CN116186281A - Dynamic knowledge graph reasoning method and system based on multiple relation selection - Google Patents
Dynamic knowledge graph reasoning method and system based on multiple relation selection Download PDFInfo
- Publication number
- CN116186281A CN116186281A CN202211684090.1A CN202211684090A CN116186281A CN 116186281 A CN116186281 A CN 116186281A CN 202211684090 A CN202211684090 A CN 202211684090A CN 116186281 A CN116186281 A CN 116186281A
- Authority
- CN
- China
- Prior art keywords
- relation
- knowledge graph
- dynamic
- selection
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000002776 aggregation Effects 0.000 claims abstract description 16
- 238000004220 aggregation Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 238000012216 screening Methods 0.000 claims abstract description 14
- 108091026890 Coding region Proteins 0.000 claims abstract description 11
- 230000004931 aggregating effect Effects 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 17
- 230000015654 memory Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 4
- 238000007477 logistic regression Methods 0.000 claims description 4
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 230000001419 dependent effect Effects 0.000 abstract description 2
- 230000000875 corresponding effect Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000001228 spectrum Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 239000004973 liquid crystal related substance Substances 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000004075 alteration Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000013213 extrapolation Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 239000002585 base Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000012458 free base Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000002123 temporal 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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention belongs to the technical field of knowledge graph reasoning, and particularly provides a dynamic knowledge graph reasoning method and system based on multiple relation selection, wherein the method comprises the following steps: extracting hidden characteristics of the target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step; encoding dynamic information of events on a time sequence using an LSTM neural network; inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted. On the basis of RGCN adjacent aggregation, a multi-relation adjacent selection aggregation device is designed, semantic and structural information of the adjacent entities in time is obtained, aggregation capacity of a plurality of relation entities in a time step is enhanced, relation structure dependent characteristics between the adjacent entities are fully utilized, and therefore performance of dynamic knowledge graph reasoning is improved.
Description
Technical Field
The invention relates to the technical field of knowledge graph reasoning, in particular to a dynamic knowledge graph reasoning method and system based on multiple relation selection.
Background
The knowledge graph technology has recently gained great attention in domestic academia and industry, has demonstrated great application potential in important industries involving national folk life, such as finance, electronic commerce, health, medical treatment, etc., provides a method for mining and expressing relationships among entities from massive data, is an effective knowledge expression form, promotes the establishment of a series of knowledge bases, such as DBpedia, YAGO, freebase, etc., and provides important support for applications such as semantic search, man-machine question answering, personal recommendation, smart phone assistant, etc.
For example, chinese patent CN112084344a discloses a knowledge graph reasoning method comprising: acquiring initial knowledge graph data to be complemented, wherein the initial knowledge graph data comprises a plurality of groups of initial data groups, and the initial data groups only comprise head entities and entity relations; and according to the initial knowledge spectrum data, invoking a knowledge spectrum reasoning model to obtain the completed target knowledge spectrum data, wherein the knowledge spectrum reasoning model is a model which is obtained by training in advance based on reinforcement learning, and each group of target data groups in the target knowledge spectrum data comprises a head entity, an entity relationship and a tail entity. In the embodiment of the patent, the knowledge spectrum reasoning model obtained based on reinforcement learning training is called to carry out knowledge reasoning on the initial knowledge spectrum, the time change is not considered, and the knowledge spectrum is perfected by adopting multiple times of training.
At present, a plurality of methods for embedding dynamic graphs can support online embedding learning of nodes on the graphs, but the methods cannot be directly applied to embedding dynamic knowledge graphs, because the node embedding supported by the methods is based on structural neighbors and cannot describe relation edge information of a semantic level. Currently, some time knowledge graph models can be used for dynamic knowledge graphs, but their goal is to mine continuously developed knowledge from time-stamped knowledge graph snapshots for chain prediction and time prediction. Therefore, the existing dynamic knowledge graph is incomplete and has weak time correlation.
Disclosure of Invention
The invention aims at the technical problems of incomplete dynamic knowledge graph and weak time relevance existing in the prior art.
The invention provides a dynamic knowledge graph reasoning method based on multi-relation selection, which comprises the following steps:
s1, extracting hidden features of a target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step;
s2, encoding dynamic information of events on a time sequence by using an LSTM neural network;
s3, inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted.
Preferably, the S1 specifically includes:
on the basis of RGCN adjacent aggregator, constructing a multi-relation adjacent selection aggregator which only retains the neighborhood information under the multi-relation corresponding to the target entity to be predicted, and then obtaining the vector representation of the target entity neighborhood information by a method of solving the cumulative mean vector.
Preferably, the S1 specifically includes:
obtaining neighborhood information with strong relevance with a target entity by screening all the neighborhood information;
and realizing multi-relation adjacent selective aggregation by fusing the acquired specific field information with the information of the past time steps.
Preferably, the S2 specifically includes:
s21, modeling time sequence knowledge by using a long-short-time memory network LSTM, and constructing a continuous knowledge triplet prediction model;
s22, obtaining the dependence of the dynamic knowledge triples on multiple time and multiple relations, and establishing a joint probability model of the dynamic knowledge graph.
Preferably, the step S21 specifically includes:
representing the dynamic knowledge graph DCIKG as a sequence of time sequence knowledge triples;
suppose a set of knowledge triples G at a point in time τ τ Obtaining a time sequence knowledge triplet existing at a time point tau by following a Markov assumption;
and calculating the joint probability distribution of the dynamic knowledge graph.
Preferably, the step S3 specifically includes:
using the fully connected layer extraction features, a logistic regression activation function (softmax) is selected to output the probability distribution of the entity or relationship to be predicted.
Preferably, the step S3 specifically includes:
and using multi-classification cross entropy loss functions to represent the predictions of the entities and the relationships, and then obtaining the overall loss function of the RS-NET dynamic knowledge graph inference model.
The invention also provides a dynamic knowledge graph reasoning system based on the multi-relation selection, which is used for realizing a dynamic knowledge graph reasoning method based on the multi-relation selection, and comprises the following steps:
the multi-relation proximity selection aggregator is used for extracting hidden features of the target entity under the correspondence of different relations, screening relation information with stronger relevance from the target entity, and aggregating neighborhood information corresponding to the multi-relation under the same time step;
a timing knowledge encoder for encoding dynamic information of events on the timing sequence using the LSTM neural network;
and the time sequence knowledge reasoning module is used for inputting the coding sequence into the multi-element classifier, extracting the characteristics and outputting the probability distribution of the entity or the relation to be predicted.
The invention also provides the electronic equipment, which comprises a memory and a processor, wherein the processor is used for realizing the steps of the dynamic knowledge graph reasoning method based on multi-relation selection when executing the computer management class program stored in the memory.
The invention also provides a computer readable storage medium having stored thereon a computer management class program which when executed by a processor implements the steps of a dynamic knowledge graph reasoning method based on multiple relational selections.
The beneficial effects are that: the invention provides a dynamic knowledge graph reasoning method and a system based on multiple relation selection, wherein the method comprises the following steps: extracting hidden characteristics of the target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step; encoding dynamic information of events on a time sequence using an LSTM neural network; inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted. On the basis of RGCN adjacent aggregation, a multi-relation adjacent selection aggregation device is designed, semantic and structural information of the adjacent entities in time is obtained, aggregation capacity of a plurality of relation entities in a time step is enhanced, relation structure dependent characteristics between the adjacent entities are fully utilized, and therefore performance of dynamic knowledge graph reasoning is improved.
Drawings
FIG. 1 is a schematic diagram of a dynamic knowledge graph reasoning method based on multiple relation selection provided by the invention;
FIG. 2 is a schematic diagram illustrating a comparison of multiple relationship proximity selection aggregation modules provided by the present invention;
FIG. 3 is a diagram of the LSTM neural network structure and calculation formula provided by the invention;
fig. 4 is a schematic hardware structure of one possible electronic device according to the present invention;
fig. 5 is a schematic hardware structure of a possible computer readable storage medium according to the present invention.
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
As shown in fig. 1, the dynamic knowledge graph reasoning method based on multi-relation selection provided by the embodiment of the invention comprises the following steps:
s1, extracting hidden features of a target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step;
s2, encoding dynamic information of events on a time sequence by using an LSTM neural network;
s3, inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted.
The core task of dynamic knowledge graph reasoning is to infer and predict knowledge elements such as missing entities, relations and the like in a time sequence fact quadruple according to known information such as entities, relations and time, and the core task mainly comprises two kinds of knowledge elements:
(1) Given (h, r, τ) or (r, t, τ), the temporal condition is known, and the missing tail entity or head entity is deduced, i.e. whether the fact quadruple (h, r, t, τ) is valid or not is judged;
the time step range given a dynamic knowledge graph is from t 0 To t T According to the time position of the reasoning facts, the dynamic knowledge graph reasoning can be divided into two modes: (1) endogenous reasoning (interaction): judging the implicit facts of the dynamic knowledge graph at time t; (2) extrapolation (Extrapolation): predicting a dynamic knowledge graph, and evolving to new fact entities h, t or relation r at tau along with time, wherein tau T ≤τ。
(2) The time conditions are known, given (h, t, τ) 1 )，(h,t,τ 2 )，And (3) reasoning out the relation of the entity pair (h, t), namely judging whether the event quadruple (h, r, t, tau) is effective.
The dynamic knowledge-graph can be expressed as: DKG= (E, R, T), wherein E, R, T respectively represent entity, relationship, time set, a time sequence knowledge in dynamic knowledge graph, can be regarded as an entity pair with time mark and relationship thereof, namely, a fact four-element group (head entity, relationship, tail entity, time) can be expressed as (h, R, T, tau) or (h) τ ,r τ ,t τ ). At time τ e T, the set of knowledge quaternions that occur represents G τ A DKG can be considered to be built on a series of chronologically-arranged factual quaternions. Thus, the goal of dynamic knowledge-graph reasoning is to learn an efficient reasoning function f (x), based on a set { G } of a set of observed event quaternions on the dynamic knowledge-graph DKG 1 ,G 2 ,G 3 ,…,G τ The fact G is given by f (x) τ Probability distribution of occurrence P (G τ )。
In a preferred embodiment, step S1 specifically includes: obtaining neighborhood information with strong relevance with a target entity by screening all the neighborhood information; and realizing multi-relation adjacent selective aggregation by fusing the acquired specific field information with the information of the past time steps. In one specific implementation scenario:
based on the RGCN proximity aggregator, a Multi-relationship proximity selection aggregator (Multi-Relational Selected Graph Aggregator) was constructed as shown in FIG. 2. The multi-relation adjacent selection aggregator only reserves the neighborhood information under the multi-relation corresponding to the target entity to be predicted, and can effectively avoid the influence of the neighborhood information of the non-relevant target entity. And finally, obtaining vector representation of the target entity neighborhood information by using a method for solving the cumulative mean value vector.
The multi-relation proximity selection aggregator can accurately aggregate neighborhood characteristics, improve reasoning effects, simplify the complexity of reasoning calculation and improve the reasoning efficiency.
The RGCN aggregator can integrate information of the multi-relation neighborhood and the multi-hop neighborhood by including information of the multi-relation and the multi-level neighborhood of the target node.
For each relation of the target node, an entity-relation local structure diagram can be obtained, and for a multi-relation diagram, information of one entity is derived from aggregation of aggregation information under all types of relations, and an aggregation function eta (·) combines information in past time steps and is defined as follows:
for each target entity node, its initial hidden layer is expressed as
The embedded vector at training is set to e t ，c h Then it is a normalization factor. According to the extracted entity-relation local structure diagram, the neighborhood information is further aggregated:
obtaining the neighborhood information with strong relevance with the target entity by screening all the neighborhood information, namely
Finally, the aggregation function η (·) is integrated with the domain-specific information obtained by fusion with the information of the past time step +.>
Multiple relationship proximity selective aggregation is achieved.
In a preferred embodiment, step S2 specifically includes: and obtaining the dependence of the dynamic knowledge triples on multiple time and multiple relations, and establishing a joint probability model of the dynamic knowledge graph. In one specific implementation scenario:
the dynamic knowledge graph can be essentially regarded as a knowledge triplet with time sequence, and has strong time dependence. For reasoning of the dynamic knowledge graph, the impending event is generally predicted based on the time relevance of the knowledge triples, namely by utilizing the dynamic relevance of the knowledge triples in the time domain. The core of the circular event encoder is to obtain the dependency of the dynamic knowledge triples on multiple times and relationships by constructing a continuous knowledge triplet prediction model. In particular, the goal of the cyclic event encoder is to build a joint probability model of the dynamic knowledge-graph.
The dynamic knowledge graph DCIKG may be represented as a sequence of time series knowledge triples, defined as g= { G 1 ,G 2 ,...,G τ ,…,G T }，G τ = { (h, r, t, τ ')egτ' =τ } represents the set of all timing knowledge triples at time step τ. The joint probability model of the whole dynamic knowledge graph can be expressed as:
suppose a set of knowledge triples G at a point in time τ τ Following the Markov assumption, i.e. event G whose probability of distribution depends on the first m steps τ-m:t-1 Further, it can be assumed that G τ Event G before being given τ-m:t-1 Independent of each other, then the conditional probability at time τ can be expressed as:
wherein E is τ And (3) representing a time sequence knowledge triplet existing at the time point tau, and carrying the formula (3) into the formula (2) to obtain the joint probability distribution of the dynamic knowledge graph:
from the above formula, it can be seen that: given all past timing facts G τ-m:t-1 First, with probability P (h τ |G t-m:t-1 ) Calculating the obtainable header entity h τ Then with P (r τ |h τ ,G t-m:t-1 ) Generates a relation r by probability prediction of (2) τ Similarly, tail entity t τ From the probability P (t τ |h τ ,r τ ,G t-m:t-1 ) And (5) generating.
Based on the above representation, tail entity t τ Probability P (t) τ |h τ ,r τ ,G t-m:t-1 ) Parameterization is expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a learnable embedded vector representing the head entity h and the relation r, respectively,/respectively>
Is a vector representation of the local relationship structure for h, r at a point in time τ1;
similarly, the probabilities of the relationship r and the header entity h can be defined as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a global graph structure G for all entity relationships at time τ1 t-m:t-1 Is a vector representation of (c).
The cyclic event encoder can effectively capture the time correlation of the knowledge triples on the basis of joint probability distribution modeling of the whole dynamic knowledge graph. Jin et al model a sequence of time series events using a recurrent neural network (Recurrent Neural Network, RNN), achieving good results, but underutilizing long-term correlated information features. Garcia et al found that Long Short-Term Memory (LSTM) networks were able to effectively capture information of timing knowledge dependence over longer periods of time. Thus, embodiments of the present application choose to model timing knowledge using LSTM. The neural network structure and calculation formula of LSTM are shown in fig. 3.
Related studies of LSTM neural networks are well established and will not be described in detail. For convenience of representation and calculation, the related operations are expressed hereinafter as a function LSTM (·), as shown in equation 9:
h＝LSTM(x n ) (8)
wherein x is n And (3) representing an input sequence, h representing an output hidden state matrix, and LSTM (·) representing a calculation function of the LSTM model.
Global representation H τ Global information of all the graphs up to the time point t can be saved, and the local representation h τ (h, r) further emphasizes the local events associated with each entity and relationship, and the global and local representations are defined as follows:
H τ ＝LSTM 1 (η(G τ ),H τ-1 ) (9)
wherein η (·) is the previously constructed aggregation function;
representing all timing knowledge triples associated with the header entity h at the current point in time τ.
In a preferred embodiment, step S3 specifically includes: the coding sequence is input into a multivariate classifier, features are extracted using a full connected layer, and a logistic regression activation function (softmax) is selected to output probability distribution of the entity or relationship to be predicted. In one specific implementation scenario:
after passing through the LSTM encoder, the encoded sequence is input into a multivariate classifier, features are extracted using a full connected layer, and a logistic regression activation function (softmax) is selected to output a probability distribution of the entity or relationship to be predicted.
The RS-NET dynamic knowledge graph reasoning model regards entity prediction and relationship prediction as a multi-classification task, and each classification corresponds to an entity or relationship object respectively. The present embodiment uses multi-classification cross entropy loss functions to represent predictions of entities and relationships, respectively:
the overall loss function of the RS-NET dynamic knowledge graph reasoning model is available:
wherein, alpha and beta are weight coefficients respectively representing different target loss functions, and the alpha and the beta have different values aiming at different dynamic knowledge graph reasoning tasks.
As shown in fig. 1, the embodiment of the present invention further provides a dynamic knowledge graph inference system based on multiple relationship selection, where the system is configured to implement a dynamic knowledge graph inference method based on multiple relationship selection as described above, and the method includes:
the multi-relation proximity selection aggregator is used for extracting hidden features of the target entity under the correspondence of different relations, screening relation information with stronger relevance from the target entity, and aggregating neighborhood information corresponding to the multi-relation under the same time step;
a timing knowledge encoder for encoding dynamic information of events on the timing sequence using the LSTM neural network;
and the time sequence knowledge reasoning module is used for inputting the coding sequence into the multi-element classifier, extracting the characteristics and outputting the probability distribution of the entity or the relation to be predicted.
Fig. 2 is a schematic diagram of an embodiment of an electronic device according to an embodiment of the present invention. As shown in fig. 2, an embodiment of the present invention provides an electronic device, including a memory 1310, a processor 1320, and a computer program 1311 stored in the memory 1310 and executable on the processor 1320, wherein the processor 1320 executes the computer program 1311 to implement the following steps: s1, extracting hidden features of a target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step;
s2, encoding dynamic information of events on a time sequence by using an LSTM neural network;
s3, inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted.
Fig. 3 is a schematic diagram of an embodiment of a computer readable storage medium according to the present invention. As shown in fig. 3, the present embodiment provides a computer-readable storage medium 1400 having stored thereon a computer program 1411, which computer program 1411, when executed by a processor, performs the steps of: s1, extracting hidden features of a target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step;
s2, encoding dynamic information of events on a time sequence by using an LSTM neural network;
s3, inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted.
In the foregoing embodiments, the descriptions of the embodiments are focused on, and for those portions of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (10)
1. A dynamic knowledge graph reasoning method based on multi-relation selection is characterized by comprising the following steps:
s1, extracting hidden features of a target entity under the correspondence of different relations, screening relation information with stronger relevance to the target entity, and aggregating neighborhood information corresponding to multiple relations under the same time step;
s2, encoding dynamic information of events on a time sequence by using an LSTM neural network;
s3, inputting the coding sequence into a multi-element classifier, extracting characteristics, and outputting probability distribution of the entity or relation to be predicted.
2. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 1, wherein the step S1 specifically includes:
on the basis of RGCN adjacent aggregator, constructing a multi-relation adjacent selection aggregator which only retains the neighborhood information under the multi-relation corresponding to the target entity to be predicted, and then obtaining the vector representation of the target entity neighborhood information by a method of solving the cumulative mean vector.
3. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 2, wherein the step S1 specifically includes:
obtaining neighborhood information with strong relevance with a target entity by screening all the neighborhood information;
and realizing multi-relation adjacent selective aggregation by fusing the acquired specific field information with the information of the past time steps.
4. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 1, wherein the step S2 specifically comprises:
s21, modeling time sequence knowledge by using a long-short-time memory network LSTM, and constructing a continuous knowledge triplet prediction model;
s22, obtaining the dependence of the dynamic knowledge triples on multiple time and multiple relations, and establishing a joint probability model of the dynamic knowledge graph.
5. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 4, wherein the step S21 specifically includes:
representing the dynamic knowledge graph DCIKG as a sequence of time sequence knowledge triples;
suppose a set of knowledge triples G at a point in time τ τ Obtaining a time sequence knowledge triplet existing at a time point tau by following a Markov assumption;
and calculating the joint probability distribution of the dynamic knowledge graph.
6. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 1, wherein the step S3 specifically comprises:
using the fully connected layer extraction features, a logistic regression activation function (softmax) is selected to output the probability distribution of the entity or relationship to be predicted.
7. The dynamic knowledge graph reasoning method based on multiple relation selection of claim 6, wherein the step S3 specifically includes:
and using multi-classification cross entropy loss functions to represent the predictions of the entities and the relationships, and then obtaining the overall loss function of the RS-NET dynamic knowledge graph inference model.
8. A dynamic knowledge graph reasoning system based on multiple relation selection, characterized in that the system is used for realizing the dynamic knowledge graph reasoning method based on multiple relation selection as claimed in any one of claims 1-7, comprising:
the multi-relation proximity selection aggregator is used for extracting hidden features of the target entity under the correspondence of different relations, screening relation information with stronger relevance from the target entity, and aggregating neighborhood information corresponding to the multi-relation under the same time step;
a timing knowledge encoder for encoding dynamic information of events on the timing sequence using the LSTM neural network;
and the time sequence knowledge reasoning module is used for inputting the coding sequence into the multi-element classifier, extracting the characteristics and outputting the probability distribution of the entity or the relation to be predicted.
9. An electronic device comprising a memory, a processor for implementing the steps of the multiple relationship selection based dynamic knowledge graph inference method of any one of claims 1-7 when executing a computer management class program stored in the memory.
10. A computer-readable storage medium, having stored thereon a computer-management-class program which, when executed by a processor, implements the steps of the multiple-relation selection-based dynamic knowledge graph inference method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211684090.1A CN116186281B (en) | 2022-12-27 | 2022-12-27 | Dynamic knowledge graph reasoning method and system based on multiple relation selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211684090.1A CN116186281B (en) | 2022-12-27 | 2022-12-27 | Dynamic knowledge graph reasoning method and system based on multiple relation selection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116186281A true CN116186281A (en) | 2023-05-30 |
| CN116186281B CN116186281B (en) | 2024-05-31 |
Family
ID=86439478
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211684090.1A Active CN116186281B (en) | 2022-12-27 | 2022-12-27 | Dynamic knowledge graph reasoning method and system based on multiple relation selection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116186281B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117594241A (en) * | 2024-01-15 | 2024-02-23 | 北京邮电大学 | Dialysis hypotension prediction method and device based on time sequence knowledge graph neighborhood reasoning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376864A (en) * | 2018-09-06 | 2019-02-22 | 电子科技大学 | A kind of knowledge mapping relation inference algorithm based on stacking neural network |
| CN109992670A (en) * | 2019-04-04 | 2019-07-09 | 西安交通大学 | A kind of map completion method of knowledge based map neighbour structure |
| US20200012953A1 (en) * | 2018-07-03 | 2020-01-09 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for generating model |
| CN114860854A (en) * | 2022-05-05 | 2022-08-05 | 中国人民解放军国防科技大学 | Attention mechanism-based time sequence knowledge graph reasoning method, device and equipment |
-
2022
- 2022-12-27 CN CN202211684090.1A patent/CN116186281B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200012953A1 (en) * | 2018-07-03 | 2020-01-09 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for generating model |
| CN109376864A (en) * | 2018-09-06 | 2019-02-22 | 电子科技大学 | A kind of knowledge mapping relation inference algorithm based on stacking neural network |
| CN109992670A (en) * | 2019-04-04 | 2019-07-09 | 西安交通大学 | A kind of map completion method of knowledge based map neighbour structure |
| CN114860854A (en) * | 2022-05-05 | 2022-08-05 | 中国人民解放军国防科技大学 | Attention mechanism-based time sequence knowledge graph reasoning method, device and equipment |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117594241A (en) * | 2024-01-15 | 2024-02-23 | 北京邮电大学 | Dialysis hypotension prediction method and device based on time sequence knowledge graph neighborhood reasoning |
| CN117594241B (en) * | 2024-01-15 | 2024-04-30 | 北京邮电大学 | Dialysis hypotension prediction method and device based on time sequence knowledge graph neighborhood reasoning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116186281B (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112529168B (en) | GCN-based attribute multilayer network representation learning method | |
| Xu et al. | Hybrid regularized echo state network for multivariate chaotic time series prediction | |
| CN112115352B (en) | Session recommendation method and system based on user interests | |
| Hassan et al. | A HMM-based adaptive fuzzy inference system for stock market forecasting | |
| US20240152732A1 (en) | Training and prediction of hybrid graph neural network model | |
| Guo et al. | A novel dynamic rough subspace based selective ensemble | |
| Li et al. | Learning robust deep state space for unsupervised anomaly detection in contaminated time-series | |
| Hallaji et al. | DLIN: Deep ladder imputation network | |
| He et al. | MTAD‐TF: Multivariate Time Series Anomaly Detection Using the Combination of Temporal Pattern and Feature Pattern | |
| Chen et al. | Multi-scale temporal features extraction based graph convolutional network with attention for multivariate time series prediction | |
| CN115168443A (en) | Anomaly detection method and system based on GCN-LSTM and attention mechanism | |
| CN116186281B (en) | Dynamic knowledge graph reasoning method and system based on multiple relation selection | |
| CN113783874A (en) | Network security situation assessment method and system based on security knowledge graph | |
| CN117175588B (en) | Space-time correlation-based electricity load prediction method and device | |
| Sriramulu et al. | Adaptive dependency learning graph neural networks | |
| CN116028315A (en) | Operation early warning method, device, medium and electronic equipment | |
| Chen et al. | LFGCN: Levitating over graphs with levy flights | |
| Liu et al. | Memory-based transformer with shorter window and longer horizon for multivariate time series forecasting | |
| You et al. | Gatae: Graph attention-based anomaly detection on attributed networks | |
| Li et al. | Robust knowledge adaptation for dynamic graph neural networks | |
| CN117422134A (en) | Knowledge graph recommendation method based on graph convolution neural network | |
| CN116701880A (en) | Time sequence data prediction method, system, equipment and medium | |
| CN116484016A (en) | Time sequence knowledge graph reasoning method and system based on automatic maintenance of time sequence path | |
| Shamsinejadbabaki et al. | Causality-based cost-effective action mining | |
| Tao et al. | HAN-CAD: hierarchical attention network for context anomaly detection in multivariate time series |
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 |