related to the candidate triple as a preset vector; if the extracted neighbor nodes in the L-th layer comprise the convergent point, extracting nodes and edges of each layer in the candidate relationship directed graph layer by layer from the convergent point of the L-th layer until the layer 1 based on the 1 st order neighbor node to the L-order neighbor node and the corresponding edge of the extracted source node; establishing a candidate relation directed graph of the candidate triple based on the nodes and edges of each layer in the extracted candidate relation directed graph, and calculating the expression of the nodes of each layer from the 1 st layer to the L th layer of the candidate relation directed graph through a graph neural network; and determining the expression of the aggregation point of the L-th layer obtained by calculation as a final expression of the candidate relation directed graph related to the candidate triple.
Optionally, the processing apparatus may be configured to obtain the representation of the relational directed graph with respect to the query by: setting the query entity in the query as a source node of a relationship directed graph; extracting neighbor nodes and corresponding edges of source nodes on each layer from the layer 1 layer by layer based on the augmented knowledge graph, and calculating the expression of neighbor nodes on each layer by layer through a graph neural network based on the extracted neighbor nodes and corresponding edges of each layer until the L-th layer; determining neighbor nodes of the L-th layer as candidate response entities of the query, and respectively determining expressions of the neighbor nodes of the L-th layer obtained through calculation as expressions of candidate relationship directed graphs aiming at the corresponding candidate response entities of the query; and setting the expression of the relation directed graph aiming at the nodes except the nodes of the L-th layer in the augmented knowledge graph as a preset vector.
According to another embodiment of the present invention, there is provided a computer-readable storage medium storing instructions that, when executed by at least one computing device, cause the at least one computing device to perform the aforementioned method of prediction based on knowledge-graph reasoning.
According to another embodiment of the present invention, there is provided a system comprising at least one computing device and at least one storage device storing instructions that, when executed by the at least one computing device, cause the at least one computing device to perform the aforementioned method of prediction based on knowledge-graph reasoning.
Advantageous effects
By applying the prediction method and the prediction system based on knowledge graph reasoning according to the exemplary embodiment of the invention, structural information and semantic information among the relations of the knowledge graph can be fully reserved, and a plurality of facts can be reasoned through a graph neural network simultaneously, so that the operational efficiency and the effect of the reasoning are greatly improved, and the reasoning process and the reasoning basis can be understood by people through visualization.
Drawings
These and/or other aspects and advantages of the present invention will become more apparent and more readily appreciated from the following detailed description of the embodiments of the invention, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a block diagram illustrating a prediction system based on knowledge-graph reasoning, according to an exemplary embodiment of the present disclosure;
FIG. 2 is an example diagram illustrating a knowledge-graph;
FIG. 3 is an exemplary relational directed graph illustrating knowledge-graph extraction based on FIG. 2 in accordance with an illustrative embodiment of the present invention;
FIG. 4 is a process for extracting a relational directed graph based on the knowledge-graph of FIG. 2 according to an exemplary embodiment of the present invention;
fig. 5 is a flowchart illustrating a prediction method based on knowledge-graph reasoning according to an exemplary embodiment of the present disclosure.
Detailed Description
In order that those skilled in the art will better understand the present invention, exemplary embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
Fig. 1 is a block diagram illustrating a prediction system 100 based on knowledge-graph reasoning, according to an exemplary embodiment of the present disclosure.
Referring to FIG. 1, a knowledge-graph inference based prediction system 100 may include an augmentation device 110, a processing device 120, a training device 130, and a prediction device 140.
In an exemplary embodiment of the present invention, augmenting means 110 may augment a given knowledge-graph to obtain an augmented knowledge-graph in which relationship paths that do not exceed L in length may be augmented to length L. A knowledge-graph, as used herein, includes nodes, which may indicate entities and which are used interchangeably herein, and edges, which indicate a relationship between two nodes to which the edge connects and which are directed edges. Further, in the following, the vector may be represented by a bold type.
In particular, the present invention relates to a method for producing,the augmentation apparatus 110 may add a relationship pointing to itself (i.e., the same relationship) to each node in the knowledge graph, and add a corresponding edge showing an opposite relationship to each edge in the knowledge graph, so that the relationship path whose length in the knowledge graph does not exceed the length L (L is a preset positive integer) may be augmented to a uniform length L in the augmentation knowledge graph. For example, for an edge (head node, relationship, tail node), an inverse relationship may be added between the head node and the tail node, thereby obtaining an edge (tail node, inverse relationship, head node). In this manner, the augmenting means 110 can obtain an augmented knowledge graph based on the original given knowledge graph, hereinafter, for convenience of explanation, used
Indicating an augmented knowledge-graph, wherein,
for a set of fact triples, a fact triplet,
and
respectively, an entity set and a relationship set of the augmented knowledge graph.
After obtaining the augmented knowledge-graph, processing device 120 may obtain an expression of a relational directed graph for a query (query) based on the augmented knowledge-graph. The query may be represented as (e)q，rq，？)，eqRefers to a query entity (query entity), rqRefers to query relationship (query relation), which are given known elements, "? "is the missing responding entity (answer entry). Query (e)q，rqIs it? ) Is to predict the responding entities missing in the query, i.e., (e)q，rqIs it? ) Should be located in "? The responding entity at "is which entity in the augmented knowledge-graph.
In the embodiment of the invention, sinceThe missing responding entities that are the targets of inference may be any node in the augmented knowledge graph (i.e., all entities in the augmented knowledge graph may be responding entities), and thus, with respect to query (e)q，rqIs it? ) May include one or more candidate relational directed graphs for different candidate responder entities, e.g., nodes in an augmented knowledge graph
Is set to query (e)q，rqIs it? ) Candidate respondent entity of (e) obtained after the candidate tripletq，rq，ea) Is used for the corresponding candidate relationship directed graph.
In an exemplary embodiment of the invention, for candidate responder entity eaThe candidate relationship directed graph(s) of (2) can be an L-level directed graph, querying entity eqA source node, candidate responder entity e, which may be set to be located at level 0 in the candidate relationship directed graphaMay be set to be located at the rendezvous point at the L-th level in the candidate relationship directed graph, the directions of all edges in the candidate relationship directed graph are along the direction from the source node to the rendezvous point, and any path from the source node to the rendezvous point corresponds to the query entity e on the augmented knowledge graphqTo candidate responding entity eaA relationship path of length L.
FIG. 2 is an example diagram illustrating a knowledge-graph. FIG. 3 is an exemplary relational directed graph illustrating knowledge-graph extraction based on FIG. 2 according to an exemplary embodiment of the invention.
Referring to FIG. 2, Sam is taken as a query entity, and direct is taken as a query relation for explanation. Assuming that the candidate responding entity is Spider-2, the candidate triplet may be (Sam, directed, Spider-2), and further assuming that the level L of the relationship directed graph is 3, the relationship directed graph for Spider-2 with respect to the query (Sam, directed,
as can be seen in fig. 3.
In FIG. 3 there are relationships directedDrawing (A)
In (3) in the subscript of each parameter indicates the number of layers of the relational directed graph, the identification of the top right corner of each parameter or entity name is the number of layers, and the same entity name indicates the same entity in the knowledge-graph, e.g., Sam, regardless of the identification of the top right corner0、 Sam1And Sam2Both indicating the entity Sam in fig. 2. Further, the dotted lines represent inverse relationships, and the relationships between the same entity names are the same relationships pointing to themselves.
With reference to figure 3 of the drawings,
is a relational directed graph
The set of edges of the l-th layer in (1),
is that
L is a positive integer and L1.
The operation of layer-by-layer connection is shown,
furthermore, in the following, for a given query entity eqUse of
And
respectively represent an augmentationFrom e in the knowledge-graphqSet of outgoing edges (out-edge) and set of entities visible after starting a travel step, i.e., eqAnd a set of corresponding edges (i.e., edges pointing from the l-1 order neighbor nodes to the l-level).
To utilize relational directed graphs
To reason about the fact (e)q，rq，ea) Relational directed graphs may be obtained in exemplary embodiments of the invention
Expression (representation).
In other words, as described above, the processing device 120 may obtain at least one candidate triplet for the query by setting each entity in the augmented knowledge-graph as a candidate answering entity for the query, respectively, and may then obtain the representation of the candidate relationship directed graph for each candidate triplet based on the augmented knowledge-graph. In an exemplary embodiment of the invention, the representation of the relational directed graph may be learned using, for example, a graph neural network GNN, which will be explained in detail below in connection with algorithm 1.
Algorithm 1 (encoding of directed graph with respect to a single relationship)
As shown in Algorithm 1, a candidate triplet (e) may be formedq，rq，ea) Query entity e inqSet as corresponding candidate relation directed graph
And the candidate response entity e in the candidate tripleaSet as corresponding candidate relationship directed graph
The convergence point of (a) is calculated.
In step 1, an initialization process may be performed. For example, the processing device 120 may express the source node at layer 0
Set to a predetermined vector (e.g., a zero vector), i.e.,
furthermore, the processing means 120 may be provided
And is provided with
Thereafter, in steps 2 to 4, the processing device 120 may extract the source node e layer by layer from the layer 1 based on the augmented knowledge graphqFrom the 1 st order neighbor node to the L th order neighbor node and corresponding edge, up to the L-th layer. In the above-described algorithm 1, the algorithm,
representing source node eqOf order l (i.e., neighbor nodes of layer l),
a set of corresponding edges (i.e., outgoing edges pointing from level l-1 to level l) representing the order l neighbor nodes.
Then, in step 5, if the extracted neighbor nodes in the L-th layer are collected
Does not include a convergence point ea(i.e., the amount of the acid,
and
the intersection between them is 0, which may indicate a candidate relationship directed graph
Null), the processing device 120 may associate the candidate triplet (e) withq，rq，ea) Is a candidate relational directed graph
Expression of
The setting is a preset vector of, for example,
and if the extracted neighbor node in the L-th layer is collected
Including a point of convergence eaThen, in steps 6 to 8, the processing device 120 may extract the candidate relationship directed graph layer by layer starting from the convergence point of the L-th layer in reverse direction based on the 1 st to L-th order neighbor nodes and the corresponding edges of the extracted source node
The nodes and edges of each layer up to layer 1. In the above-described algorithm 1, the algorithm,
representing candidate relational directed graphs
Junction of the first layerThe set of points is then selected from the group,
representing candidate relational directed graphs
The set of edges in the l-th layer.
Thus, after step 8 above, the processing means 120 may establish a triplet (e) for the candidate based on the nodes and edges of the layers in the extracted candidate relationship directed graphq，rq，ea) Is a candidate relational directed graph
(e.g., a relational directed graph as in FIG. 3). At this time, in step 9 to step 11, the processing device 120 may direct the graph from the candidate relationship through the graph neural network
The expression of the node of each layer is calculated layer by layer from the 1 st layer to the L th layer of the hierarchy.
In Algorithm 1, for nodes in layer I
The expression thereof can be expressed as the following equation (1):
[ equation 1]
In the case of the equation 1, the,
is node e of layer l-1s(ii) is expressed in terms of (a),
is node esThe expression of the relationship r with node e at level l,
is a node relationship pair (e) for the inputq，rq) A message function of WlIs the weight matrix of the l-th layer.
In step 12, the processing device 120 may calculate the obtained convergence point e of the lth layeraExpression of
Determined as relating to a candidate triple (e)q，rq，ea) Is a candidate relational directed graph
The final expression of (1).
Using Algorithm 1, as above, the query (e) can be obtainedq，rqIs it? ) Against candidate responding entities eaIs the expression of directed graphs
And in this way, the query (e) can be obtained individually in an enumerated mannerq，rqIs it? ) By expressing a relational directed graph against other entities in the augmented knowledge graph, a query (e) can be obtainedq，rqIs it? ) Expression of all relationships directed graph.
However, it should be understood that algorithm 1 simply obtains information about the query (e)q，rqIs it? ) The present application is not limited thereto, and other manners can be used to obtain the query (e)q，rqIs it? ) Is expressed in a directed graph, which is explained in detail below in connection with algorithm 2.
Algorithm 2 (encoding directed graph relation based on recursion)
As shown in Algorithm 2, Algorithm 2 may query (e)q，rqIs it? ) The query entity in (2) is set as a source node of the relational directed graph for calculation.
In step 1, an initialization process may be performed. For example, the processing device 120 may express the source node at layer 0
Set to a predetermined vector (e.g., a zero vector), i.e.,
furthermore, the processing means 120 may be provided
Thereafter, in steps 2 to 5, the processing device 120 may extract neighbor nodes and corresponding edges of the source node in each layer from layer 1, layer by layer, based on the augmented knowledge graph, and based on the extracted set of neighbor nodes in each layer
And the collecting platform of the corresponding edge
(step 3), calculating the expression of the neighbor nodes of each layer by layer through the graph neural network (step 4) until the L-th layer.
Alternatively, in step 3 above, the processing device 120 may not need to collect all edges from the augmented knowledge graph, but may sample a uniform portion from all edges, thereby increasing efficiency.
In Algorithm 2, for nodes in layer I
The expression of which can be expressed as the following equation 2:
[ equation 2]
Through the recursive calculations of the above-mentioned steps 2 to 5, the processing device 120 can obtain the nodes of the layers extracted through the step 3 (i.e., the nodes belonging to the layers)
Node(s) of the expression. At this time, the processing device 120 may extract nodes in the lth layer (i.e., belonging to the same layer as the node in the first layer)
Node (e) is determined to be directed to the query (e)q，rqIs it? ) And may determine the expressions of the nodes of the L-th layer obtained by the calculation as related to the query (e), respectivelyq，rqIs it? ) For each corresponding candidate responding entity.
At step 6, processing device 120 may assign augmented knowledge-graph nodes other than the node at level L (i.e., not belonging to nodes at level L)
Node(s) is set to a preset vector, e.g., a zero vector.
Expression obtained using algorithm 2 according to an exemplary embodiment of the present invention
And expression obtained using Algorithm 1
The same, and algorithm 2 can be computed at the same time with the same (e)q，rq) Different from eaIs predicted from the fact thatqAnd the efficiency is greatly improved by expressing all starting relation directed graphs.
Fig. 4 shows a process of extracting a relational directed graph using algorithm 2 based on the knowledge-graph of fig. 2. As shown in FIG. 4, using Algorithm 2, from an ellipseThe relation directed graph formed by the circular frames is
And the relation directed graph formed by the rectangular frames is
That is, all the relational directed graphs from Sam can be obtained at the same time based on the algorithm 2, and thus, expressions of the relational directed graphs can be obtained at the same time, and the calculation efficiency is improved.
Further, in an exemplary embodiment of the present invention, to learn the query dependent expression of the relationship directed graph, further conversion of equation 2 to the following equation 3 may be used:
[ equation 3]
Therein, in equation 3, to find important edges, an attention-based mechanism may be used in each layer, based on the query relation rqWith respect to each edge (e) is calculateds，r，eo) Attention weight of
The following were used:
here, the weight vector
Weight matrix
d is the dimension of the vector expression, dαIs a dimension of the intermediate expression of attention,
is a join operator, σ () is a Sigmoid function, the output value is between 0 and 1, and ReLU () is an activation function.
It should be understood that the expressions of equations 1 to 3 above are merely examples, and the present application is not limited thereto, and other variant forms may be used to obtain the expression of the relational directed graph.
Through the above described computations, the processing device 120 may obtain an expression of all the relationship directed graphs for the query. The training device 130 may establish a loss function for the graph network model of the original given knowledge-graph based on the expression of the relational directed graph obtained by the processing device 120, and train the graph network model of the knowledge-graph based on the established loss function.
For example, the training device 130 may apply a scoring function f (e) to the modelq，rq，ea) Is determined as
Is a weight vector and can be obtained by model training learning.
Thus, the training device 130 can function the model's loss
Determined as equation 4 below:
[ equation 4]
The first part in equation 4 is to the training set
Positive sample triplet (e) of (a)q，rq，ea) A second part comprising a score for the query (e)q，rqIs it? ) All triplets of (2) scores, model parameters
Is randomly initialized and optimized during model training by minimizing equation 4 using, for example, a random gradient descent. It should be appreciated that in the second part of the loss function described above, not all triples may be employed, but rather a triplet with positive samples (e) may be usedq，rq，ea) Corresponding negative sample triplets.
After the training device 130 completes model training using the above-mentioned loss function, the prediction device 140 may predict the query using the trained graph network model (e)q，rqIs it? ) The responding entity of (1).
Fig. 5 is a flow diagram illustrating a prediction method 500 based on knowledge-graph reasoning, according to an exemplary embodiment of the present disclosure.
As shown in fig. 5, in step S510, the knowledge-map may be augmented by the augmenting means 110 to obtain an augmented knowledge-map. In an exemplary embodiment of the invention, the augmenting means 110 may add a relationship to each node in the knowledge-graph that points to itself and a corresponding edge that exhibits an inverse relationship to each edge in the knowledge-graph, such that relationship paths in the knowledge-graph that do not exceed the length L are augmented to the length L in the augmented knowledge-graph.
At step S520, an expression of a relational directed graph about the query may be obtained by the processing device 120 based on the augmented knowledge-graph. In an exemplary embodiment of the present invention, the query may include a query entity, a query relationship, and a response entity, and the relationship directed graph with respect to the query may include candidate relationship directed graphs for different candidate response entities.
Thereafter, in step S530, the training device 130 may establish a loss function for the graph network model of the knowledge-graph based on the expression of the relational directed graph obtained in step S520, and train the graph network model of the knowledge-graph based on the loss function.
Finally, in step S540, the predicting device 140 may predict the response entity in the query by using the trained graph network model.
The specific processes of detailed operations performed by the respective constituent elements of the prediction system 100 based on the knowledge graph reasoning according to the exemplary embodiment of the present disclosure have been described in detail above with reference to fig. 1 to 5, and therefore, for brevity, will not be described again here.
The prediction system and method based on the knowledge-graph inference according to the exemplary embodiments of the present disclosure may be applied to perform inference tasks of relational retrieval, semantic retrieval, intelligent recommendation, intelligent question answering, personalized recommendation, multi-hop prediction, node classification, content distribution, and the like.
By way of example only, in various application scenarios of the prediction method and system based on knowledge-graph inference according to exemplary embodiments of the present disclosure, for example, for retrieval (such as relational retrieval, semantic retrieval, intelligence, etc.), another entity may be retrieved by entering two keywords, for example, entering (mother of Zhang triplet) may retrieve another entity "Liqu" (mother of Zhang triplet).
For example, for intelligent question-answering, input "where are the capital of china? The user can accurately return to Beijing, so that the intention of the user can be really understood through the knowledge graph.
By applying the prediction method and the prediction system based on knowledge graph reasoning according to the exemplary embodiment of the invention, the structural information and the semantic information among the relations of the knowledge graph can be fully reserved, and a plurality of facts can be simultaneously reasoned through the graph neural network, so that the operational efficiency and the effect of reasoning are greatly improved.
A prediction method and system based on knowledge graph reasoning according to an exemplary embodiment of the present disclosure has been described above with reference to fig. 1 to 5. However, it should be understood that: the apparatus and systems shown in the figures may each be configured as software, hardware, firmware, or any combination thereof that performs the specified function. For example, the systems and apparatuses may correspond to an application-specific integrated circuit, a pure software code, or a module combining software and hardware. Further, one or more functions implemented by these systems or apparatuses may also be collectively performed by a component in a physical entity device (e.g., a processor, a client, a server, or the like).
Further, the above method may be implemented by instructions recorded on a computer-readable storage medium, for example, according to an exemplary embodiment of the present application, there may be provided a computer-readable storage medium storing instructions that, when executed by at least one computing device, cause the at least one computing device to perform the steps of: augmenting a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph, the lengths of which do not exceed L, are augmented to a length L in the augmented knowledge graph; obtaining an expression of a relationship directed graph about a query based on the augmented knowledge graph, wherein the query comprises a query entity, a query relationship and a response entity, and the relationship directed graph about the query comprises candidate relationship directed graphs for different candidate response entities; establishing a loss function for the knowledge-graph based on the expression of the relational directed graph; training a graph network model of the knowledge graph based on the loss function; predicting a response entity in the query by using the trained graph network model, wherein the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a convergence point located at the L-th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the convergence point, and any path from the source node to the convergence point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
The instructions stored in the computer-readable storage medium may be executed in an environment deployed in a computer device such as a client, a host, a proxy device, a server, and the like, and it should be noted that the instructions may also be used to perform additional steps other than the above steps or perform more specific processing when the above steps are performed, and the content of the additional steps and the further processing are mentioned in the description of the related method with reference to fig. 1 to 5, and therefore will not be described again in order to avoid repetition.
It should be noted that the prediction system based on the knowledge-graph inference according to the exemplary embodiments of the present disclosure may fully rely on the execution of computer programs or instructions to implement the corresponding functions, i.e., each device corresponds to each step in the functional architecture of the computer programs, so that the entire system is called by a specialized software package (e.g., lib library) to implement the corresponding functions.
On the other hand, when the system and apparatus shown in fig. 1 are implemented in software, firmware, middleware or microcode, program code or code segments to perform the corresponding operations may be stored in a computer-readable medium such as a storage medium, so that at least one processor or at least one computing device may perform the corresponding operations by reading and executing the corresponding program code or code segments.
For example, according to an exemplary embodiment of the present application, a system may be provided comprising at least one computing device and at least one storage device storing instructions, wherein the instructions, when executed by the at least one computing device, cause the at least one computing device to perform the steps of: augmenting a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph, the lengths of which do not exceed L, are augmented to a length L in the augmented knowledge graph; obtaining an expression of a relation directed graph related to query based on the augmented knowledge graph, wherein the query comprises a query entity, a query relation and a response entity, and the relation directed graph related to query comprises candidate relation directed graphs aiming at different candidate response entities; establishing a loss function for the knowledge-graph based on the expression of the relational directed graph; training a graph network model of the knowledge-graph based on the loss function; predicting a response entity in the query by using the trained graph network model, wherein the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a convergence point located at the L-th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the convergence point, and any path from the source node to the convergence point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
In particular, the above-described system may be deployed on a server or a client, or on a node in a distributed network environment. Further, the system may be a PC computer, tablet device, personal digital assistant, smart phone, web application, or other device capable of executing the set of instructions. In addition, the system may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). In addition, all components of the system may be connected to each other via a bus and/or a network.
The system here need not be a single system, but can be any collection of devices or circuits capable of executing the above instructions (or sets of instructions) either individually or in combination. The system may also be part of an integrated control system or system manager, or may be configured as a portable electronic device that interfaces with local or remote (e.g., via wireless transmission).
In the system, the at least one computing device may comprise a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller, or a microprocessor. By way of example, and not limitation, the at least one computing device may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, and the like. The computing device may execute instructions or code stored in one of the storage devices, which may also store data. Instructions and data may also be transmitted and received over a network via a network interface device, which may employ any known transmission protocol.
The memory device may be integrated with the computing device, for example, by having RAM or flash memory disposed within an integrated circuit microprocessor or the like. Further, the storage device may comprise a stand-alone device, such as an external disk drive, storage array, or any other storage device usable by a database system. The storage device and the computing device may be operatively coupled or may communicate with each other, such as through I/O ports, network connections, etc., so that the computing device can read instructions stored in the storage device.
While exemplary embodiments of the present application have been described above, it should be understood that the above description is exemplary only, and not exhaustive, and that the present application is not limited to the exemplary embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the present application. Therefore, the protection scope of the present application shall be subject to the scope of the claims.
Claims (12)
1. A prediction method based on knowledge-graph reasoning, the method comprising:
augmenting a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph, the lengths of which do not exceed L, are augmented to a length L in the augmented knowledge graph;
obtaining an expression of a relationship directed graph about a query based on the augmented knowledge graph, wherein the query comprises a query entity, a query relationship and a response entity, and the relationship directed graph about the query comprises candidate relationship directed graphs for different candidate response entities;
establishing a loss function for a graph network model of the knowledge-graph based on the expression of the relational directed graph;
training a graph network model of the knowledge-graph based on the loss function;
predicting response entities in the query using the trained graph network model;
the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a convergence point located at the L th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the convergence point, and any path from the source node to the convergence point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
2. The method of claim 1, wherein the step of augmenting the knowledge-graph comprises: adding a self-pointing relationship to each node in the knowledge graph, and adding a corresponding edge presenting an opposite relationship to each edge in the knowledge graph, so that a relationship path of which the length does not exceed the length L in the knowledge graph is extended to the length L in the augmented knowledge graph.
3. The method of claim 1, wherein obtaining the representation of the relational directed graph for the query comprises:
obtaining at least one candidate triplet for the query by setting an entity in the augmented knowledge graph as a candidate answer entity for the query;
obtaining a representation of a candidate relationship directed graph for each candidate triplet based on the augmented knowledge-graph.
4. The method of claim 3, wherein obtaining the representation of the candidate relationship directed graph with respect to each candidate triplet comprises:
setting a query entity in the candidate triple as a source node of the candidate relationship directed graph and setting a candidate response entity in the candidate triple as an aggregation point of the candidate relationship directed graph;
extracting 1-order neighbor nodes of the source nodes, L-order neighbor nodes and corresponding edges layer by layer from the 1 st layer until the L-th layer based on the augmented knowledge graph;
if the extracted neighbor nodes in the L-th layer do not comprise the rendezvous point, setting the expression of the candidate relation directed graph related to the candidate triple as a preset vector;
if the extracted neighbor nodes in the L-th layer comprise the convergent point, extracting nodes and edges of each layer in the candidate relationship directed graph layer by layer from the convergent point of the L-th layer until the layer 1 based on the 1 st order neighbor node to the L-order neighbor node and the corresponding edge of the extracted source node;
establishing a candidate relation directed graph of the candidate triple based on the nodes and edges of each layer in the extracted candidate relation directed graph, and calculating the expression of the nodes of each layer from the 1 st layer to the L th layer of the candidate relation directed graph through a graph neural network;
and determining the expression of the aggregation point of the L-th layer obtained by calculation as a final expression of the candidate relation directed graph related to the candidate triple.
5. The method of claim 1, wherein obtaining the representation of the relational directed graph for the query comprises:
setting the query entity in the query as a source node of a relationship directed graph;
extracting neighbor nodes and corresponding edges of source nodes on each layer from the layer 1 layer by layer based on the augmented knowledge graph, and calculating the expression of the neighbor nodes on each layer by layer through a graph neural network based on the extracted neighbor nodes and corresponding edges of each layer until the L-th layer;
determining neighbor nodes of the L-th layer as candidate response entities of the query, and respectively determining expressions of the neighbor nodes of the L-th layer obtained through calculation as expressions of candidate relationship directed graphs aiming at the corresponding candidate response entities of the query;
and setting the expression of the relation directed graph aiming at the nodes except the nodes of the L-th layer in the augmented knowledge graph as a preset vector.
6. A prediction system based on knowledge-graph reasoning, the system comprising:
an augmenting device configured to augment a knowledge graph to obtain an augmented knowledge graph, wherein the knowledge graph comprises nodes and edges, the nodes indicate entities, the edges indicate relationships between nodes to which the edges are connected and are directed edges, and relationship paths in the knowledge graph whose length does not exceed L are augmented to a length L in the augmented knowledge graph;
a processing device configured to obtain, based on the augmented knowledge-graph, a representation of a relational directed graph with respect to a query, wherein the query comprises a query entity, a query relation, and a response entity, and the relational directed graph with respect to the query comprises candidate relational directed graphs for different candidate response entities;
a training device configured to establish a loss function for the graph network model of the knowledge-graph based on the expression of the relational directed graph and train the graph network model of the knowledge-graph based on the loss function;
a prediction device configured to predict responding entities in the query using the trained graph network model,
the candidate relationship directed graph is a directed graph of an L layer, the query entity is set as a source node located at the 0 th layer in the candidate relationship directed graph, the corresponding candidate response entity is set as a convergence point located at the L th layer in the candidate relationship directed graph, directions of all edges in the candidate relationship directed graph are along a direction from the source node to the convergence point, and any path from the source node to the convergence point corresponds to a relationship path with a length of L from the query entity to the corresponding candidate response entity on the augmented knowledge graph, wherein L is a positive integer.
7. The system of claim 6, wherein the augmentation device is configured to: adding a self-directed relation to each node in the knowledge graph, and adding a corresponding side presenting an opposite relation to each side in the knowledge graph, so that a relation path of which the length does not exceed the length L in the knowledge graph is extended to the length L in the augmented knowledge graph, thereby obtaining the augmented knowledge graph.
8. The system of claim 6, wherein the processing device is configured to obtain the representation of the relational directed graph for the query by:
obtaining at least one candidate triplet for the query by setting an entity in the augmented knowledge graph as a candidate answer entity for the query;
obtaining a representation of a candidate relationship directed graph for each candidate triplet based on the augmented knowledge-graph.
9. The system of claim 8, wherein the processing device is configured to obtain the representation of the candidate relationship directed graph for each candidate triplet by:
setting a query entity in the candidate triple as a source node of the candidate relationship directed graph and setting a candidate response entity in the candidate triple as an aggregation point of the candidate relationship directed graph;
extracting 1-order neighbor nodes of the source nodes, L-order neighbor nodes and corresponding edges layer by layer from the 1 st layer until the L-th layer based on the augmented knowledge graph;
if the extracted neighbor nodes in the L-th layer do not comprise the rendezvous point, setting the expression of the candidate relation directed graph related to the candidate triple as a preset vector;
if the extracted neighbor nodes in the L-th layer comprise the convergent point, extracting nodes and edges of each layer in the candidate relationship directed graph layer by layer from the convergent point of the L-th layer until the layer 1 based on the 1 st order neighbor node to the L-order neighbor node and the corresponding edge of the extracted source node;
establishing a candidate relation directed graph of the candidate triple based on the nodes and edges of each layer in the extracted candidate relation directed graph, and calculating the expression of the nodes of each layer from the 1 st layer to the L th layer of the candidate relation directed graph through a graph neural network;
and determining the expression of the aggregation point of the L-th layer obtained by calculation as a final expression of the candidate relation directed graph related to the candidate triple.
10. The system of claim 6, wherein the processing device is configured to obtain the representation of the relational directed graph for the query by:
setting the query entity in the query as a source node of a relationship directed graph;
extracting neighbor nodes and corresponding edges of source nodes on each layer from the layer 1 layer by layer based on the augmented knowledge graph, and calculating the expression of the neighbor nodes on each layer by layer through a graph neural network based on the extracted neighbor nodes and corresponding edges of each layer until the L-th layer;
determining neighbor nodes of the L-th layer as candidate response entities of the query, and respectively determining expressions of the neighbor nodes of the L-th layer obtained through calculation as expressions of candidate relationship directed graphs aiming at the corresponding candidate response entities of the query;
and setting the expression of the relation directed graph aiming at the nodes except the nodes of the L-th layer in the augmented knowledge graph as a preset vector.
11. A computer-readable storage medium storing instructions that, when executed by at least one computing device, cause the at least one computing device to perform the method of any of claims 1 to 5.
12. A system comprising at least one computing device and at least one storage device storing instructions that, when executed by the at least one computing device, cause the at least one computing device to perform the method of any of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110931960.XA CN113626612A (en) | 2021-08-13 | 2021-08-13 | Prediction method and system based on knowledge graph reasoning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110931960.XA CN113626612A (en) | 2021-08-13 | 2021-08-13 | Prediction method and system based on knowledge graph reasoning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113626612A true CN113626612A (en) | 2021-11-09 |
Family
ID=78385444
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110931960.XA Pending CN113626612A (en) | 2021-08-13 | 2021-08-13 | Prediction method and system based on knowledge graph reasoning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113626612A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114757911A (en) * | 2022-04-14 | 2022-07-15 | 电子科技大学 | Magnetic resonance image auxiliary processing system based on graph neural network and contrast learning |
| CN114817512A (en) * | 2022-06-28 | 2022-07-29 | 清华大学 | Question-answer reasoning method and device |
| CN116011548A (en) * | 2023-03-24 | 2023-04-25 | 北京澜舟科技有限公司 | Multi-knowledge-graph question-answering model training method, system and storage medium |
| CN116564408A (en) * | 2023-04-28 | 2023-08-08 | 上海科技大学 | Synthetic lethal gene pair prediction method, device, equipment and medium based on knowledge-graph reasoning |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106886572A (en) * | 2017-01-18 | 2017-06-23 | 中国人民解放军信息工程大学 | Knowledge mapping relationship type estimation method and its device based on Markov Logic Networks |
| CN110059264A (en) * | 2019-04-24 | 2019-07-26 | 东南大学 | Location search method, equipment and the computer storage medium of knowledge based map |
| CN110837567A (en) * | 2019-11-18 | 2020-02-25 | 第四范式（北京）技术有限公司 | Method and system for embedding knowledge graph |
| US20200242444A1 (en) * | 2019-01-30 | 2020-07-30 | Baidu Usa Llc | Knowledge-graph-embedding-based question answering |
| US20200293517A1 (en) * | 2019-02-06 | 2020-09-17 | Parsons Corporation | Goal-directed semantic search |
| CN111858947A (en) * | 2019-04-26 | 2020-10-30 | 第四范式（北京）技术有限公司 | Automatic knowledge graph embedding method and system |
| CN112115328A (en) * | 2020-08-24 | 2020-12-22 | 苏宁金融科技(南京)有限公司 | Page flow map construction method and device and computer readable storage medium |
| CN112417171A (en) * | 2020-11-23 | 2021-02-26 | 南京大学 | Data augmentation method for knowledge graph representation learning |
| WO2021042546A1 (en) * | 2019-09-06 | 2021-03-11 | 平安科技（深圳）有限公司 | Knowledge graph link error checking and repairing method and apparatus, and storage medium |
| CN112784065A (en) * | 2021-02-01 | 2021-05-11 | 东北大学 | Unsupervised knowledge graph fusion method and unsupervised knowledge graph fusion device based on multi-order neighborhood attention network |
| CN112818137A (en) * | 2021-04-19 | 2021-05-18 | 中国科学院自动化研究所 | Entity alignment-based multi-source heterogeneous knowledge graph collaborative reasoning method and device |
| CN112905809A (en) * | 2021-03-30 | 2021-06-04 | 第四范式（北京）技术有限公司 | Knowledge graph learning method and system |
-
2021
- 2021-08-13 CN CN202110931960.XA patent/CN113626612A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106886572A (en) * | 2017-01-18 | 2017-06-23 | 中国人民解放军信息工程大学 | Knowledge mapping relationship type estimation method and its device based on Markov Logic Networks |
| US20200242444A1 (en) * | 2019-01-30 | 2020-07-30 | Baidu Usa Llc | Knowledge-graph-embedding-based question answering |
| US20200293517A1 (en) * | 2019-02-06 | 2020-09-17 | Parsons Corporation | Goal-directed semantic search |
| CN110059264A (en) * | 2019-04-24 | 2019-07-26 | 东南大学 | Location search method, equipment and the computer storage medium of knowledge based map |
| CN111858947A (en) * | 2019-04-26 | 2020-10-30 | 第四范式（北京）技术有限公司 | Automatic knowledge graph embedding method and system |
| WO2021042546A1 (en) * | 2019-09-06 | 2021-03-11 | 平安科技（深圳）有限公司 | Knowledge graph link error checking and repairing method and apparatus, and storage medium |
| CN110837567A (en) * | 2019-11-18 | 2020-02-25 | 第四范式（北京）技术有限公司 | Method and system for embedding knowledge graph |
| CN112115328A (en) * | 2020-08-24 | 2020-12-22 | 苏宁金融科技(南京)有限公司 | Page flow map construction method and device and computer readable storage medium |
| CN112417171A (en) * | 2020-11-23 | 2021-02-26 | 南京大学 | Data augmentation method for knowledge graph representation learning |
| CN112784065A (en) * | 2021-02-01 | 2021-05-11 | 东北大学 | Unsupervised knowledge graph fusion method and unsupervised knowledge graph fusion device based on multi-order neighborhood attention network |
| CN112905809A (en) * | 2021-03-30 | 2021-06-04 | 第四范式（北京）技术有限公司 | Knowledge graph learning method and system |
| CN112818137A (en) * | 2021-04-19 | 2021-05-18 | 中国科学院自动化研究所 | Entity alignment-based multi-source heterogeneous knowledge graph collaborative reasoning method and device |
Non-Patent Citations (2)
| Title |
|---|
| 官赛萍;靳小龙;贾岩涛;王元卓;程学旗;: "面向知识图谱的知识推理研究进展", 软件学报, no. 10 * |
| 郑玉艳;田莹;石川;: "一种元路径下基于频繁模式的实体集扩展方法", 软件学报, no. 10 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114757911A (en) * | 2022-04-14 | 2022-07-15 | 电子科技大学 | Magnetic resonance image auxiliary processing system based on graph neural network and contrast learning |
| CN114757911B (en) * | 2022-04-14 | 2023-04-07 | 电子科技大学 | Magnetic resonance image auxiliary processing system based on graph neural network and contrast learning |
| CN114817512A (en) * | 2022-06-28 | 2022-07-29 | 清华大学 | Question-answer reasoning method and device |
| CN114817512B (en) * | 2022-06-28 | 2023-03-14 | 清华大学 | Question-answer reasoning method and device |
| CN116011548A (en) * | 2023-03-24 | 2023-04-25 | 北京澜舟科技有限公司 | Multi-knowledge-graph question-answering model training method, system and storage medium |
| CN116564408A (en) * | 2023-04-28 | 2023-08-08 | 上海科技大学 | Synthetic lethal gene pair prediction method, device, equipment and medium based on knowledge-graph reasoning |
| CN116564408B (en) * | 2023-04-28 | 2024-03-01 | 上海科技大学 | Synthetic lethal gene pair prediction method, device, equipment and medium based on knowledge-graph reasoning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022068196A1 (en) | Cross-modal data processing method and device, storage medium, and electronic device | |
| CN110837550B (en) | Knowledge graph-based question answering method and device, electronic equipment and storage medium | |
| WO2021190451A1 (en) | Method and apparatus for training image processing model | |
| CN113626612A (en) | Prediction method and system based on knowledge graph reasoning | |
| CN111382868B (en) | Neural network structure searching method and device | |
| CN111563192B (en) | Entity alignment method, device, electronic equipment and storage medium | |
| CN110472002B (en) | Text similarity obtaining method and device | |
| WO2024041479A1 (en) | Data processing method and apparatus | |
| WO2023051369A1 (en) | Neural network acquisition method, data processing method and related device | |
| WO2021190433A1 (en) | Method and device for updating object recognition model | |
| CN114417986A (en) | Artificial intelligence-based medicine characteristic information determination method and device | |
| CN114547267A (en) | Intelligent question-answering model generation method and device, computing equipment and storage medium | |
| CN113254716A (en) | Video clip retrieval method and device, electronic equipment and readable storage medium | |
| CN113627422A (en) | Image classification method and related equipment thereof | |
| CN110837567A (en) | Method and system for embedding knowledge graph | |
| CN113360683B (en) | Method for training cross-modal retrieval model and cross-modal retrieval method and device | |
| WO2024114659A1 (en) | Summary generation method and related device | |
| CN116935188B (en) | Model training method, image recognition method, device, equipment and medium | |
| WO2024051655A1 (en) | Method and apparatus for processing histopathological whole-slide image, and medium and electronic device | |
| CN116186295B (en) | Attention-based knowledge graph link prediction method, attention-based knowledge graph link prediction device, attention-based knowledge graph link prediction equipment and attention-based knowledge graph link prediction medium | |
| WO2023197857A1 (en) | Model partitioning method and related device thereof | |
| WO2023143570A1 (en) | Connection relationship prediction method and related device | |
| CN115273148B (en) | Pedestrian re-recognition model training method and device, electronic equipment and storage medium | |
| CN115186738B (en) | Model training method, device and storage medium | |
| CN113298265B (en) | Heterogeneous sensor potential correlation learning method based on deep learning |
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 |