CN106909622A - Knowledge mapping vector representation method, knowledge mapping relation inference method and system - Google Patents
Knowledge mapping vector representation method, knowledge mapping relation inference method and system Download PDFInfo
- Publication number
- CN106909622A CN106909622A CN201710041593.XA CN201710041593A CN106909622A CN 106909622 A CN106909622 A CN 106909622A CN 201710041593 A CN201710041593 A CN 201710041593A CN 106909622 A CN106909622 A CN 106909622A
- Authority
- CN
- China
- Prior art keywords
- entity
- relation
- path
- multistep
- knowledge mapping
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/22—Indexing; Data structures therefor; Storage structures
- G06F16/2228—Indexing structures
- G06F16/2237—Vectors, bitmaps or matrices
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a kind of vector representation method of knowledge mapping.The method includes：Entity in knowledge mapping is expressed as initial low-dimensional vector to, the multistep relation path between relation and entity pair；Using the loss function training entity of variable spaced to the model of, the low-dimensional vector representation of multistep relation path between relation and entity pair.The model learnt using the present invention carries out relation inference can improve the accuracy of the reasoning in different knowledge mappings.
Description
Technical field
The present invention relates to technical field of information processing, and in particular to a kind of pass of the knowledge mapping based on particular path translation
It is inference method and system.
Background technology
Knowledge mapping is the knowledge cluster organized in graph form in knowledge engineering, and it is by different types of entity as section
Point, relation are constituted as the side of connecting node.In knowledge mapping, entity refer in real world objective object (for example,
Barack Obama), or the abstract concept (for example, the 44th president in the U.S.) in human thought.Relation is then description
Actual relationship between two entities is (for example, Barack Obama is the 44th president in the U.S., i.e. Barack Obama
There is the relation of "Yes" and the 44th president in the U.S. between).
In known knowledge mapping, entity type has personage, event, organization, place etc., and the pass between them
Set type is also very diversified.Different entity types relation of interest is also different.For example, for people entities it
Between, common relation has relatives and friends；For between people and organization, common relation has work unit, graduation universities and colleges
Deng.The relation of these known inter-entity in original knowledge mapping than sparse, and actually inter-entity also exist it is a large amount of
Implication relation, can be excavated by existing knowledge in knowledge mapping and relation or reasoning these implication relations.
The most frequently used inference method is rule-based method, i.e., by the way that to acquainted analysis, formulation is suitably pushed away
Disconnected rule, is finally released the relation of inter-entity by these rules.But this method is laid down a regulation by artificial, workload it is very big and
The regular limited amount that can be formulated, covering scope is smaller, with larger limitation.In order to reduce the artificial mark amount of rule,
The relation inference method of another conventional knowledge mapping is automatically to obtain rule, example by machine learning according to existing knowledge
Such as, more depended on using the model based on translation such as existing transE, transR, transH, but the effect of this method
The selection of the parameter of selection and model to feature, migration needs to spend more energy, example in the knowledge mapping of different field
Such as, for the relation inference of sphere of learning, such as cooperative relationship is more focused on the content similarity feature of study hotspot, and have
Effect path length is generally shorter；And the relation inference in character relation field is more focused on structural similarity feature, and active path
Length may be more long, therefore, in actual applications, the migration between the knowledge mapping of different field has limitation.Additionally,
In the model of traditional Auto-learning Method, the accuracy of study is weighed typically by the loss function based on interval,
Every being generally pre-selected from candidate value, and, this be spaced in learning process in be changeless.This changeless
Every the accuracy of the study of the different knowledge mappings of regulation, different entities and the relation for being unable to self adaptation.
The content of the invention
It is an object of the invention to overcome above-mentioned defect of the prior art, there is provided a kind of relation of improved knowledge mapping
Inference method.
According to the first aspect of the invention, there is provided a kind of vector representation method of knowledge mapping, including：
Step 1：Entity in knowledge mapping is expressed as just to, the multistep relation path between relation and the entity pair
Beginning low-dimensional vector；
Step 2：The entity is trained to, the multistep between relation and the entity pair using the loss function of variable spaced
The model of the low-dimensional vector representation of relation path.
Preferably, the loss function includes loss and the damage of entity pair and multistep relation path of entity pair and relation
Lose.
Preferably, the loss of the entity pair and relation is defined as：
Wherein, Δ is the training set that triple (h, r, t) is constituted, and h is head entity, and t is tail entity, and r is represented between the two
Relation；Z represents the modulus of training set Δ；The negative example triple of Δ ' presentation-entity pair and relation, triple (h ', r ', t ') ∈
Δ ', is that, by the h in (h, r, t) ∈ Δs, r, t replaces with h ', r ', what t ' was obtained；h,r,t,h′,r′,t′∈Rd, RdRepresent dimension
Number is the low-dimensional vector space of d；[x]+Return to the higher value in both x and 0；| | | | represent L1Or L2Normal form；γ is triple
Positive example and negative example between interval.
Preferably, the loss of the entity pair and multistep relation path is defined as：
Lp,r=[| | p-r | |+Mpath(p)-||p-r′||]+
Wherein, MpathP () is the interval between the positive counter-example of multistep relation path, be defined as Mpath(p)=minr,r′|||p-
R ' | |-| | p-r | | |, p is the vector representation of multistep relation path；R ' is negative example relation r ' ∈ Nh,tIn low-dimensional vector representation to
Amount, Nh,tTo bear example triple (h, r ', set t) in knowledge mapping；| | | | represent L1Or L2Normal form.
Preferably, the length of the multistep relation path is less than threshold value.
Preferably, in training process in step 2, the loss function is updated using gradient descent method.
According to the second aspect of the invention, there is provided a kind of relation inference method of knowledge mapping.The method includes：
Step 11：All institutes for having relationship by objective (RBO) and the target entity between are found out from knowledge mapping according to target entity
There is candidate's entity as candidate's entity set；
Step 12：Vector representation method according to knowledge mapping may be with the target reality in candidate's entity set to infer
There is candidate's entity of the relationship by objective (RBO) in body.
Preferably, in step 12, infer that the candidate's entity in the Candidate Set exists with target entity using following formula
The possibility of the relationship by objective (RBO)：
F (h, r, t)=| | h+r-t | |
Wherein, runic h, r, t represent h, r, t low-dimensional vector space vector representation, | | | | represent L1Or L2Normal form.
According to the third aspect of the invention we, there is provided a kind of relation inference system of knowledge mapping.The system includes：For
All all candidate's entity conducts for having relationship by objective (RBO) and the target entity between are found out from knowledge mapping according to target entity
The device of candidate's entity set；In inferring candidate's entity set for the vector representation method of knowledge mapping of the invention
May there is the device of candidate's entity of the relationship by objective (RBO) with the target entity.
Compared with prior art, the advantage of the invention is that：
Learn rule of inference automatically using existing entity relationship and entity in knowledge mapping, using the loss of variable spaced
Function carries out the contact that study can be adaptively set up between the relation between entity pair and multistep relation path, improves study
Accuracy.For different knowledge mappings, the present invention can be adaptively calculated the optimal interval value of loss function, it is not necessary to carry
It is preceding to define any candidate interval value.Compared with other models, complexity is identical, but relation inference effect is significantly improved.
Brief description of the drawings
The accompanying drawing for being combined in the description and constituting a part for specification shows embodiments of the invention, and even
It is used to explain principle of the invention together with its explanation.
Fig. 1 shows the flow chart of the relation inference method of knowledge mapping according to an embodiment of the invention.
Fig. 2 shows the schematic diagram at particular path interval according to an embodiment of the invention.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with accompanying drawing by specific real
The present invention is described in more detail to apply example.It should be appreciated that specific embodiment described herein is only used to explain the present invention, and
It is not used in the restriction present invention.
In short, the method according to inventive embodiment includes：By entity to, the multistep relation between relation and entity pair
Path representation is initial low-dimensional vector；Design learning target (loss function), trains based on the model of particular path translation
Practise, by constantly adjustment, the value of more novel entities, relation and vector makes loss function minimum；Using these vectors for training
Carry out entity prediction or relation inference etc..
Herein, described particular path translation refers to represent in the knowledge mapping translated based on existing path
During habit, specific interval is calculated each multistep relation path, preferably to learn the low-dimensional vector of multistep relation path,
That is the interval of each multistep relation path is specific, therefore, it can the referred to as model based on particular path translation.The present invention is adopted
Study is trained with the loss function of variable spaced, the adaptivity for further increasing learning process is accurate with study
Degree.Specifically, Fig. 1 shows the flow chart of the relation inference method of knowledge mapping according to an embodiment of the invention.
1) step S110, generates multistep relation path set.
Herein, multistep relation path refers to the path that there is relationship by objective (RBO) between entity pair, and the length in path can be with
Value is more than or equal to 1.For example,, the length in its path is 2, generally with "-work in-work
In-1- " represent.When the length in path exceedes threshold value, it is believed that the incidence relation between the entity pair is weaker, can not examine
Consider.
In one embodiment, the mode for obtaining the set of multistep path relation is from existing according to given relationship by objective (RBO)
Knowledge base in find out it is all exist the relations entities pair；For each entity pair, from the beginning entity starts, until reaching tail reality
Body or the path threshold more than setting, obtain the set of multistep relation path.Existing knowledge base is included but is not limited to
Freebase, Google Knowledge Graph, GeneOntology etc..
If for example, the relationship by objective (RBO) searched is " coauthor (collaboration) ", traveling through all entities that there is the relation
It is right, it is understood that there may be "<A-B>", "<C-D>" etc..For example, for<A-B>Entity pair, can be used breadth-first search traversal method
To obtain all set of paths between them.The multistep relation path of all entities pair can be finally obtained in this way
Set.
2) step S120, calculates the importance degree of multistep relation path and selects important path.
In this step, weighed by calculating the score R (p | h, t) of every multistep relation path between each pair entity
Its importance degree.R (p | h, t) is the stock number that from the beginning entity h (head) reaches tail entity t (tail), for weighing entity to it
Between relation path importance degree, score value it is higher represent the paths it is more important.
For example, for entity to (h, t), if the relation path p between the entity pair isOrder
Rp(h)=1, R ([| h, t)=Rp(t), and for entity m ∈ Si, the annexation r of miPreposition node be expressed as Si-1(·,
M), then the score of this relation path is calculated as：
Wherein, the span of score is between 0 to 1；Represent arbitrary node.For example, Si(n) represent with n for
The set of the arbitrfary point of point.
Specifically, if in knowledge mapping, for a certain entity pair, there is pathWithAnd
So, S1(, t)={ a, b }, also, S1(h)={ a, b }, S2(a)={ t, t ' }, S2(b)=
{ t }, then：
Every score of relation path can be calculated by the formula (1), in the present embodiment, alternatively can be only
Important path (for example, for each pair entity pair, selecting three paths of highest scoring) is selected to perform procedure below.Pass through
This mode, can reduce amount of calculation, improve the efficiency of study.
3) step S130, low-dimensional vector representation is initialized as by entity, relation and multistep relation path.
Entity, relation, multistep relation path in the knowledge mapping that will have been obtained in this step, be initialized as with to
Measure to represent, the dimension of vector is general between 0-300.In one embodiment, using be evenly distributed or Bernoulli Jacob distribution come
Initialized.Entity end to end and relation are represented with triple (h, r, t), path is represented with p.
All objects (entity, relation etc.) can be mapped to by this low-dimensional by the method for this low-dimensional vector representation empty
Between in.
4) step S140, construction training set.
In this embodiment, two training sets are constructed and share training in model, first training set by triple (h,
R, t) positive example and negative example composition；Second training set is by being that the positive example and negative example of entity pair and multistep relation path are constituted.
For example, as it is known that Lee and Zhang have " collaboration " relation, then triple (h, r, t) is (Lee collaborates, Zhang),
This triple is necessary being, therefore, it is positive example triple.Can be obtained by least one in random replacement h, r, t
To corresponding negative example triple, for example, (Lee collaborates, Shen).The entity may have p to corresponding multistep relation path1=
"-work in-work in-1-”、p2="-tutor-tutor-1- ", p3="-be published in-be published in-1- " etc., can be with by replacing r
Entity pair negative example corresponding with multistep relation path is obtained, for example, (Lee, man and wife, Zhang), the multistep relation path of the entity pair
Should be corresponding with " collaboration ", it is now corresponding with " man and wife ".
5) step S150, the low-dimensional vector representation of learning object, relation and multistep relation path.
The purpose of this step is to make it by the value of the vectors such as continuous adjustment, more novel entities, relation, multistep relation path
Closer to legitimate reading.The degree of closeness with legitimate reading, i.e. learning objective are weighed with loss function to be made by optimization
Loss function value is minimum.
Defining loss function is：
Wherein, (E z) represents the loss function of low-dimensional vector learning process to L；E represents low-dimensional vector representation；Z represents ternary
Group (h, r, t), h is head entity, and t is tail entity, and r is relation between the two；Δ is the collection of the known triple in knowledge mapping
Close, as the training set in learning process；Z represents the number of triple in the modulus of training set Δ, i.e. Δ；Ph,t={ p1,
p2,…,pmRepresent the set of multistep relation path from h to t；Lp,rRepresent the loss of multistep relation path p and relation r；Δ″
The negative example of presentation-entity pair and multistep relation path, for optimizing
In formula (2), Part I Eh,r,tThe loss of presentation-entity pair and relation；Part IIThe loss of presentation-entity pair and multistep relation path, wherein, R (p | h, t)
To weigh score of the entity to the relation path p importance degrees between (h, t).
The specific steps of computing formula (2) include：
A) the loss E of computational entity pair and relationh,r,t。
The loss function of entity pair and relation is defined as：
Wherein, Δ is the training set that triple is constituted；Z represents the number of triple in the modulus of training set Δ, i.e. Δ；
The negative example triple of Δ ' presentation-entity pair and relation, triple (h ', r ', t ') ∈ Δs ', be by h, the r in (h, r, t) ∈ Δs,
T replaces with h ', r ', what t ' was obtained, and h is head entity, and t is tail entity, and r represents relation between the two；h,r,t,h′,r′,t′
∈Rd, RdRepresentation dimension is the low-dimensional vector space of d；[x]+Return to the higher value in both x and 0；| | | | represent L1Or L2Model
Formula；γ is interval, that is, distinguish a non-negative of positive example triple (h, r, t) ∈ Δs and negative example triple (h ', r ', t ') ∈ Δs '
Value, the value of γ can preset.
From formula (3), for the example that known Lee and Zhang have " collaboration " relation, its optimization aim is positive example
The loss of (Lee collaborates, Zhang) is more than the loss of negative example (Lee collaborates, Shen) plus interval γ.
B) interval of multistep relation path is calculated.
Every optimal interval of relation path can be calculated, it is also possible to according to obtaining for above-mentioned multistep relation path importance degree
Divide the optimal interval for only calculating the important relation path of a portion.
In the present invention, the optimal interval of relation path is defined as：
Mpath(p)=minr,r′|||p-r′||-||p-r||| (4)
Wherein, positive example relation r ∈ Rh,t, Rh,tIt is the set of the relation r of positive example triple (h, r, t) in knowledge mapping；It is negative
Example relation r ' ∈ Nh,t, Nh,tFor born in knowledge mapping example triple (h, r ', the set of relation r ' t), (h, r ', t) be by (h,
R, t) in r replace with what r ' was obtained；| | | | represent L1Or L2Normal form.
More specifically, in preferable knowledge mapping low-dimensional vector representation, for specific multistep relation path p, with
Its related positive example relation r should be as close as possible in low-dimensional vector space, and relative negative example relation r ' low-dimensional to
Should be as mutually remote as possible in quantity space.Therefore, the vector representation of relation path p causes corresponding positive example relation r ∈ Rh,tCluster
Together, and with negative example relation r ' ∈ Nh,tBetween have certain interval.
(shown with X-Y scheme) as shown in Figure 2, the optimal interval M of particular pathpathP () is equal to two concentric hyperspheres
The super radius mould of body difference long, positive example relation r (being represented with open circles) is respectively positioned on inner side spheroid, bear example relation r ' (with it is hollow just
Square expression) it is respectively positioned on beyond outside spheroid, optimal interval MpathP () is equal to the distance between inside and outside hypersphere body interval.Cause
This, can be use up positive example and negative example by setting appropriate interval to every multistep relation path or important multistep relation path
It is possible to separate, such that it is able to improve the accuracy of low-dimensional vector study.
C) loss between multistep relation path and relation.
LP, r=[| | p-r | |+Mpath(p)-||p-r′||]+ (5)
Wherein, MpathP () is the optimal interval of multistep relation path；P is by relation r1, r2..., rlThe relation road of composition
Footpath p={ r1, r2..., rlLow-dimensional vector representation vector, p ∈ Rd, and p=r1+r2+…+rl, d is the dimension of low-dimensional vector space
Degree；R ' is negative example relation r ' ∈ NH, tIn low-dimensional vector representation vector, NH, tFor born in knowledge mapping example triple (h, r ', t)
Set；| | | | represent L1Or L2Normal form.
From formula (5), by defining Mpath(p), for the relation road of different knowledge mapping or different entities pair
Footpath, its interval is variable, i.e., loss function is variable.Therefore, this method can adaptively learning knowledge vector
Low-dimensional is represented.
In sum, in an embodiment of the present invention, loss function L (E, z) phase of the low-dimensional vector learning process of definition
When in there is two constraints：One is so that the triple of entity pair and relation composition closest to truth；Two are so that entity pair
Multistep relation path closest to truth.Wherein, the loss reduction of entity pair and relation is finger entity vector plus relation
Vector should be close with tail entity vector.The minimum of the loss between entity pair and relation path is by the relation between entity pair
To weigh, such as the vector of "-father-father-" this paths should distance be approached in lower dimensional space with the vector of " grandfather ".
In learning process, loss function can be optimized by renewal vector.It is for instance possible to use gradient descent method
To be updated, vectorial update mode is as follows：
hi=hi+μ*2*|ti-hi-ri|
ri=ri+μ*2*|ti-hi-ri|
ti=ti-μ*2*|ti-hi-ri|
h′i=h 'i-μ*2*|t′i-h′i-r′i|
r′i=r 'i-μ*2*|t′i-h′i-r′i|
t′i=t 'i+μ*2*|t′i-h′i-r′i|
Wherein, dim is the dimension of vector space, hiRepresent the i-th dimension vector of h.μ is learning rate, typically 0.1,
0.01 .0.001 } selection in.
Step S150 is performed by iteration, until convergence, you can entity, relation, multistep relation path after being trained
Vector.
6) step S160, relation inference is carried out based on scoring functions.
The purpose of this step be according to above-mentioned study obtain entity, relation, relation path low-dimensional vector representation, come
Carry out relation inference.Specifically include：
A) for target entity and relation to be inferred, all entities for meeting candidate's entity type are chosen as candidate's reality
Body set.
B), according to following scoring functions and study obtain entity, relation low-dimensional vector representation, calculate candidate's entity and
The score value of target entity, the triple of relationship by objective (RBO) composition.
F (h, r, t)=| | h+r-t | | (6)
Wherein, runic h, r, t represent h, r, t low-dimensional vector space vector representation, | | | | represent L1Or L2Normal form.
C) candidate's entity is sorted according to the score value for calculating, orderly candidate's list of entities is obtained.For example, candidate's entity
First in the list entity of to be most probable produce with target entity relationship by objective (RBO).
For example, relationship by objective (RBO) r=places of birth to be inferred, target head entity Lee h=to be inferred, candidate's entity set
Conjunction is probably { Beijing, Hangzhou, Qingdao, Hainan ... }, calculates the orderly list of entities { Hangzhou, Beijing } obtained after score value.
The effect of relation inference, data set of the inventor in Freebase are carried out using the present invention in order to further illustrate
Verified on FB15K.Freebase data sets are the Typical Representatives in existing knowledge storehouse, are one and increase income and constantly update
Open knowledge base.The method that inventor is provided using entity Predicting Technique and the present invention, using the average sequence value of positive entity
(Mean Rank) is verified as evaluation index.Experiment parameter is as follows：In data set FB15K, exist 1345 kinds of relations and
14951 entities.To FB15K data sets, learning rate λ=0.001 that learning process is used, the optional dimension d=50 of vector, batch
Treatment size B=4800, parameter γ=1, path length is 2, from L1Normal form weighs similarity.
By experiment, the method according to the invention average sequence value under the conditions of " Filter " is 85, is put down under the conditions of " Raw "
Equal sequence value is 24.(wherein, " Filter " represented that before being ranked up to candidate's entity positive triple has been filtered out (example
Such as, Lee has two children A and B, and entity is predicted to (Lee, child, B), gives " Lee " and relation " child ", waits
Selected works remove positive example A so that candidate collection only has B mono- correct entity, referred to as filter；Conversely, Candidate Set is not processed,
So that there is the correct entities of A, B two in candidate collection, possible A is come before B during sequence so that mean rank are deteriorated, referred to as
raw).The average sequence value for using TransR methods popular in the prior art to obtain is 226 (Filter), 78 (Raw)；Using another
The average sequence value that a kind of existing HOLE methods are obtained is 259 (Filter), 116 (Raw).Therefore, the present invention is used to provide
Method compared with TransR technologies, sequence value significantly reduces 50~110；Compared with using HOLE technologies, sequence value is significantly reduced
90~180.
In sum, relation and multistep relation road between the method according to the invention, first modeled target entity pair
Connecting each other for footpath, is used low-dimensional vector representation, re-defines the optimal spacing value of self adaptation, can improve model learning
Accuracy such that it is able to improve the accuracy that the relation of knowledge mapping is inferred.
It is described above various embodiments of the present invention, described above is exemplary, and non-exclusive, and
It is not limited to disclosed each embodiment.In the case of without departing from the scope and spirit of illustrated each embodiment, for this skill
Many modifications and changes will be apparent from for the those of ordinary skill in art field.The selection of term used herein, purport
Best explaining the principle of each embodiment, practical application or to the technological improvement in market, or make the art its
Its those of ordinary skill is understood that each embodiment disclosed herein.The scope of the present invention be defined by the appended claims.
Claims (10)
1. a kind of vector representation method of knowledge mapping, including：
Step 1：Entity in knowledge mapping is expressed as to, the multistep relation path between relation and the entity pair initial low
Dimensional vector；
Step 2：The entity is trained to, the multistep relation between relation and the entity pair using the loss function of variable spaced
The model of the low-dimensional vector representation in path.
2. method according to claim 1, wherein, the loss function includes the damage of entity pair and multistep relation path
Lose.
3. method according to claim 2, wherein, the loss of the entity pair and multistep relation path is defined as：
Lp,r=[| | p-r | |+Mpath(p)-||p-r′||]+
Wherein, MpathP () is the interval between the positive counter-example of multistep relation path, be defined as Mpath(p)=minr,r′|||p-r′||-
| | p-r | | |, p is the vector representation of multistep relation path；R ' is negative example relation r ' ∈ Nh,tIn low-dimensional vector representation vector,
Nh,tTo bear example triple (h, r ', set t) in knowledge mapping；| | | | represent L1Or L2Normal form.
4. method according to claim 2, wherein, the loss function also includes the loss of entity pair and relation.
5. method according to claim 4, wherein, the loss of the entity pair and relation is defined as：
Wherein, Δ is the training set that triple (h, r, t) is constituted, and h is head entity, and t is tail entity, and r represents relation between the two；
Z represents the modulus of training set Δ；The negative example triple of Δ ' presentation-entity pair and relation, triple (h ', r ', t ') ∈ Δs ' are
H in (h, r, t) ∈ Δs, r, t are replaced with into h ', r ', what t ' was obtained；h,r,t,h′,r′,t′∈Rd, RdRepresentation dimension is d
Low-dimensional vector space；[x]+Return to the higher value in both x and 0；| | | | represent L1Or L2Normal form；γ be triple just
Interval between example and negative example.
6. method according to claim 1, wherein, the length of the multistep relation path is less than threshold value.
7. method according to claim 1, wherein, in the training process of step 2, using gradient descent method to update
State loss function.
8. a kind of relation inference method of knowledge mapping, including：
Step 11：All all times for having relationship by objective (RBO) and the target entity between are found out from knowledge mapping according to target entity
Entity is selected as candidate's entity set；
Step 12：The model of the low-dimensional vector representation obtained according to any one of claim 1-7 infers candidate's reality
Body concentrates the candidate's entity that may there is the relationship by objective (RBO) with the target entity.
9. method according to claim 8, wherein, in step 12, the time in the Candidate Set is inferred using following formula
Select entity that there is the relationship by objective (RBO) with target entity：
F (h, r, t)=| | h+r-t | |
Wherein, runic h, r, t represent h, r, t low-dimensional vector space vector representation, | | | | represent L1Or L2Normal form.
10. the relation inference system of a kind of knowledge mapping, including：
For all candidates for having relationship by objective (RBO) and the target entity between to be found out from knowledge mapping according to target entity
Entity as candidate's entity set device；
Candidate's entity is inferred for the model of the low-dimensional vector representation obtained according to any one of claim 1-7
May there is the device of candidate's entity of the relationship by objective (RBO) with the target entity in concentration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041593.XA CN106909622A (en) | 2017-01-20 | 2017-01-20 | Knowledge mapping vector representation method, knowledge mapping relation inference method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041593.XA CN106909622A (en) | 2017-01-20 | 2017-01-20 | Knowledge mapping vector representation method, knowledge mapping relation inference method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106909622A true CN106909622A (en) | 2017-06-30 |
Family
ID=59207024
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710041593.XA Pending CN106909622A (en) | 2017-01-20 | 2017-01-20 | Knowledge mapping vector representation method, knowledge mapping relation inference method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106909622A (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480191A (en) * | 2017-07-12 | 2017-12-15 | 清华大学 | A kind of entity alignment model of iteration |
| CN107545033A (en) * | 2017-07-24 | 2018-01-05 | 清华大学 | A kind of computational methods based on the knowledge base entity classification for representing study |
| CN107590139A (en) * | 2017-09-21 | 2018-01-16 | 桂林电子科技大学 | A kind of knowledge mapping based on circular matrix translation represents learning method |
| CN107590237A (en) * | 2017-09-11 | 2018-01-16 | 桂林电子科技大学 | A kind of knowledge mapping based on dynamic translation principle represents learning method |
| CN107729444A (en) * | 2017-09-30 | 2018-02-23 | 桂林电子科技大学 | Recommend method in a kind of personalized tourist attractions of knowledge based collection of illustrative plates |
| CN107943874A (en) * | 2017-11-13 | 2018-04-20 | 平安科技（深圳）有限公司 | Knowledge mapping processing method, device, computer equipment and storage medium |
| CN108073711A (en) * | 2017-12-21 | 2018-05-25 | 北京大学深圳研究生院 | A kind of Relation extraction method and system of knowledge based collection of illustrative plates |
| CN108228877A (en) * | 2018-01-22 | 2018-06-29 | 北京师范大学 | Knowledge base complementing method and device based on study sort algorithm |
| CN108304933A (en) * | 2018-01-29 | 2018-07-20 | 北京师范大学 | A kind of complementing method and complementing device of knowledge base |
| CN108446769A (en) * | 2018-01-23 | 2018-08-24 | 深圳市阿西莫夫科技有限公司 | Knowledge mapping relation inference method, apparatus, computer equipment and storage medium |
| CN108509483A (en) * | 2018-01-31 | 2018-09-07 | 北京化工大学 | The mechanical fault diagnosis construction of knowledge base method of knowledge based collection of illustrative plates |
| CN108510110A (en) * | 2018-03-13 | 2018-09-07 | 浙江禹控科技有限公司 | A kind of water table trend analysis method of knowledge based collection of illustrative plates |
| CN108846000A (en) * | 2018-04-11 | 2018-11-20 | 中国科学院软件研究所 | A kind of common sense semanteme map construction method and device based on supernode and the common sense complementing method based on connection prediction |
| CN108959472A (en) * | 2018-06-20 | 2018-12-07 | 桂林电子科技大学 | Knowledge mapping based on multistep relation path indicates learning method |
| CN109146078A (en) * | 2018-07-19 | 2019-01-04 | 桂林电子科技大学 | A kind of knowledge mapping expression learning method based on dynamic route |
| CN109492027A (en) * | 2018-11-05 | 2019-03-19 | 南京邮电大学 | It is a kind of based on weak trust data across the potential character relation analysis method of community |
| CN110019982A (en) * | 2017-12-05 | 2019-07-16 | 航天信息股份有限公司 | The determination method and device of node coordinate |
| CN110796254A (en) * | 2019-10-30 | 2020-02-14 | 南京工业大学 | Knowledge graph reasoning method and device, computer equipment and storage medium |
| WO2020147594A1 (en) * | 2019-01-16 | 2020-07-23 | 阿里巴巴集团控股有限公司 | Method, system, and device for obtaining expression of relationship between entities, and advertisement retrieval system |
| CN111753094A (en) * | 2019-03-27 | 2020-10-09 | 杭州海康威视数字技术股份有限公司 | Method and device for constructing event knowledge graph and method and device for determining event |
| CN112073415A (en) * | 2020-09-08 | 2020-12-11 | 北京天融信网络安全技术有限公司 | Method and device for constructing network security knowledge graph |
| CN112380355A (en) * | 2020-11-20 | 2021-02-19 | 华南理工大学 | Method for representing and storing time slot heterogeneous knowledge graph |
| CN112417163A (en) * | 2020-11-13 | 2021-02-26 | 中译语通科技股份有限公司 | Entity clue fragment-based candidate entity alignment method and device |
| CN113298426A (en) * | 2021-06-17 | 2021-08-24 | 华能澜沧江水电股份有限公司 | Knowledge graph driven dam safety evaluation weight dynamic drafting method and system |
| CN113641707A (en) * | 2018-01-25 | 2021-11-12 | 北京百度网讯科技有限公司 | Knowledge graph disambiguation method, device, equipment and storage medium |
| CN114219089A (en) * | 2021-11-11 | 2022-03-22 | 山东人才发展集团信息技术有限公司 | Construction method and equipment of new-generation information technology industry knowledge graph |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105468583A (en) * | 2015-12-09 | 2016-04-06 | 百度在线网络技术（北京）有限公司 | Entity relationship obtaining method and device |
| CN105630901A (en) * | 2015-12-21 | 2016-06-01 | 清华大学 | Knowledge graph representation learning method |
| CN105824802A (en) * | 2016-03-31 | 2016-08-03 | 清华大学 | Method and device for acquiring knowledge graph vectoring expression |
-
2017
- 2017-01-20 CN CN201710041593.XA patent/CN106909622A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105468583A (en) * | 2015-12-09 | 2016-04-06 | 百度在线网络技术（北京）有限公司 | Entity relationship obtaining method and device |
| CN105630901A (en) * | 2015-12-21 | 2016-06-01 | 清华大学 | Knowledge graph representation learning method |
| CN105824802A (en) * | 2016-03-31 | 2016-08-03 | 清华大学 | Method and device for acquiring knowledge graph vectoring expression |
Non-Patent Citations (2)
| Title |
|---|
| YANKAI LIN等: "Modeling Relation Paths for Representation Learning of Knowledge Bases", 《PROCEEDINGS OF THE 2015 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING》 * |
| YANTAO JIA等: "Locally Adaptive Translation for Knowledge Graph Embedding", 《PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE》 * |
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480191A (en) * | 2017-07-12 | 2017-12-15 | 清华大学 | A kind of entity alignment model of iteration |
| CN107480191B (en) * | 2017-07-12 | 2020-08-21 | 清华大学 | Iterative entity alignment model |
| CN107545033A (en) * | 2017-07-24 | 2018-01-05 | 清华大学 | A kind of computational methods based on the knowledge base entity classification for representing study |
| CN107545033B (en) * | 2017-07-24 | 2020-12-01 | 清华大学 | Knowledge base entity classification calculation method based on representation learning |
| CN107590237A (en) * | 2017-09-11 | 2018-01-16 | 桂林电子科技大学 | A kind of knowledge mapping based on dynamic translation principle represents learning method |
| CN107590237B (en) * | 2017-09-11 | 2021-04-30 | 桂林电子科技大学 | Knowledge graph representation learning method based on dynamic translation principle |
| CN107590139A (en) * | 2017-09-21 | 2018-01-16 | 桂林电子科技大学 | A kind of knowledge mapping based on circular matrix translation represents learning method |
| CN107590139B (en) * | 2017-09-21 | 2020-10-27 | 桂林电子科技大学 | Knowledge graph representation learning method based on cyclic matrix translation |
| CN107729444B (en) * | 2017-09-30 | 2021-01-12 | 桂林电子科技大学 | Knowledge graph-based personalized tourist attraction recommendation method |
| CN107729444A (en) * | 2017-09-30 | 2018-02-23 | 桂林电子科技大学 | Recommend method in a kind of personalized tourist attractions of knowledge based collection of illustrative plates |
| WO2019091019A1 (en) * | 2017-11-13 | 2019-05-16 | 平安科技（深圳）有限公司 | Knowledge graph processing method and device, computer device and computer storage medium |
| CN107943874A (en) * | 2017-11-13 | 2018-04-20 | 平安科技（深圳）有限公司 | Knowledge mapping processing method, device, computer equipment and storage medium |
| CN110019982B (en) * | 2017-12-05 | 2021-07-06 | 航天信息股份有限公司 | Node coordinate determination method and device |
| CN110019982A (en) * | 2017-12-05 | 2019-07-16 | 航天信息股份有限公司 | The determination method and device of node coordinate |
| CN108073711A (en) * | 2017-12-21 | 2018-05-25 | 北京大学深圳研究生院 | A kind of Relation extraction method and system of knowledge based collection of illustrative plates |
| CN108228877A (en) * | 2018-01-22 | 2018-06-29 | 北京师范大学 | Knowledge base complementing method and device based on study sort algorithm |
| CN108446769A (en) * | 2018-01-23 | 2018-08-24 | 深圳市阿西莫夫科技有限公司 | Knowledge mapping relation inference method, apparatus, computer equipment and storage medium |
| CN108446769B (en) * | 2018-01-23 | 2020-12-08 | 深圳市阿西莫夫科技有限公司 | Knowledge graph relation inference method, knowledge graph relation inference device, computer equipment and storage medium |
| CN113641707B (en) * | 2018-01-25 | 2023-07-21 | 北京百度网讯科技有限公司 | Knowledge graph disambiguation method, device, equipment and storage medium |
| CN113641707A (en) * | 2018-01-25 | 2021-11-12 | 北京百度网讯科技有限公司 | Knowledge graph disambiguation method, device, equipment and storage medium |
| CN108304933A (en) * | 2018-01-29 | 2018-07-20 | 北京师范大学 | A kind of complementing method and complementing device of knowledge base |
| CN108509483A (en) * | 2018-01-31 | 2018-09-07 | 北京化工大学 | The mechanical fault diagnosis construction of knowledge base method of knowledge based collection of illustrative plates |
| CN108510110A (en) * | 2018-03-13 | 2018-09-07 | 浙江禹控科技有限公司 | A kind of water table trend analysis method of knowledge based collection of illustrative plates |
| CN108846000A (en) * | 2018-04-11 | 2018-11-20 | 中国科学院软件研究所 | A kind of common sense semanteme map construction method and device based on supernode and the common sense complementing method based on connection prediction |
| CN108959472A (en) * | 2018-06-20 | 2018-12-07 | 桂林电子科技大学 | Knowledge mapping based on multistep relation path indicates learning method |
| CN108959472B (en) * | 2018-06-20 | 2021-11-19 | 桂林电子科技大学 | Knowledge graph representation learning method based on multi-step relation path |
| CN109146078B (en) * | 2018-07-19 | 2021-04-30 | 桂林电子科技大学 | Knowledge graph representation learning method based on dynamic path |
| CN109146078A (en) * | 2018-07-19 | 2019-01-04 | 桂林电子科技大学 | A kind of knowledge mapping expression learning method based on dynamic route |
| CN109492027B (en) * | 2018-11-05 | 2022-02-08 | 南京邮电大学 | Cross-community potential character relation analysis method based on weak credible data |
| CN109492027A (en) * | 2018-11-05 | 2019-03-19 | 南京邮电大学 | It is a kind of based on weak trust data across the potential character relation analysis method of community |
| WO2020147594A1 (en) * | 2019-01-16 | 2020-07-23 | 阿里巴巴集团控股有限公司 | Method, system, and device for obtaining expression of relationship between entities, and advertisement retrieval system |
| CN111753094A (en) * | 2019-03-27 | 2020-10-09 | 杭州海康威视数字技术股份有限公司 | Method and device for constructing event knowledge graph and method and device for determining event |
| CN111753094B (en) * | 2019-03-27 | 2024-02-02 | 杭州海康威视数字技术股份有限公司 | Method and device for constructing event knowledge graph and method and device for determining event |
| CN110796254B (en) * | 2019-10-30 | 2024-02-27 | 南京工业大学 | Knowledge graph reasoning method and device, computer equipment and storage medium |
| CN110796254A (en) * | 2019-10-30 | 2020-02-14 | 南京工业大学 | Knowledge graph reasoning method and device, computer equipment and storage medium |
| CN112073415A (en) * | 2020-09-08 | 2020-12-11 | 北京天融信网络安全技术有限公司 | Method and device for constructing network security knowledge graph |
| CN112417163A (en) * | 2020-11-13 | 2021-02-26 | 中译语通科技股份有限公司 | Entity clue fragment-based candidate entity alignment method and device |
| CN112380355A (en) * | 2020-11-20 | 2021-02-19 | 华南理工大学 | Method for representing and storing time slot heterogeneous knowledge graph |
| CN113298426A (en) * | 2021-06-17 | 2021-08-24 | 华能澜沧江水电股份有限公司 | Knowledge graph driven dam safety evaluation weight dynamic drafting method and system |
| CN114219089A (en) * | 2021-11-11 | 2022-03-22 | 山东人才发展集团信息技术有限公司 | Construction method and equipment of new-generation information technology industry knowledge graph |
| CN114219089B (en) * | 2021-11-11 | 2022-07-22 | 山东人才发展集团信息技术有限公司 | Construction method and equipment of new-generation information technology industry knowledge graph |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106909622A (en) | Knowledge mapping vector representation method, knowledge mapping relation inference method and system | |
| CN110147450B (en) | Knowledge complementing method and device for knowledge graph | |
| CN104298873B (en) | A kind of attribute reduction method and state of mind appraisal procedure based on genetic algorithm and rough set | |
| Lobato et al. | Multi-objective genetic algorithm for missing data imputation | |
| Rotshtein et al. | Fuzzy evidence in identification, forecasting and diagnosis | |
| CN108256065A (en) | Knowledge mapping inference method based on relationship detection and intensified learning | |
| CN109710741A (en) | A kind of mask method the problem of study based on deeply towards online answer platform | |
| CN113190688B (en) | Complex network link prediction method and system based on logical reasoning and graph convolution | |
| Fang | Intelligent online English teaching system based on SVM algorithm and complex network | |
| Owen | Hyperparameter Tuning with Python: Boost your machine learning model's performance via hyperparameter tuning | |
| Ilangkumaran et al. | Machine tool selection using AHP and VIKOR methodologies under fuzzy environment | |
| CN108460462A (en) | A kind of Interval neural networks learning method based on interval parameter optimization | |
| Zhang et al. | Multi-state deterioration prediction for infrastructure asset: Learning from uncertain data, knowledge and similar groups | |
| CN114118088A (en) | Document level entity relation extraction method and device based on hypergraph convolutional neural network | |
| Huang et al. | Research of data mining and web technology in university discipline construction decision support system based on MVC model | |
| Pan et al. | Learning first-order rules with relational path contrast for inductive relation reasoning | |
| Kalantari et al. | The unreasonable effectiveness of inverse reinforcement learning in advancing cancer research | |
| CN114817571A (en) | Method, medium, and apparatus for predicting achievement quoted amount based on dynamic knowledge graph | |
| Wang et al. | Graph neural network method for the intelligent selection of river system | |
| Kareem et al. | Evaluation Of Bayesian Network Structure Learning | |
| CN108122613A (en) | Health forecast method and apparatus based on health forecast model | |
| Ahmad et al. | A novel adaptive learning path method | |
| Niemeijer et al. | Constructing and predicting school advice for academic achievement: A comparison of item response theory and machine learning techniques | |
| CN114841148A (en) | Text recognition model training method, model training device and electronic equipment | |
| CN107480768A (en) | Bayesian network structure adaptive learning method and device, storage device and terminal device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170630 |