CN107885759A - A kind of knowledge mapping based on multiple-objection optimization represents learning method - Google Patents
A kind of knowledge mapping based on multiple-objection optimization represents learning method Download PDFInfo
- Publication number
- CN107885759A CN107885759A CN201611189474.0A CN201611189474A CN107885759A CN 107885759 A CN107885759 A CN 107885759A CN 201611189474 A CN201611189474 A CN 201611189474A CN 107885759 A CN107885759 A CN 107885759A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- entity
- relation
- vector
- 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
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2452—Query translation
-
- 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
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)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of knowledge mapping based on multiple-objection optimization and represents learning method, it uses the model based on translation between entity vector and relation vector, it using relationship map type definition in triple structure and best show interrelated between the vector of entity in triple and relation vector, and more parameters need not be introduced;Then entity vector sum relation vector is associated using loss function, and optimize the loss function, when reaching optimization aim, the vector of the vector sum relation of each entity in can learned knowledge collection of illustrative plates, so as to the contact preferably between presentation-entity and relation, and preferably it is applied in extensive knowledge mapping completion.The present invention solves the problems, such as that prior art is excessively simple or excessively complicated and can not represent the entity in knowledge mapping and relation well and can not be advantageously applied in extensive knowledge mapping have good practicality.
Description
Technical field
The present invention relates to knowledge mapping and machine learning techniques field, and in particular to a kind of knowledge based on multiple-objection optimization
Collection of illustrative plates represents learning method.
Background technology
With the rapid development of the technologies such as mobile Internet, Internet of Things, numerous new opplications are in a manner of unprecedented and speed
Degree is produced and have accumulated substantial amounts of data, and effective information how is filtered out from big data and has become many field faces
Opportunities and challenges, thereby produce knowledge mapping.
Knowledge mapping carries out structured modeling to the world, will be expressed as reality with abstract concept by specific things in the world
Body, knowledge network is constructed using the relation between entity, its essence is a digraph, with node on behalf entity, uses Lian Bianbiao
Show relation.Therefore, the knowledge in knowledge mapping is often represented with triple (h, r, t) i.e. (entity 1, relation, entity 2), right
Answer a company side in digraph and its 2 entities of connection.For example, leaf jasmine is this knowledge of the wife of Yao Ming, in knowledge
Triple relation (Ye Li, is ... wife, Yao Ming) can be utilized to represent in collection of illustrative plates.Knowledge mapping can be to entity and entity
Between relation establish more accurately inner link, be currently widely used for the various fields such as intelligent answer, information retrieval.Although
Existing knowledge mapping scale is very huge, but still very sparse, and knowledge mapping completion is current important study hotspot.
Similar with other kinds of large-scale data, extensive knowledge mapping also faces serious Sparse and calculates effect
Rate problem.And in recent years, represent that study obtains extensively in the research of extensive knowledge mapping using deep learning as representative
Concern, and make remarkable progress.Represent that study is intended to all entities in knowledge mapping and relationship map a to low-dimensional
, in continuous, real value vector space, the Sparse faced before solving well in knowledge mapping research and calculating
Efficiency.In existing representation of knowledge learning art, because optimization aim is relatively simple, do not consider sufficiently and sharp
With the triple structural information of knowledge mapping, and the difference of semantic space between entity and relation is only considered, without thin
The map type of relation in the consideration triple structural information of cause, and learning parameter is excessive, causes knowledge mapping to represent study side
In study, model excessively can not simply model method to complex relationship or model excessively complexity can not be applied to extensive knowledge graph
In spectrum.
The content of the invention
It is to be solved by this invention that to be that existing knowledge represents that learning method has optimization aim single and can not be applied to big
The problem of in scale knowledge mapping, there is provided a kind of knowledge mapping based on multiple-objection optimization represents learning method.
To solve the above problems, the present invention is achieved by the following technical solutions:
A kind of knowledge mapping based on multiple-objection optimization represents learning method, comprises the following steps:
Step 1, using the model based on translation between entity vector and relation vector, establish the triple knot of knowledge mapping
The correlation function of entity vector and relation vector in structure;
Step 2, the loss function for establishing entity vector and the correlation function of relation vector, and lose letter by minimizing
Number, with the vector representation of learn entity and relation, reaches optimization aim.
In step 1, used translation model is TransE translation models or TransH translation models.
In step 1, when optimization aim is t-r=α h,
It is using the correlation function f (h, r, t) constructed by TransE translation models:
It is using the correlation function f (h, r, t) constructed by TransH translation models:
Wherein:(h, r, t) represents the triple of knowledge mapping, and h represents head entity, and t represents tail entity, and r represents head entity h
Relation between tail entity t;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2Normal form;lrFor
Relation vector in relation r corresponding relations space, lhrIt is expressed as a normals of the entity h along hyperplane and projects to corresponding relation space
In head entity vector, ltrRepresent the tail entity vector that normals of the tail entity t along hyperplane is projected in corresponding relation space.
In step 1, when optimization aim is α t-r=h,
It is using the correlation function f (h, r, t) constructed by TransE translation models:
It is using the correlation function f (h, r, t) constructed by TransH translation models:
Wherein:(h, r, t) represents that the triple of knowledge mapping is positive example triple, and h represents head entity, and t represents tail entity,
R represents head entity h and tail entity t relation;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2
Normal form;lrFor the relation vector in relation r corresponding relations space, lhrA normals of the entity h along hyperplane is expressed as to project to pair
Answer the head entity vector in relation space, ltrRepresent the tail that normals of the tail entity t along hyperplane is projected in corresponding relation space
Entity vector.
In step 2, the loss function L of foundation is:
Wherein:[f(h,r,t)+γ-f(h′,r,t′)]+=max (0, f (h, r, t)+γ-f (h ', r, t '));γ is to set
Fixed marginal value;(h, r, t) represents that the triple of knowledge mapping is positive example triple, and h represents head entity, and t represents tail entity, r
Head entity h and tail entity t relation is represented, f (h, r, t) represents the correlation function of positive example triple, S(h,r,t)Represent positive example three
Tuple-set;(h ', r, t ') represents the negative example triple that random replacement turns around constructed by entity h and tail entity t, f (h ', r, t ')
Represent the correlation function of negative example triple, S(h′,r,t′)Represent negative example triplet sets.
Relation r map type include one-to-one (1-1), one-to-many (1-N), many-one (N-1) or multi-to-multi (N-N) this
4 types.
In step 2, during loss function is minimized:
When relation r type is many-one, by constantly adjusting h, t and r, make t-r as equal with α h as possible;
When relation r type is one-to-many, by constantly adjusting h, t and r, make h+r as equal with α t as possible;
When relation r type for one-to-one or multipair polymorphic type when, by constantly adjusting h, t and r, make h+r as far as possible with t
It is equal.
The present invention considers relationship map type, using relationship map type in knowledge mapping triple structure and defines more
The knowledge mapping of objective optimization represents learning method, best show the interrelated of entity and relation.The present invention is using real
Model based on translation between body vector and relation vector, utilizes relationship map type definition in triple structure and fine earth's surface
Show that entity is vectorial interrelated between relation vector in triple, and more parameters need not be introduced;Then damage is utilized
Lose function to associate entity vector sum relation vector, and optimize the loss function, when reaching optimization aim, it is possible to learn
The vector of the vector sum relation of each entity in knowledge mapping is obtained, so as to the contact preferably between presentation-entity and relation, and
Preferably it is applied in extensive knowledge mapping completion.The present invention solve prior art it is excessively simple or excessively complicated and can not be very
Entity and relation in good expressions knowledge mapping and the problem of can not be advantageously applied in extensive knowledge mapping, with good
Good practicality.
Brief description of the drawings
Fig. 1 is the exemplary plot of relation triple in knowledge mapping.
Fig. 2 is the flow chart that knowledge mapping of the present invention represents learning method.
Fig. 3 a are the exemplary plot that the triple table obtained according to prior art knowledge mapping expression learning method advises knowledge.
Fig. 3 b are the exemplary plot that the triple table obtained according to knowledge mapping of the present invention expression learning method advises knowledge.
Embodiment
For the goal of the invention of the present invention and technical scheme is more clearly understood, illustrate referring to the drawings, to the present invention
It is further described.
The present invention considers the triple structural information of knowledge mapping, and using typical (entity 1, relation, entity 2) three
The form of tuple represents knowledge, and relation is used for connecting two entities, portray the association between two entities.Fig. 1 is knowledge graph
The exemplary plot of typical triple in spectrum.Wherein, node such as " Bush ", " U.S. " and " Obama " that circle represents are to be real
Body, the Lian Bianru " president ", " birthplace " and " wife " between two entities etc. is all relation.In addition, it will be seen that
" president " relation pair has answered multiple entities, map type N-N.
The invention provides a kind of knowledge mapping based on multiple-objection optimization to represent learning method, referring to Fig. 2, this method bag
Include:Using entity vector relation vector between the model based on translation, using relationship map type in triple structure (including
1-1 simple relation, 1-N, N-1 and N-N complex relationship) define in triple (h, r, t) entity vector and relation vector it
Between it is interrelated, and more parameters need not be introduced;Entity vector sum relation vector is associated by loss function,
And loss function is minimized, with the vector representation of learn entity and relation, reach optimization aim.Can be preferably using the present invention
Contact between presentation-entity and relation, and can be applied in extensive knowledge mapping.
Embodiment one:
A kind of knowledge mapping based on multiple-objection optimization represents learning method, comprises the following steps:
Step 1, the knot using triple in model and knowledge mapping based on translation between entity vector and relation vector
Structure information, establish interrelated between entity vector and relation vector in triple (h, r, t).
Step 11, definition optimization aim are:
T-r=α h
Wherein, α represents parameter when relationship map type is N-1;α span is α ∈ (0,1).
Step 12, utilize the model based on translation, definition measurement relation r and entity pair between entity vector and relation vector
Correlation function f (h, r, t) between (h, t).
There is a variety of the model based on translation, between entity vector and relation vector for example, TransE and TransH etc..
If using TransE energy function, then f (h, r, t) can be defined as:
If using TransH energy function, then f (h, r, t) can be defined as:
Wherein, (h, r, t) represents the triple of knowledge mapping, and h represents head entity, and t represents tail entity, and r represents head entity h
Relation between tail entity t;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2Normal form;lrFor
Relation vector in relation r corresponding relations space, lhrIt is expressed as a normals of the entity h along hyperplane and projects to corresponding relation space
In head entity vector, ltrRepresent the tail entity vector that normals of the tail entity t along hyperplane is projected in corresponding relation space.
Step 2, by loss function entity vector is associated with relation vector, and minimize loss function, with
Entity vector sum relation vector is obtained, reaches optimization aim.
Step 21, definition loss function are:
Wherein, [f (h, r, t)+γ-f (h ', r, t ')]+=max (0, f (h, r, t)+γ-f (h ', r, t ')).γ is side
Actual value, positive example triple and negative example triple to be distinguished.(h, r, t) represents that the triple of knowledge mapping is positive example ternary
Group, h represent head entity, and t represents tail entity, and r represents head entity h and tail entity t relation, and f (h, r, t) represents positive example triple
Correlation function, S(h,r,t)Represent positive example triplet sets.(h ', r, t ') represents that random replacement turns around entity h and tail entity t institutes
The negative example triple of structure, f (h ', r, t ') represent the correlation function of negative example triple, S(h′,r,t′)Represent negative example triple collection
Close.
Step 22, the loss function is minimized, study obtains each entity vector sum relation vector in knowledge mapping.Most
The process of smallization loss function is exactly to reach the process of optimization aim.
During loss function is minimized:
When relation r type is N-1, by constantly adjusting h, t and r, make t-r as equal with α h as possible, reduction t-r with
Association between h;
When relation r type is 1-N, by constantly adjusting h, t and r, make h+r as equal with α t as possible;
When relation r is other types (1-1 and N-N type), by constantly adjusting h, t and r, make h+r as far as possible with t phases
Deng.
Embodiment two:
A kind of knowledge mapping based on multiple-objection optimization represents learning method, comprises the following steps:
Step 1, the knot using triple in model and knowledge mapping based on translation between entity vector and relation vector
Structure information, establish interrelated between entity vector and relation vector in triple (h, r, t).
Step 11, definition optimization aim are:
T-r=α h
Wherein, α represents parameter when relationship map type is N-1;α span is α ∈ (0,1).
Step 12, utilize the model based on translation, definition measurement relation r and entity pair between entity vector and relation vector
Correlation function f (h, r, t) between (h, t).
There is a variety of the model based on translation, between entity vector and relation vector for example, TransE and TransH etc..
If using TransE energy function, then f (h, r, t) can be defined as:
If using TransH energy function, then f (h, r, t) can be defined as:
Wherein, (h, r, t) represents the triple of knowledge mapping, and h represents head entity, and t represents tail entity, and r represents head entity h
Relation between tail entity t;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2Normal form;lrFor
Relation vector in relation r corresponding relations space, lhrIt is expressed as a normals of the entity h along hyperplane and projects to corresponding relation space
In head entity vector, ltrRepresent the tail entity vector that normals of the tail entity t along hyperplane is projected in corresponding relation space.
Step 2, by loss function entity vector is associated with relation vector, and minimize loss function, with
Entity vector sum relation vector is obtained, reaches optimization aim.
Step 21, definition loss function are:
Wherein, [f (h, r, t)+γ-f (h ', r, t ')]+=max (0, f (h, r, t)+γ-f (h ', r, t ')).γ is side
Actual value, positive example triple and negative example triple to be distinguished.(h, r, t) represents that the triple of knowledge mapping is positive example ternary
Group, h represent head entity, and t represents tail entity, and r represents head entity h and tail entity t relation, and f (h, r, t) represents positive example triple
Correlation function, S(h,r,t)Represent positive example triplet sets.(h ', r, t ') represents that random replacement turns around entity h and tail entity t institutes
The negative example triple of structure, f (h ', r, t ') represent the correlation function of negative example triple, S(h′,r,t′)Represent negative example triple collection
Close.
Step 22, the loss function is minimized, study obtains each entity vector sum relation vector in knowledge mapping.Most
The process of smallization loss function is exactly to reach the process of optimization aim.
During loss function is minimized:
When relation r type is N-1, by constantly adjusting h, t and r, make t-r as equal with α h as possible, reduction t-r with
Association between h;
When relation r type is 1-N, by constantly adjusting h, t and r, make h+r as equal with α t as possible;
When relation r is other types (1-1 and N-N type), by constantly adjusting h, t and r, make h+r as far as possible with t phases
Deng.
As a result emulate:
Learn to obtain each entity vector sum relation vector in knowledge mapping by the above method.Fig. 3 a are prior art
The triple table that method obtains advises the exemplary plot of knowledge.Fig. 3 b are to be represented according to knowledge mapping of the present invention based on multiple-objection optimization
The triple table that learning method obtains advises the exemplary plot of knowledge.In Fig. 3 a, the pass in knowledge mapping triple structure is not accounted for
It is map type, due to the U.S.+president=Obama, meanwhile, the U.S.+president=Bush, therefore finally draw Obama and Bu
It is assorted equal, but in fact, Obama and Bush also make a big difference in other side.And in Fig. 3 b, it is contemplated that knowledge mapping
Relationship map type in triple structure, when relation r is complex relationship, by weakening associating between h+r and t, both may be used
It is that can to express Bush again be US President to US President to express Obama, while and can distinguishes Obama and Bush.Thus
As can be seen that Fig. 3 a prior art is compared with Fig. 3 b present invention, knowledge mapping of the invention represents learning method, Ke Yigeng
Connecting each other between entity and relation is represented well, and considers that the model after relationship map type is more flexible.
The foregoing is only the present invention preferable case study on implementation, be not intended to limit the scope of the present invention within.
Claims (7)
1. a kind of knowledge mapping based on multiple-objection optimization represents learning method, it is characterized in that, comprise the following steps:
Step 1, using entity vector relation vector between the model based on translation, establish in the triple structure of knowledge mapping
The correlation function of entity vector and relation vector;
Step 2, the loss function for establishing entity vector and the correlation function of relation vector, and by minimizing loss function, with
The vector representation for entity and the relation of learning, reaches optimization aim.
2. a kind of knowledge mapping based on multiple-objection optimization according to claim 1 represents learning method, it is characterized in that,
In step 1, used translation model is TransE translation models or TransH translation models.
3. a kind of knowledge mapping based on multiple-objection optimization according to claim 1 or 2 represents learning method, its feature
It is, in step 1, when optimization aim is t-r=α h,
It is using the correlation function f (h, r, t) constructed by TransE translation models:
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<mi>&alpha;</mi>
<mi>h</mi>
<mo>+</mo>
<mi>r</mi>
<mo>-</mo>
<mi>t</mi>
<mo>|</mo>
<msub>
<mo>|</mo>
<mrow>
<msub>
<mi>L</mi>
<mn>1</mn>
</msub>
<mo>/</mo>
<msub>
<mi>L</mi>
<mn>2</mn>
</msub>
</mrow>
</msub>
</mrow>
It is using the correlation function f (h, r, t) constructed by TransH translation models:
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>&alpha;l</mi>
<mrow>
<mi>h</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>l</mi>
<mi>r</mi>
</msub>
<mo>-</mo>
<msub>
<mi>l</mi>
<mrow>
<mi>t</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mrow>
<msub>
<mi>L</mi>
<mn>1</mn>
</msub>
<mo>/</mo>
<msub>
<mi>L</mi>
<mn>2</mn>
</msub>
</mrow>
</msub>
</mrow>
Wherein:(h, r, t) represents the triple of knowledge mapping, and h represents head entity, and t represents tail entity, and r represents head entity h and tail
Entity t relation;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2Normal form;lrIt is corresponding for relation r
Relation vector in relation space, lhrIt is real to be expressed as the head that a normals of the entity h along hyperplane is projected in corresponding relation space
Body vector, ltrRepresent the tail entity vector that normals of the tail entity t along hyperplane is projected in corresponding relation space.
4. a kind of knowledge mapping based on multiple-objection optimization according to claim 1 or 2 represents learning method, its feature
It is, in step 1, when optimization aim is α t-r=h,
It is using the correlation function f (h, r, t) constructed by TransE translation models:
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>&alpha;l</mi>
<mrow>
<mi>h</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>l</mi>
<mi>r</mi>
</msub>
<mo>-</mo>
<msub>
<mi>l</mi>
<mrow>
<mi>t</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mrow>
<msub>
<mi>L</mi>
<mn>1</mn>
</msub>
<mo>/</mo>
<msub>
<mi>L</mi>
<mn>2</mn>
</msub>
</mrow>
</msub>
</mrow>
It is using the correlation function f (h, r, t) constructed by TransH translation models:
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>l</mi>
<mrow>
<mi>h</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>l</mi>
<mi>r</mi>
</msub>
<mo>-</mo>
<msub>
<mi>&alpha;l</mi>
<mrow>
<mi>t</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mrow>
<msub>
<mi>L</mi>
<mn>1</mn>
</msub>
<mo>/</mo>
<msub>
<mi>L</mi>
<mn>2</mn>
</msub>
</mrow>
</msub>
</mrow>
Wherein:(h, r, t) represents that the triple of knowledge mapping is positive example triple, and h represents head entity, and t represents tail entity, r tables
Show the relation between an entity h and tail entity t;α represents the parameter of different relationship map types;L1Represent L1Normal form, L2Represent L2
Normal form;lrFor the relation vector in relation r corresponding relations space, lhrA normals of the entity h along hyperplane is expressed as to project to pair
Answer the head entity vector in relation space, ltrRepresent the tail that normals of the tail entity t along hyperplane is projected in corresponding relation space
Entity vector.
5. a kind of knowledge mapping based on multiple-objection optimization according to claim 1 represents learning method, it is characterized in that,
In step 2, the loss function L of foundation is:
<mrow>
<mi>L</mi>
<mo>=</mo>
<munder>
<mi>&Sigma;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
<mo>&Element;</mo>
<msub>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msub>
</mrow>
</munder>
<munder>
<mi>&Sigma;</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>h</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<msup>
<mi>t</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
<mo>&Element;</mo>
<msubsup>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>&prime;</mo>
</msubsup>
</mrow>
</munder>
<msub>
<mrow>
<mo>&lsqb;</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&gamma;</mi>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>h</mi>
<mo>&prime;</mo>
</msup>
<mo>,</mo>
<mi>r</mi>
<mo>,</mo>
<msup>
<mi>t</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mo>+</mo>
</msub>
</mrow>
Wherein:[f(h,r,t)+γ-f(h′,r,t′)]+=max (0, f (h, r, t)+γ-f (h ', r, t '));γ is the side of setting
Actual value;(h, r, t) represents that the triple of knowledge mapping is positive example triple, and h represents head entity, and t represents tail entity, and r represents head
Entity h and tail entity t relation, f (h, r, t) represent the correlation function of positive example triple, S(h,r,t)Represent positive example triple collection
Close;(h ', r, t ') represents the negative example triple that random replacement turns around constructed by entity h and tail entity t, and f (h ', r, t ') represents negative
The correlation function of example triple, S(h′,r,t′)Represent negative example triplet sets.
6. a kind of knowledge mapping based on multiple-objection optimization according to claim 1 represents learning method, it is characterized in that, close
It is that r map type includes one-to-one, one-to-many, many-one and multi-to-multi this 4 type.
7. a kind of knowledge mapping based on multiple-objection optimization according to claim 6 represents learning method, it is characterized in that,
In step 2, during loss function is minimized:
When relation r type is many-one, by constantly adjusting h, t and r, make t-r as equal with α h as possible;
When relation r type is one-to-many, by constantly adjusting h, t and r, make h+r as equal with α t as possible;
When relation r type for one-to-one or multipair polymorphic type when, by constantly adjusting h, t and r, make h+r as far as possible with t phases
Deng.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611189474.0A CN107885759A (en) | 2016-12-21 | 2016-12-21 | A kind of knowledge mapping based on multiple-objection optimization represents learning method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611189474.0A CN107885759A (en) | 2016-12-21 | 2016-12-21 | A kind of knowledge mapping based on multiple-objection optimization represents learning method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107885759A true CN107885759A (en) | 2018-04-06 |
Family
ID=61770154
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611189474.0A Pending CN107885759A (en) | 2016-12-21 | 2016-12-21 | A kind of knowledge mapping based on multiple-objection optimization represents learning method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107885759A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108509654A (en) * | 2018-04-18 | 2018-09-07 | 上海交通大学 | The construction method of dynamic knowledge collection of illustrative plates |
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 |
CN108984745A (en) * | 2018-07-16 | 2018-12-11 | 福州大学 | A kind of neural network file classification method merging more knowledge mappings |
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 |
CN109165278A (en) * | 2018-09-07 | 2019-01-08 | 桂林电子科技大学 | It is a kind of that learning method is indicated based on entity and the knowledge mapping of relational structure information |
CN109255002A (en) * | 2018-09-11 | 2019-01-22 | 浙江大学 | A method of it is excavated using relation path and solves knowledge mapping alignment task |
CN110888942A (en) * | 2019-11-05 | 2020-03-17 | 天津大学 | Ontology inclusion axiom learning method based on linear programming |
CN111444181A (en) * | 2020-03-20 | 2020-07-24 | 腾讯科技(深圳)有限公司 | Knowledge graph updating method and device and electronic equipment |
CN111914094A (en) * | 2019-05-10 | 2020-11-10 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
CN112765398A (en) * | 2021-01-04 | 2021-05-07 | 珠海格力电器股份有限公司 | Information recommendation method and device and storage medium |
CN113033914A (en) * | 2021-04-16 | 2021-06-25 | 哈尔滨工业大学 | Entity and relation prediction method for machining process knowledge graph |
-
2016
- 2016-12-21 CN CN201611189474.0A patent/CN107885759A/en active Pending
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108509654A (en) * | 2018-04-18 | 2018-09-07 | 上海交通大学 | The construction method of dynamic knowledge collection of illustrative plates |
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 |
CN108984745A (en) * | 2018-07-16 | 2018-12-11 | 福州大学 | A kind of neural network file classification method merging more knowledge mappings |
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 |
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 |
CN109165278A (en) * | 2018-09-07 | 2019-01-08 | 桂林电子科技大学 | It is a kind of that learning method is indicated based on entity and the knowledge mapping of relational structure information |
CN109165278B (en) * | 2018-09-07 | 2021-11-09 | 桂林电子科技大学 | Knowledge graph representation learning method based on entity and relation structure information |
CN109255002B (en) * | 2018-09-11 | 2021-08-27 | 浙江大学 | Method for solving knowledge graph alignment task by utilizing relationship path mining |
CN109255002A (en) * | 2018-09-11 | 2019-01-22 | 浙江大学 | A method of it is excavated using relation path and solves knowledge mapping alignment task |
CN111914094A (en) * | 2019-05-10 | 2020-11-10 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
CN111914094B (en) * | 2019-05-10 | 2023-09-26 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
CN110888942A (en) * | 2019-11-05 | 2020-03-17 | 天津大学 | Ontology inclusion axiom learning method based on linear programming |
CN111444181B (en) * | 2020-03-20 | 2021-05-11 | 腾讯科技(深圳)有限公司 | Knowledge graph updating method and device and electronic equipment |
CN111444181A (en) * | 2020-03-20 | 2020-07-24 | 腾讯科技(深圳)有限公司 | Knowledge graph updating method and device and electronic equipment |
CN112765398A (en) * | 2021-01-04 | 2021-05-07 | 珠海格力电器股份有限公司 | Information recommendation method and device and storage medium |
CN113033914A (en) * | 2021-04-16 | 2021-06-25 | 哈尔滨工业大学 | Entity and relation prediction method for machining process knowledge graph |
CN113033914B (en) * | 2021-04-16 | 2022-03-25 | 哈尔滨工业大学 | Entity and relation prediction method for machining process knowledge graph |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107885759A (en) | A kind of knowledge mapping based on multiple-objection optimization represents learning method | |
CN108763376B (en) | Knowledge representation learning method for integrating relationship path, type and entity description information | |
Diudea et al. | Omega and related counting polynomials | |
CN109033129A (en) | Multi-source Information Fusion knowledge mapping based on adaptive weighting indicates learning method | |
CN105512289A (en) | Image retrieval method based on deep learning and Hash | |
CN109255002B (en) | Method for solving knowledge graph alignment task by utilizing relationship path mining | |
CN105631037A (en) | Image retrieval method | |
CN102842043B (en) | Particle swarm classifying method based on automatic clustering | |
CN113177132A (en) | Image retrieval method based on depth cross-modal hash of joint semantic matrix | |
CN107741568A (en) | A kind of lithium battery SOC estimation method that optimization RBF neural is shifted based on state | |
CN113065974B (en) | Link prediction method based on dynamic network representation learning | |
CN112000689B (en) | Multi-knowledge graph fusion method based on text analysis | |
CN106649550A (en) | Joint knowledge embedded method based on cost sensitive learning | |
CN104463208A (en) | Multi-view semi-supervised collaboration classification algorithm with combination of agreement and disagreement label rules | |
CN108764295A (en) | A kind of soft-measuring modeling method based on semi-supervised integrated study | |
CN108052683A (en) | A kind of knowledge mapping based on cosine measurement rule represents learning method | |
CN113987203A (en) | Knowledge graph reasoning method and system based on affine transformation and bias modeling | |
CN104504251A (en) | Community dividing method based on PageRank algorithm | |
CN112581368A (en) | Multi-robot grid map splicing method based on optimal map matching | |
CN113887698B (en) | Integral knowledge distillation method and system based on graph neural network | |
CN107590237B (en) | Knowledge graph representation learning method based on dynamic translation principle | |
CN109165278A (en) | It is a kind of that learning method is indicated based on entity and the knowledge mapping of relational structure information | |
CN111693868A (en) | Lithium battery state of charge estimation method based on density feature clustering integration | |
CN109472712A (en) | A kind of efficient Markov random field Combo discovering method strengthened based on structure feature | |
CN114596473A (en) | Network embedding pre-training method based on graph neural network hierarchical loss function |
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: 20180406 |