CN109947948A - A kind of knowledge mapping expression learning method and system based on tensor - Google Patents
A kind of knowledge mapping expression learning method and system based on tensor Download PDFInfo
- Publication number
- CN109947948A CN109947948A CN201910148591.XA CN201910148591A CN109947948A CN 109947948 A CN109947948 A CN 109947948A CN 201910148591 A CN201910148591 A CN 201910148591A CN 109947948 A CN109947948 A CN 109947948A
- Authority
- CN
- China
- Prior art keywords
- entity
- tensor
- code matrix
- mask code
- 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.)
- Granted
Links
- 238000013507 mapping Methods 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 106
- 238000012549 training Methods 0.000 claims abstract description 11
- 230000000694 effects Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000001186 cumulative effect Effects 0.000 claims description 6
- 230000009191 jumping Effects 0.000 claims description 6
- 230000007812 deficiency Effects 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000002585 base Substances 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000012458 free base Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Machine Translation (AREA)
Abstract
The invention discloses a kind of, and the knowledge mapping based on tensor indicates learning method and system, the present invention indicates that learning method only considers triple self-information, entity text description information, entity pictorial information, entity hierarchy information for existing knowledge mapping, and the problem that the local network structure information for having ignored map causes representation of knowledge learning effect bad, the present invention is in the expression study of knowledge mapping, it is contemplated that the local network structure information of each entity.Firstly, generating the mask code matrix of each entity according to the data set of knowledge mapping；Secondly, training tensor R, indicates the semantic information of entire data set, each entity can be acted on by mask code matrix and R, obtain the semantic tensor expression of each entity；Then, training vector r indicates the map vector mapped to each Entity Semantics tensor in certain semantic space；It is finally trained in specific semantic space, improving knowledge mapping indicates the accuracy of study.
Description
Technical field
The present invention relates to representation of knowledge learning areas, indicate more specifically to a kind of knowledge mapping based on tensor
The method and system of study.
Background technique
With the high speed development of science and technology, the arrival of big data era, information consumption has been increasingly becoming can not in people's life
Or scarce a part, but the energy of people is limited, in order to allow machine faster and better people can more be helped to collect knowledge, reason
Knowledge and processing knowledge are solved, representation of knowledge learning art comes into being, and the knowledge mapping of structuring contains a large amount of language
The expression study of adopted information, knowledge mapping is the tasks such as intelligent answer, Intelligent dialogue, Web search, semantic analysis and semantic reasoning
In important component.Knowledge mapping indicates that study is not only a research hotspot of academia, also receives in the field NLP
The concern of emphasis.
Knowledge mapping indicates application of the study in artificial intelligence field, can preferably meet people and disappear to information
Take, intelligent Answer System, conversational system, semantic search systematic difference, improves the efficiency that people obtain information, meanwhile, base
It can be used as the important auxiliary information of people's decision in the knowledge reasoning of big data.
Although knowledge mapping indicate study be widely used in many fields, various expression learning methods also by
It proposes, such as has the TransE of the pure structure of knowledge based map, there are also the DKRL etc. for merging entity description information.But current
Knowledge mapping indicates learning method, and there is also many problems.Be first knowledge reasoning accuracy it is not high, this allow expression learn
It is difficult to play a role in practical applications, followed by current many knowledge mappings are all manually to erect, either
WekiPedia or FreeBase needs to be solved in knowledge mapping that data are rare to ask with knowledge mapping completion not enough completely
Topic.
Summary of the invention
The technical problem to be solved in the present invention is that not high for the accuracy of knowledge reasoning in the prior art, this allows table
Dendrography habit is difficult to play a role in practical applications, and current many knowledge mappings are all manually to erect, also
It is sufficiently complete, it needs to be solved the technological deficiency of data scarcity problem in knowledge mapping with knowledge mapping completion, provides and be based on opening
The knowledge mapping of amount indicates learning method and system.
The present invention solves its technical problem, and the used knowledge mapping based on tensor indicates learning method, comprising as follows
Step:
S1, knowledge mapping is pre-processed, generates the mask code matrix Se of each entity；
S2, the tensor R for indicating knowledge mapping whole semantic information is obtained；
S3, each entity are multiplied with tensor R respectively by itself mask code matrix Se, obtain the semantic tensor of each entity；
It S4, is that one corresponding map vector v of each semantic information setting respectively will be each by the effect of each { v }
The semantic tensor of a entity is mapped in corresponding each semantic space, obtains the vector expression e in corresponding semantic space；
S5, each entity triple in each semantic space there are relationship is trained, obtains knowledge graph stave
The model that dendrography is practised, the model are used for input head entity and relationship, export corresponding tail entity；Wherein for any one language
The corresponding any entity triple (e1, r, e2) of adopted space r, training is so that e1+r is consistent with e2, i.e. e1+r ≈ e2；Wherein, e1
Head entity is represented, e2 represents tail entity, and the corresponding vector space of relationship r between e1 and e2 is the semantic space.
Preferably, the knowledge mapping of the invention based on tensor indicates in the step S1 of learning method, the entity of generation
Mask code matrix contains local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is pair
Angular moment battle array, dimension are m × m；
(2) single order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
The side that body is connected directly, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
Body two jumps connected side, and just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity together
Mask code matrix.
Preferably, knowledge mapping of the invention based on tensor indicates in learning method, the corresponding single order of the entity
Corresponding position specifically refers in mask code matrix or second-order matrix:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
Preferably, the knowledge mapping of the invention based on tensor indicates to utilize tensor pair in the step S3 of learning method
Entity is indicated, and is the semantic information branch by the various levels of entity in specified riding position.
According to another aspect of the present invention, the present invention is to solve its technical problem, the used knowledge graph based on tensor
Spectral representation learning system, which is characterized in that comprise the following steps:
Preprocessing module generates the mask code matrix Se of each entity for pre-processing to knowledge mapping；
Tensor obtains module, for obtaining the tensor R for indicating knowledge mapping whole semantic information；
Semantic tensor obtains module, is multiplied, is obtained every with tensor R respectively by itself mask code matrix Se for each entity
The semantic tensor of a entity；
Vector obtains module, for setting a corresponding map vector v for each semantic information, passes through each { v }
Effect, the semantic tensor of each entity is mapped in corresponding each semantic space respectively, obtains corresponding semantic space
In vector express e；
Model training module, for being trained to each entity triple in each semantic space there are relationship,
Obtaining knowledge mapping indicates that the model of study, the model are used for input head entity and relationship, export corresponding tail entity；Wherein
Any entity triple (e1, r, e2) corresponding for any one semantic space r, training is so that e1+r is consistent with e2；Wherein,
E1 represents head entity, and e2 represents tail entity, and the corresponding vector space of relationship r between e1 and e2 is the semantic space.
Preferably, the knowledge mapping of the invention based on tensor indicates in learning system, in preprocessing module, generation
Entity mask code matrix contains local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is pair
Angular moment battle array, dimension are m × m；
(2) single order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
The side that body is connected directly, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
Body two jumps connected side, and just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity together
Mask code matrix.
Preferably, knowledge mapping of the invention based on tensor indicates in learning system, the corresponding single order of the entity
Corresponding position specifically refers in mask code matrix or second-order matrix:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
Preferably, the knowledge mapping of the invention based on tensor indicates in learning system, and semantic tensor obtains in module,
Entity is indicated using tensor, is the semantic information branch by the various levels of entity in specified riding position.
Previous knowledge mapping indicates that learning method often only considered the description information of entity, pictorial information, context
The information such as keyword and hypernym, these information are incomplete.The present invention is from the angle of figure, by the local area network of each entity
Network structural information imparts entity, is that entity information is more comprehensively and abundant.The addition of entity local network structure information makes up
The deficiency expressed in entity structure, solves that because structural information deficiency causes knowledge mapping to do semantic reasoning accuracy rate is low to ask
Topic.Text is not available for mathematical operation in general, this allows computer to be difficult to handle.Text conversion is tensor by the present invention
And vector form, text can be performed mathematical calculations.The basic component units of knowledge mapping are entity and relationship, and the present invention will
Entity and transformation into numerical space, and the semantic relation between entity and entity still can in numerical space body
It is existing.The present invention, using the local network structure information of each entity as a part of entity information, allows entity to believe from the angle of figure
Breath is more perfect, and is indicated in the form of tensor to entity, and the semantic information of entity different level is specified in semantic tensor
Specified position, the expression that can be entity are more specifically more accurate.To allow the expression of knowledge mapping to obtain network structure letter
The supplement of breath is more comprehensively complete.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is the specific steps figure that the knowledge mapping based on tensor indicates one embodiment of learning method；
Fig. 2 is that the knowledge mapping of the invention based on tensor indicates that the mask code matrix of study generates schematic diagram.
Specific embodiment
For a clearer understanding of the technical characteristics, objects and effects of the present invention, now control attached drawing is described in detail
A specific embodiment of the invention.
With reference to Fig. 1, the specific steps of one embodiment of learning method are indicated for the knowledge mapping of the invention based on tensor
Figure.Knowledge mapping indicate the destination of study be by triple entity and relationship map into the continuous Real-value space of low-dimensional,
Knowledge mapping used by the present embodiment based on tensor indicates that learning method comprises the following steps:
S1, knowledge mapping is pre-processed, generates the mask code matrix Se of each entity.The entity mask code matrix packet of generation
Contain local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is pair
Angular moment battle array, dimension are m × m；
(2) single order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
The side that body is connected directly, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
Body two jumps connected side, and just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity together
Mask code matrix.
Preferably, corresponding position specifically refers in the corresponding single order mask code matrix of entity or second-order matrix:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
It is following above-mentioned principle to be illustrated in conjunction with Fig. 2.
(1) count first a shared r1, r2 in entire map, r3 ..., 6 kinds of relationships such as r6, set entity mask code matrix
Dimension be 6 × 6, and be diagonal matrix；
(2) the single order relationship one of entity A shares 5, including 2 r2,2 r3,1 r4.Therefore the single order mask of entity A
In matrix Se: Se (1,1)=0, Se (2,2)=2, Se (3,3)=2, Se (4,4)=1, Se (5,5)=0, Se (6,6)=0；
(3) the second order relationship one of entity A shares 6, including 1 r1,1 r3,2 r4,2 r5,1 r6.Therefore real
In the second order mask code matrix Se of body A: Se (1,1)=0.5, Se (2,2)=0, Se (3,3)=0.5, Se (4,4)=1, Se (5,5)
=1, Se (6,6)=0.5；
(4) mask code matrix of entity be single order mask code matrix and second order mask code matrix and: Se (1,1)=0.5, Se (2,
2)=2, Se (3,3)=2.5, Se (4,4)=2, Se (5,5)=1, Se (6,6)=0.5.
S2, the tensor R for indicating knowledge mapping whole semantic information is obtained.
S3, each entity are multiplied with tensor R respectively by itself mask code matrix Se, obtain the semantic tensor of each entity.
In step S3, unlike existing knowledge mapping expression learning method, the present invention is indicated entity using tensor, will
The semantic information branch of the various levels of entity is in specified riding position.
It S4, is that one corresponding map vector v of each semantic information setting respectively will be each by the effect of each { v }
The semantic tensor of a entity is mapped in corresponding each semantic space, obtains the vector expression e in corresponding semantic space.
S5, each entity triple in each semantic space there are relationship is trained, obtains knowledge graph stave
The model that dendrography is practised, the model are used for input head entity and relationship, export corresponding tail entity；Wherein for any one language
The corresponding any entity triple (e1, r, e2) of adopted space r, training is so that e1+r is consistent with e2；Wherein, e1 represents head entity,
E2 represents tail entity, and the corresponding vector space of relationship r between e1 and e2 is the semantic space.
About semantic space and semantic tensor, semantic space refers to the corresponding vector space of some relationship, semanteme
Amount refers to the tensor representation of entity, it is the tensor before being mapped to semantic space, contains multiple semantic informations.For example,
(Xiao Ming, father, Zhang San), (Xiao Ming, the form master, Li Si), the entity tensor of Xiao Ming contain semantemes such as " father " " forms master "
Information, and relationship " father ", " form master " have corresponding semantic space respectively.
According to another aspect of the present invention, the present invention is to solve its technical problem, the used knowledge graph based on tensor
Spectral representation learning system, which is characterized in that comprise the following steps:
Preprocessing module generates the mask code matrix Se of each entity for pre-processing to knowledge mapping；
Tensor obtains module, for obtaining the tensor R for indicating knowledge mapping whole semantic information；
Semantic tensor obtains module, is multiplied, is obtained every with tensor R respectively by itself mask code matrix Se for each entity
The semantic tensor of a entity；
Vector obtains module, for setting a corresponding map vector v for each semantic information, passes through each { v }
Effect, the semantic tensor of each entity is mapped in corresponding each semantic space respectively, obtains corresponding semantic space
In vector express e；
Model training module, for being trained to each entity triple in each semantic space there are relationship,
Obtaining knowledge mapping indicates that the model of study, the model are used for input head entity and relationship, export corresponding tail entity；Wherein
Any entity triple (e1, r, e2) corresponding for any one semantic space r, training is so that e1+r is consistent with e2；Wherein,
E1 represents head entity, and e2 represents tail entity, and the corresponding vector space of relationship r between e1 and e2 is the semantic space.
Preferably, the knowledge mapping of the invention based on tensor indicates in learning system, in preprocessing module, generation
Entity mask code matrix contains local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is pair
Angular moment battle array, dimension are m × m；
(2) single order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
The side that body is connected directly, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the reality
Body two jumps connected side, and just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity together
Mask code matrix.
Preferably, knowledge mapping of the invention based on tensor indicates in learning system, the corresponding single order of the entity
Corresponding position specifically refers in mask code matrix or second-order matrix:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
Preferably, the knowledge mapping of the invention based on tensor indicates in learning system, and semantic tensor obtains in module,
Entity is indicated using tensor, is the semantic information branch by the various levels of entity in specified riding position.
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form, all of these belong to the protection of the present invention.
Claims (8)
1. a kind of knowledge mapping based on tensor indicates learning method, which is characterized in that comprise the following steps:
S1, knowledge mapping is pre-processed, generates the mask code matrix Se of each entity；
S2, the tensor R for indicating knowledge mapping whole semantic information is obtained；
S3, each entity are multiplied with tensor R respectively by itself mask code matrix Se, obtain the semantic tensor of each entity；
S4, a corresponding map vector v is set for each semantic information, by the effect of each { v }, respectively by each reality
The semantic tensor of body is mapped in corresponding each semantic space, obtains the vector expression e in corresponding semantic space；
S5, each entity triple in each semantic space there are relationship is trained, obtaining knowledge mapping indicates to learn
The model of habit, the model are used for input head entity and relationship, export corresponding tail entity；It is wherein semantic empty for any one
Between corresponding any entity triple (e1, r, e2), training so that e1+r is consistent with e2；Wherein, e1 represents head entity, and e2 is represented
Tail entity, the corresponding vector space of relationship r between e1 and e2 is the semantic space.
2. the knowledge mapping according to claim 1 based on tensor indicates learning method, which is characterized in that in step S1,
The entity mask code matrix of generation contains local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is to angular moment
Battle array, dimension are m × m；
(2) calculate single order entity mask code matrix: in knowledge mapping, to any one entity, every traversal one is straight with the entity
Connected side is connect, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the entity two
Connected side is jumped, just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity mask together
Matrix.
3. the knowledge mapping according to claim 2 based on tensor indicates learning method, which is characterized in that the entity pair
Corresponding position specifically refers in the single order mask code matrix or second-order matrix answered:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
4. the knowledge mapping according to claim 1 based on tensor indicates learning method, which is characterized in that in step S3,
Entity is indicated using tensor, is the semantic information branch by the various levels of entity in specified riding position.
5. a kind of knowledge mapping based on tensor indicates learning system, which is characterized in that comprise the following steps:
Preprocessing module generates the mask code matrix Se of each entity for pre-processing to knowledge mapping；
Tensor obtains module, for obtaining the tensor R for indicating knowledge mapping whole semantic information；
Semantic tensor obtains module, is multiplied respectively with tensor R for each entity by itself mask code matrix Se, obtains each reality
The semantic tensor of body；
Vector obtains module, for setting a corresponding map vector v for each semantic information, passes through the work of each { v }
With the semantic tensor of each entity is mapped in corresponding each semantic space respectively, is obtained in corresponding semantic space
Vector expresses e；
Model training module is obtained for being trained to each entity triple in each semantic space there are relationship
Knowledge mapping indicates that the model of study, the model are used for input head entity and relationship, export corresponding tail entity；Wherein for
The corresponding any entity triple (e1, r, e2) of any one semantic space r, training is so that e1+r is consistent with e2；Wherein, e1 generation
Gauge outfit entity, e2 represent tail entity, and the corresponding vector space of relationship r between e1 and e2 is the semantic space.
6. the knowledge mapping according to claim 5 based on tensor indicates learning system, which is characterized in that preprocessing module
In, the entity mask code matrix of generation contains local network structure information, the specific steps are as follows:
(1) it sets mask code matrix dimension: obtaining the relationship type sum m that knowledge mapping includes, entity mask code matrix is to angular moment
Battle array, dimension are m × m；
(2) calculate single order entity mask code matrix: in knowledge mapping, to any one entity, every traversal one is straight with the entity
Connected side is connect, just corresponding position starts from scratch every time cumulative 1 in the corresponding single order mask code matrix of entity；
(3) second order entity mask code matrix is calculated: in knowledge mapping, to any one entity, every traversal one and the entity two
Connected side is jumped, just corresponding position starts from scratch and adds up 0.5 every time in the corresponding second order mask code matrix of entity；
(4) the corresponding single order mask code matrix of each entity and second order mask code matrix are added to and obtain final entity mask together
Matrix.
7. the knowledge mapping according to claim 6 based on tensor indicates learning system, which is characterized in that the entity pair
Corresponding position specifically refers in the single order mask code matrix or second-order matrix answered:
For any entity, directly connected side shares K item, this K side includes: k1 item relationship r1, k2 item relationship
R2 ..., km item relationship rm, wherein k1+k2+ ...+km=K, then the element that line n n-th arranges in mask code matrix is kn, n=1,
... and m 2,；
For any entity, with secondly jumping the side being connected shares P item, this P side includes: p1 item relationship r1, p2 item relationship
R2 ..., pm item relationship rm, wherein p1+p2+ ...+pm=P, n < m, then the element that line n n-th arranges in mask code matrix is kn/2, n
=1,2 ... and m.
8. the knowledge mapping according to claim 5 based on tensor indicates learning system, which is characterized in that semantic tensor obtains
In modulus block, entity is indicated using tensor, is the semantic information branch by the various levels of entity in specified placement
Position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910148591.XA CN109947948B (en) | 2019-02-28 | 2019-02-28 | Knowledge graph representation learning method and system based on tensor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910148591.XA CN109947948B (en) | 2019-02-28 | 2019-02-28 | Knowledge graph representation learning method and system based on tensor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109947948A true CN109947948A (en) | 2019-06-28 |
| CN109947948B CN109947948B (en) | 2021-04-09 |
Family
ID=67008118
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910148591.XA Expired - Fee Related CN109947948B (en) | 2019-02-28 | 2019-02-28 | Knowledge graph representation learning method and system based on tensor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109947948B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021056389A1 (en) * | 2019-09-27 | 2021-04-01 | Intel Corporation | Methods and apparatus to process machine learning model in web-browser environment |
| CN112612900A (en) * | 2020-12-10 | 2021-04-06 | 大连理工大学 | Knowledge graph guided multi-scene image generation method |
| CN113194334A (en) * | 2021-04-16 | 2021-07-30 | 厦门智瞳科技有限公司 | Image processing method and device for protecting privacy, terminal and storage medium |
| CN114330323A (en) * | 2022-03-08 | 2022-04-12 | 成都数联云算科技有限公司 | Entity relationship joint extraction method and device, computer terminal and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120158791A1 (en) * | 2010-12-21 | 2012-06-21 | Microsoft Corporation | Feature vector construction |
| US8862662B2 (en) * | 2012-10-29 | 2014-10-14 | The Boeing Company | Determination of latent interactions in social networks |
| US20170038750A1 (en) * | 2014-04-22 | 2017-02-09 | Siemens Aktiengesellschaft | Method, controller, and computer program product for controlling a target system |
| CN106528610A (en) * | 2016-09-28 | 2017-03-22 | 厦门理工学院 | Knowledge graph representation learning method based on path tensor decomposition |
| CN108572981A (en) * | 2017-03-10 | 2018-09-25 | 中移（杭州）信息技术有限公司 | A kind of information recommendation method and device |
-
2019
- 2019-02-28 CN CN201910148591.XA patent/CN109947948B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120158791A1 (en) * | 2010-12-21 | 2012-06-21 | Microsoft Corporation | Feature vector construction |
| US8862662B2 (en) * | 2012-10-29 | 2014-10-14 | The Boeing Company | Determination of latent interactions in social networks |
| US20170038750A1 (en) * | 2014-04-22 | 2017-02-09 | Siemens Aktiengesellschaft | Method, controller, and computer program product for controlling a target system |
| CN106528610A (en) * | 2016-09-28 | 2017-03-22 | 厦门理工学院 | Knowledge graph representation learning method based on path tensor decomposition |
| CN108572981A (en) * | 2017-03-10 | 2018-09-25 | 中移（杭州）信息技术有限公司 | A kind of information recommendation method and device |
Non-Patent Citations (3)
| Title |
|---|
| TROUILLON T, DANCE C R, GAUSSIER É, ET AL.: "Knowledge graph completion via complex tensor factorization", 《THE JOURNAL OF MACHINE LEARNING RESEARCH》 * |
| WANG Q, MAO Z, WANG B, ET AL.: "Knowledge graph embedding: A survey of approaches and applications", 《IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING》 * |
| 刘峤 等: "基于双层随机游走的关系推理算法", 《计算机学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021056389A1 (en) * | 2019-09-27 | 2021-04-01 | Intel Corporation | Methods and apparatus to process machine learning model in web-browser environment |
| CN112612900A (en) * | 2020-12-10 | 2021-04-06 | 大连理工大学 | Knowledge graph guided multi-scene image generation method |
| CN113194334A (en) * | 2021-04-16 | 2021-07-30 | 厦门智瞳科技有限公司 | Image processing method and device for protecting privacy, terminal and storage medium |
| CN113194334B (en) * | 2021-04-16 | 2023-06-06 | 厦门智瞳科技有限公司 | Privacy-protecting image processing method, privacy-protecting image processing device, terminal and storage medium |
| CN114330323A (en) * | 2022-03-08 | 2022-04-12 | 成都数联云算科技有限公司 | Entity relationship joint extraction method and device, computer terminal and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109947948B (en) | 2021-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902171B (en) | Text relation extraction method and system based on hierarchical knowledge graph attention model | |
| CN109947948A (en) | A kind of knowledge mapping expression learning method and system based on tensor | |
| Liu et al. | Sign language recognition with long short-term memory | |
| CN108009285B (en) | Forest Ecology man-machine interaction method based on natural language processing | |
| CN106980858B (en) | Language text detection and positioning system and language text detection and positioning method using same | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| CN106777275A (en) | Entity attribute and property value extracting method based on many granularity semantic chunks | |
| CN106650789A (en) | Image description generation method based on depth LSTM network | |
| CN107679462A (en) | A kind of depth multiple features fusion sorting technique based on small echo | |
| CN104899298A (en) | Microblog sentiment analysis method based on large-scale corpus characteristic learning | |
| CN107766320A (en) | A kind of Chinese pronoun resolution method for establishing model and device | |
| CN114092707A (en) | Image text visual question answering method, system and storage medium | |
| CN109710744A (en) | A kind of data matching method, device, equipment and storage medium | |
| CN105975457A (en) | Information classification prediction system based on full-automatic learning | |
| CN110188348A (en) | A kind of Chinese language processing model and method based on deep neural network | |
| CN109712108A (en) | It is a kind of that vision positioning method is directed to based on various distinctive candidate frame generation network | |
| CN107357785A (en) | Theme feature word abstracting method and system, feeling polarities determination methods and system | |
| CN114463545A (en) | Image semantic segmentation algorithm and system based on multi-channel depth weighted aggregation | |
| CN108829810A (en) | File classification method towards healthy public sentiment | |
| CN107066979A (en) | A kind of human motion recognition method based on depth information and various dimensions convolutional neural networks | |
| CN107480194A (en) | The construction method and system of the multi-modal automatic learning model of the representation of knowledge | |
| CN113987167A (en) | Dependency perception graph convolutional network-based aspect-level emotion classification method and system | |
| CN113837290A (en) | Unsupervised unpaired image translation method based on attention generator network | |
| CN103065009B (en) | Intelligent design system and method of traffic sign lines | |
| CN117033609A (en) | Text visual question-answering method, device, computer equipment and storage medium |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210409 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |