CN106354710A - Neural network relation extracting method - Google Patents
Neural network relation extracting method Download PDFInfo
- Publication number
- CN106354710A CN106354710A CN201610685532.2A CN201610685532A CN106354710A CN 106354710 A CN106354710 A CN 106354710A CN 201610685532 A CN201610685532 A CN 201610685532A CN 106354710 A CN106354710 A CN 106354710A
- Authority
- CN
- China
- Prior art keywords
- sentence
- vector
- entity
- pair
- relation
- 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
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a neural network relation extracting method. A method for relation extracting of a neural network of an attention mechanism is selected based on the levels of sentences. The method comprises the specific steps that for each sentence and a pair of entities related to the sentence, sentence vector representation of the pair of entities is established with a convolutional neural network; sentence vector representation expressing the relation of the pair of entities is selected with a set sentence level attention mechanism, and comprehensive sentence vector representation of the pair of entities is obtained; the relation of the pair of entities is predicted according to the comprehensive sentence vector representation of the pair of entities. In this way, by means of the method, the interference of noise in remote supervision data can be reduced in neural network relation extracting, information of different sentences can be considered at the same time, the stability of a model can be improved, and good practicability is achieved.
Description
Technical field
The present invention relates to natural language processing technique field, particularly to a kind of neutral net Relation extraction method.
Background technology
Develop rapidly with society, come into the information explosion epoch at present, have the new data of magnanimity to produce daily.Mutually
As presently the most easily information acquisition platform, user is increasingly urgent with the demand of conclusion to effective information screening, such as networking
What extracts effective information from mass data becomes a difficult problem.In order to solve this difficult problem it is proposed that knowledge mapping concept, knowledge
The noun of the specific term in mass data and things is expressed as entity by collection of illustrative plates, specific term be in the world all persons,
Name, title or team etc., the internal relation between entity are expressed as relation it is intended to mass data is expressed as profit between entity
With relation as bridge ternary relation group.For example, " Beijing is the capital of China " this data, then utilizes in knowledge mapping
Tlv triple relation is expressed as: Beijing, is ... capital, China.
Although the knowledge mapping having built up contains more than one hundred million data, compared to endless data, it is still remote
It is far from perfect.Automatically perfect in order to constantly carry out to knowledge mapping, employ a lot of technology.Wherein just there is Relation extraction skill
Art, this technology can automatically extract from natural language text structured data.
At present, usually adopt the Relation extraction technology based on supervised learning in Relation extraction technology, this technology needs
Substantial amounts of artificial mark training data, unusual time and effort consuming.For this problem, the Relation extraction technology based on remote supervisory
Propose and can automatically generate the technology of training data by contrasting between plain text and configured knowledge base,
If the Relation extraction technology based on remote supervisory assumes that two entities simultaneously appear in a sentence, then this sentence is just
Express the relation between this two entities, by this hypothesis, it is possible to use all in the knowledge base of setting comprise this two realities
The sentence of body is as the training data of this two entity relationships.But, trained based on the Relation extraction technology of remote supervisory
Data has that one serious it is simply that the noise of the training data producing is very serious, because not all comprises
The sentence of two entities all can reflect the relation between this two entities.In order to reduce the noise of training data, producing training
By the method for probability graph model, the Relation extraction technology based on remote supervisory is processed during data, that is, pass through joint probability
The relation changed between sentence and two entities solves.
Further, if Relation extraction technology adopts non-neural net method, because non-neutral net introduces
Natural language processing instrument extracts relationship characteristic, also inevitably introduces some noises.At present, Relation extraction technology adopts
The model realization of neutral net, but, the model of neutral net is not also gone on processing the noise that remote supervisory data produces
Effective method.Therefore, how to solve application in remote supervisory data for the neural net method be one highly important
Problem.
Content of the invention
In view of this, the embodiment of the present invention provides a kind of neutral net Relation extraction method, and the method can be supervised to long-range
The noise superintending and directing data is processed, and improves Relation extraction effect.
According to above-mentioned purpose, the present invention is achieved in that
A kind of neutral net Relation extraction method, the method includes:
A pair of entity to each sentence and its correlation, using convolutional neural networks build the pair of entity sentence to
Amount represents；
Sentence level attention mechanism using setting selects the sentence of the relation that have expressed the pair of inter-entity therein
Subvector represents, the synthetic sentence subvector obtaining the pair of entity represents；
The Relationship Prediction carrying out the pair of inter-entity is represented according to the synthetic sentence subvector of the pair of entity, obtains pre-
Measured value.
It is preferred that the sentence vector representation of the pair of entity of described structure is:
Term vector is built respectively to a pair of entity in sentence, described term vector includes the spelling of meaning of a word vector and position vector
Connect information；
By convolution, pond and non-enjoy operation described term vector is converted to sentence vector representation.
It is preferred that described by convolution, pond with non-enjoy operation and described term vector is converted to sentence vector representation is:
First, will be operated between term vector sequence w and convolution matrix w, convolution operation can be long by one
Spend the sliding window for l local feature is extracted.Define q firstiIt is all of term vector in i-th window interior
Concatenation information:
qi=wi-l+1:i(1≤i≤m+l-1)；
Secondly, the term vector of all words beyond sentence boundary is regarded as and seems null vector；
Then, the i-th dimension characterizing definition that convolution obtains is: pi=[wq+b]i；Wherein b is bias vector；
Again, the i-th dimension of sentence vector representation is defined as by pond: [x]i=max (pi)；Or adopt segmentation pond,
Every one-dimensional characteristic p that convolution is obtainediFrom the beginning entity and tail entity are divided into three sections of (pi1,pi2,pi3), then respectively to each section
Carry out pond: [x]ij=max (pij)；
Finally, [x]iIt is defined as [x]ijSplicing, non-linearization is being carried out to vector x, is obtaining final sentence vector table
Show.
The pair of inter-entity is have expressed using the sentence level attention mechanism selection arranging is therein described at an angle
Relation sentence vector representation, obtaining the process that the synthetic sentence subvector of the pair of entity represents is:
The resultant vector defining all sentences represents；
The weight of each sentence vector is defined by the way of average；
Define the weight of each sentence vector representation using the sentence level attention mechanism of setting.
It is preferred that the described resultant vector defining all sentences is expressed as:
Assume all sentences expression vector be x1,x2,…,xn, the resultant vector of sentence is represented and is defined as all sentences
The weighted sum of vector representation vector:
Wherein αiIt is defined as each sentence vector xiWeight.
It is preferred that the described vectorial weight of each sentence that defined by the way of average is:
Assume the contribution that all sentence vector representations represent for the described synthetic sentence subvector to entity finally giving
Equalization, the described synthetic sentence subvector to entity obtaining represents the meansigma methodss as each sentence vector representation, public as follows
Shown in formula:
It is preferred that the described sentence level attention mechanism using setting defines the weight of each sentence vector representation
For:
The bilinear function that selects using setting weighs the vector representation x of each sentenceiDescribed in finally to be predicted
To the correlation degree between relation r of inter-entity, equation below:
ei=xiar；
Wherein, a is diagonal parameter matrix, and r is the vector representation of relationship by objective (RBO) r of inquiry, by Selective attention power machine
The weight formulating each sentence vector representation adopted is as follows:
It is preferred that the described synthetic sentence subvector according to the pair of entity represents the relation carrying out the pair of inter-entity
Prediction, obtaining predictive value is:
First, define the probability of the relation of final prediction, described definition of probability is:
Wherein nrFor the number of all relation species, o is the input of final neutral net, is defined as follows:
O=ms+d
Wherein d is bias vector, and m is all relation vector representing matrixs；
Secondly, by stochastic gradient descent, minimize evaluation function, all parameters are learnt and updates, comprising: be logical
Cross minimum evaluation function and learn all of parameter.Evaluation function formula is as follows:
Wherein s is the number of all training sentence set, and θ represents all of model
Parameter, carries out parameter optimization using stochastic gradient descent algorithm.
Finally, prevent from training over-fitting by dropout, comprising: using dropout mechanism, final output o enters one
Step is defined as:Wherein h is the vector of Bernoulli Jacob's distribution that every one-dimension probability is p, and final is owned
Sentence expression vector is multiplied by p, obtains
As can be seen from the above scheme, the embodiment of the present invention proposes the nerve based on sentence level Selective attention power mechanism
Cyberrelationship abstracting method, particularly as follows: a pair of entity to each sentence and its correlation, is built described using convolutional neural networks
The sentence vector representation of a pair of entity；Have expressed the pair of reality using the sentence level attention mechanism selection of setting is therein
The sentence vector representation of the relation between body, the synthetic sentence subvector obtaining the pair of entity represents；According to the pair of entity
Synthetic sentence subvector represent the Relationship Prediction carrying out the pair of inter-entity.So, the embodiment of the present invention not only can be in god
Reduce the interference of noise in remote supervisory data in extracting through cyberrelationship, the information of different sentences can also be considered simultaneously, carry
The stability of high model, has good practicality.
Brief description
Fig. 1 is the method flow diagram that neutral net provided in an embodiment of the present invention extracts relation；
Fig. 2 is that the selection of the sentence level attention mechanism using setting provided in an embodiment of the present invention is therein have expressed institute
After stating the sentence vector representation of the relation to inter-entity, obtain the schematic diagram that the described synthetic sentence subvector to entity represents；
Fig. 3 is the method specific example schematic diagram described in Fig. 1 provided in an embodiment of the present invention.
Specific embodiment
For making the objects, technical solutions and advantages of the present invention become more apparent, develop simultaneously embodiment referring to the drawings, right
The present invention is described in further detail.
Fig. 1 is the method flow diagram that neutral net provided in an embodiment of the present invention extracts relation, and it concretely comprises the following steps:
Step 101, a pair of entity to each sentence and its correlation, build the pair of entity using convolutional neural networks
Sentence vector representation；
Step 102, select therein to have expressed the pair of inter-entity using the sentence level attention mechanism of setting
The sentence vector representation of relation, the synthetic sentence subvector obtaining the pair of entity represents；
Step 103, represent that the relation carrying out the pair of inter-entity is pre- according to the synthetic sentence subvector of the pair of entity
Survey, obtain predictive value.
In the method, the detailed process of described step 101 is:
Step 1011, the word to the input of sentence are indicated, and obtain term vector；
In this step, the input of convolutional neural networks is all words of sentence.First by all words in sentence
It is converted into continuous vector representation.Here, the word of each input is converted into the vector in a term vector matrix；
Further, this step is also distinguished to the position of two entities using position vector.Here, term vector is used for identifying
The syntactic and semantic information of each word, is obtained using text depth representing model (word2vec) study；Position vector is used for
The positional information of presentation-entity, is defined as the vector representation of the mutual alignment difference between each word and head entity, tail entity.?
Whole term vector is defined as the vectorial concatenation information with position vector of the meaning of a word that word2vec learns.
Step 1012, convolution of passing through, the word of input is represented the vector table being converted into sentence by pondization and nonlinear operation
Show；
In this step, convolution operation is defined as the operation between term vector sequence w and convolution matrix w, convolution
Operation can be extracted to local feature by the sliding window that a length is l.Define q firstiIt is in i-th window
The concatenation information of all of term vector in portion:
qi=wi-l+1:i(1≤i≤m+l-1)
Due to the border of sentence may be exceeded during window sliding, add some blank words on the border of sentence, that is,
Say, the term vector of all words beyond sentence boundary is regarded as and seems null vector.Then, the i-th dimension characterizing definition that convolution obtains
For:
pi=[wq+b]i
Wherein b is bias vector；
Further, the i-th dimension of the expression of final sentence is defined as by pond:
[x]i=max (pi)
Additionally, in Relation extraction task it is contemplated that after the position of two entities, pondization can be improved to point further
Duan Chihua, every one-dimensional characteristic p that convolution is obtainediFrom the beginning entity and tail entity are divided into three sections of (pi1,pi2,pi3), then right respectively
Each section carries out pond:
[x]ij=max (pij)
Then [x]iIt is defined as [x]ijSplicing.
In this step last, is carrying out such as the non-linearization of tanh function, is obtaining final sentence vector to vector x
Represent.
In the method, described step 102 particularly as follows:
Step 1021, the resultant vector of all sentences of definition represent；
Very intuitively it is assumed that the expression vector of all sentences is x1,x2,…,xn, the resultant vector of sentence is represented definition
Weighted sum for all sentence vector representations vector:
Wherein αiIt is defined as each sentence vector xiWeight；
Step 1022, define the weight of each sentence vector by the way of average；
In this step it is assumed that all sentence vector representations are for the described synthetic sentence subvector to entity finally giving
The contribution representing is impartial, then the described synthetic sentence subvector to entity finally giving represents as each sentence vector table
The meansigma methodss shown, shown in equation below:
Step 1023, the weight of each sentence vector representation is defined using the sentence level attention mechanism of setting；
In this step, weigh the vector representation x of each sentence based on the function of inquiry using oneiWant with final
Correlation degree between described relation r to inter-entity of prediction.Here to be defined using selection bilinear function:
ei=xiar；
Wherein, a is diagonal parameter matrix, and r is the vector representation of relationship by objective (RBO) r of inquiry, can pass through Selective attention
The weight that power mechanism defines each sentence vector representation is as follows:
Due to considering the information of described relation r to inter-entity finally to be predicted, using the sentence level note of setting
After meaning power mechanism is selected, the final weight of noise sentence can effectively be reduced by the weight of each sentence vector.
In the method, the further step of described step 103:
Step 1031, the probability of the relation of the final prediction of definition；
In this step, by give all sentences set and described to entity, relation that each of final prediction obtains
Definition of probability be:
Wherein nrFor the number of all relation species, o is the input of final neutral net, is defined as follows:
O=ms+d
Wherein d is bias vector, and m is all relation vector representing matrixs；
Step 1032, pass through stochastic gradient descent, minimize evaluation function, all parameters are learnt and are updated；
Specifically, learn all of parameter by minimizing evaluation function.Evaluation function formula is as follows:
Wherein s is the number of all training sentence set, and θ is represented all of model parameter, calculated using stochastic gradient descent
Method carries out parameter optimization.
Step 1033, by dropout prevent train over-fitting；
In this step, in order to prevent from training over-fitting, employ existing dropout mechanism, final output o enters
One step is defined as:
Wherein h is the vector of Bernoulli Jacob's distribution that every one-dimension probability is p.In final prediction, by final all sentences
Subrepresentation vector is multiplied by p, that is,
Fig. 2 is that the selection of the sentence level attention mechanism using setting provided in an embodiment of the present invention is therein have expressed institute
After stating the sentence vector representation of the relation to inter-entity, obtain the schematic diagram that the described synthetic sentence subvector to entity represents, its
In, m1,m2..., mc is each sentence vector representation, r1,r2,…,rcFor representing to the relation of inter-entity each described, incite somebody to action both
After being associated, using formulaEach sentence vector representation is defined by Selective attention power mechanism, obtains
Final r.
Fig. 3 is the method specific example schematic diagram described in Fig. 1 provided in an embodiment of the present invention, as illustrated, from down to up,
A pair of entity in one sentence according to the method described in Fig. 1, after coarse grain layer by layer, finally given described to inter-entity
Relationship Prediction value.
As can be seen from the above scheme, the embodiment of the present invention proposes the nerve based on sentence level Selective attention power mechanism
Cyberrelationship abstracting method, not only can reduce the interference of noise in remote supervisory data in neutral net Relation extraction, also
The information of different sentences can be considered simultaneously, improve the stability of model, there is good practicality.
The object, technical solutions and advantages of the present invention are further described, institute by above act preferred embodiment
It should be understood that the foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all the present invention's
Spirit and principle within, any modification, equivalent and improvement of being made etc., should be included in protection scope of the present invention it
Interior.
Claims (8)
1. a kind of neutral net Relation extraction method is it is characterised in that the method includes:
A pair of entity to each sentence and its correlation, builds the sentence vector table of the pair of entity using convolutional neural networks
Show；
Using setting sentence level attention mechanism select the relation that have expressed the pair of inter-entity therein sentence to
Amount represents, the synthetic sentence subvector obtaining the pair of entity represents；
The Relationship Prediction carrying out the pair of inter-entity is represented according to the synthetic sentence subvector of the pair of entity, is predicted
Value.
2. the method for claim 1 is it is characterised in that the sentence vector representation of the pair of entity of described structure is:
Term vector is built respectively to a pair of entity in sentence, described term vector includes the splicing letter of meaning of a word vector and position vector
Breath；
By convolution, pond and non-enjoy operation described term vector is converted to sentence vector representation.
3. the method for claim 1 is it is characterised in that described operated institute's predicate by convolution, pond and non-enjoying
Vector is converted to sentence vector representation:
First, will be operated between term vector sequence w and convolution matrix w, convolution operation can be l by a length
Sliding window local feature is extracted.Define q firstiIt is the splicing of all of term vector in i-th window interior
Information:
qi=wi-l+1:i(1≤i≤m+l-1)；
Secondly, the term vector of all words beyond sentence boundary is regarded as and seems null vector；
Then, the i-th dimension characterizing definition that convolution obtains is: pi=[wq+b]i；Wherein b is bias vector；
Again, the i-th dimension of sentence vector representation is defined as by pond: [x]i=max (pi)；Or adopt segmentation pond, will roll up
The long-pending every one-dimensional characteristic p obtainingiFrom the beginning entity and tail entity are divided into three sections of (pi1,pi2,pi3), then respectively each section is carried out
Chi Hua: [x]ij=max (pij)；
Finally, [x]iIt is defined as [x]ijSplicing, non-linearization is being carried out to vector x, is obtaining final sentence vector representation.
4. the method for claim 1 is it is characterised in that the described sentence level attention mechanism using setting selects it
In the relation that have expressed the pair of inter-entity sentence vector representation, obtain the synthetic sentence subvector table of the pair of entity
The process shown is:
The resultant vector defining all sentences represents；
The weight of each sentence vector is defined by the way of average；
Define the weight of each sentence vector representation using the sentence level attention mechanism of setting.
5. method as claimed in claim 4 is it is characterised in that the resultant vector of all sentences of described definition is expressed as:
Assume all sentences expression vector be x1,x2,…,xn, the resultant vector of sentence is represented and is defined as all sentence vectors
The weighted sum of expression vector:
Wherein αiIt is defined as each sentence vector xiWeight.
6. method as claimed in claim 4 is it is characterised in that described each sentence vector of being defined by the way of average
Weight is:
Assume that the contribution that all sentence vector representations represent for the described synthetic sentence subvector to entity finally giving is impartial,
The described synthetic sentence subvector to entity obtaining represents the meansigma methodss as each sentence vector representation, equation below institute
Show:
7. method as claimed in claim 4 is it is characterised in that the described sentence level attention mechanism definition using setting is every
The weight of one sentence vector representation is:
The bilinear function that selects using setting weighs the vector representation x of each sentenceiFinally to be predicted described to entity
Between relation r between correlation degree, equation below:
ei=xiar；
Wherein, a is diagonal parameter matrix, and r is the vector representation of relationship by objective (RBO) r of inquiry, fixed by Selective attention power mechanism
The weight of each sentence vector representation adopted is as follows:
8. the method for claim 1 is it is characterised in that the described synthetic sentence subvector according to the pair of entity represents
Carry out the Relationship Prediction of the pair of inter-entity, obtaining predictive value is:
First, define the probability of the relation of final prediction, described definition of probability is:
Wherein nrFor the number of all relation species, o is the input of final neutral net, is defined as follows:
O=ms+d
Wherein d is bias vector, and m is all relation vector representing matrixs；
Secondly, by stochastic gradient descent, minimize evaluation function, all parameters are learnt and updates, comprising: by
Littleization evaluation function learns all of parameter.Evaluation function formula is as follows:
Wherein s is the number of all training sentence set, and θ represents all of model parameter,
Parameter optimization is carried out using stochastic gradient descent algorithm.
Finally, prevent from training over-fitting by dropout, comprising: using dropout mechanism, final output o is fixed further
Justice is:Wherein h is the vector of Bernoulli Jacob's distribution that every one-dimension probability is p, by final all sentences
Represent that vector is multiplied by p, obtain
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610685532.2A CN106354710A (en) | 2016-08-18 | 2016-08-18 | Neural network relation extracting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610685532.2A CN106354710A (en) | 2016-08-18 | 2016-08-18 | Neural network relation extracting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106354710A true CN106354710A (en) | 2017-01-25 |
Family
ID=57843370
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610685532.2A Pending CN106354710A (en) | 2016-08-18 | 2016-08-18 | Neural network relation extracting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106354710A (en) |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844738A (en) * | 2017-02-14 | 2017-06-13 | 华南理工大学 | The sorting technique of Junker relation between food materials based on neutral net |
| CN106970981A (en) * | 2017-03-28 | 2017-07-21 | 北京大学 | A kind of method that Relation extraction model is built based on transfer matrix |
| CN107133211A (en) * | 2017-04-26 | 2017-09-05 | 中国人民大学 | A kind of composition methods of marking based on notice mechanism |
| CN107194422A (en) * | 2017-06-19 | 2017-09-22 | 中国人民解放军国防科学技术大学 | A kind of convolutional neural networks relation sorting technique of the forward and reverse example of combination |
| CN107220237A (en) * | 2017-05-24 | 2017-09-29 | 南京大学 | A kind of method of business entity's Relation extraction based on convolutional neural networks |
| CN107239446A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of intelligence relationship extracting method based on neutral net Yu notice mechanism |
| CN107256228A (en) * | 2017-05-02 | 2017-10-17 | 清华大学 | Answer selection system and method based on structuring notice mechanism |
| CN107273800A (en) * | 2017-05-17 | 2017-10-20 | 大连理工大学 | A kind of action identification method of the convolution recurrent neural network based on attention mechanism |
| CN107273349A (en) * | 2017-05-09 | 2017-10-20 | 清华大学 | A kind of entity relation extraction method and server based on multilingual |
| CN107392229A (en) * | 2017-06-21 | 2017-11-24 | 清华大学 | A kind of network representation method based on the Relation extraction that most gears to the needs of the society |
| CN107766506A (en) * | 2017-10-20 | 2018-03-06 | 哈尔滨工业大学 | A kind of more wheel dialog model construction methods based on stratification notice mechanism |
| CN107944559A (en) * | 2017-11-24 | 2018-04-20 | 国家计算机网络与信息安全管理中心 | A kind of entity relationship automatic identifying method and system |
| CN108304911A (en) * | 2018-01-09 | 2018-07-20 | 中国科学院自动化研究所 | Knowledge Extraction Method and system based on Memory Neural Networks and equipment |
| CN108399180A (en) * | 2017-02-08 | 2018-08-14 | 腾讯科技（深圳）有限公司 | A kind of knowledge mapping construction method, device and server |
| CN108491680A (en) * | 2018-03-07 | 2018-09-04 | 安庆师范大学 | Drug relationship abstracting method based on residual error network and attention mechanism |
| CN108536754A (en) * | 2018-03-14 | 2018-09-14 | 四川大学 | Electronic health record entity relation extraction method based on BLSTM and attention mechanism |
| CN108563653A (en) * | 2017-12-21 | 2018-09-21 | 清华大学 | A kind of construction method and system for knowledge acquirement model in knowledge mapping |
| CN108733792A (en) * | 2018-05-14 | 2018-11-02 | 北京大学深圳研究生院 | A kind of entity relation extraction method |
| CN108875809A (en) * | 2018-06-01 | 2018-11-23 | 大连理工大学 | The biomedical entity relationship classification method of joint attention mechanism and neural network |
| CN109062897A (en) * | 2018-07-26 | 2018-12-21 | 苏州大学 | Sentence alignment method based on deep neural network |
| CN109376864A (en) * | 2018-09-06 | 2019-02-22 | 电子科技大学 | A kind of knowledge mapping relation inference algorithm based on stacking neural network |
| CN109522920A (en) * | 2018-09-18 | 2019-03-26 | 义语智能科技（上海）有限公司 | Training method and equipment based on the synonymous discrimination model for combining semantic feature |
| CN109522557A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Training method, device and the readable storage medium storing program for executing of text Relation extraction model |
| CN109597894A (en) * | 2018-09-30 | 2019-04-09 | 阿里巴巴集团控股有限公司 | A kind of correlation model generation method and device, a kind of data correlation method and device |
| CN109635124A (en) * | 2018-11-30 | 2019-04-16 | 北京大学 | A kind of remote supervisory Relation extraction method of combination background knowledge |
| CN109670050A (en) * | 2018-12-12 | 2019-04-23 | 科大讯飞股份有限公司 | A kind of entity relationship prediction technique and device |
| CN109902171A (en) * | 2019-01-30 | 2019-06-18 | 中国地质大学（武汉） | Text Relation extraction method and system based on layering knowledge mapping attention model |
| CN109992629A (en) * | 2019-02-28 | 2019-07-09 | 中国科学院计算技术研究所 | A kind of neural network Relation extraction method and system of fusion entity type constraint |
| CN110196978A (en) * | 2019-06-04 | 2019-09-03 | 重庆大学 | A kind of entity relation extraction method for paying close attention to conjunctive word |
| CN110210016A (en) * | 2019-04-25 | 2019-09-06 | 中国科学院计算技术研究所 | Bilinearity neural network Deceptive news detection method and system based on style guidance |
| CN110245292A (en) * | 2019-05-28 | 2019-09-17 | 华东师范大学 | A kind of natural language Relation extraction method based on neural network filtering noise characteristic |
| CN110263019A (en) * | 2019-06-18 | 2019-09-20 | 中南民族大学 | Construction method, device and the storage medium of entity relation extraction model |
| CN110275960A (en) * | 2019-06-11 | 2019-09-24 | 中国电子科技集团公司电子科学研究院 | Representation method and system based on the knowledge mapping and text information for censuring sentence |
| CN110309512A (en) * | 2019-07-05 | 2019-10-08 | 北京邮电大学 | A kind of Chinese grammer error correction method thereof based on generation confrontation network |
| CN110309516A (en) * | 2019-05-30 | 2019-10-08 | 清华大学 | Training method, device and the electronic equipment of Machine Translation Model |
| CN110580340A (en) * | 2019-08-29 | 2019-12-17 | 桂林电子科技大学 | neural network relation extraction method based on multi-attention machine system |
| CN110852066A (en) * | 2018-07-25 | 2020-02-28 | 清华大学 | Multi-language entity relation extraction method and system based on confrontation training mechanism |
| CN111091007A (en) * | 2020-03-23 | 2020-05-01 | 杭州有数金融信息服务有限公司 | Method for identifying relationships among multiple enterprises based on public sentiment and enterprise portrait |
| CN111192692A (en) * | 2020-01-02 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Entity relationship determination method and device, electronic equipment and storage medium |
| CN111191461A (en) * | 2019-06-06 | 2020-05-22 | 北京理工大学 | Remote supervision relation extraction method based on course learning |
| CN111274812A (en) * | 2018-12-03 | 2020-06-12 | 阿里巴巴集团控股有限公司 | Character relation recognition method, device and storage medium |
| CN111831783A (en) * | 2020-07-07 | 2020-10-27 | 北京北大软件工程股份有限公司 | Chapter-level relation extraction method |
| CN112328784A (en) * | 2019-08-05 | 2021-02-05 | 上海智臻智能网络科技股份有限公司 | Data information classification method and device |
| CN112347263A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN112347265A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN112347196A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Entity relation extraction method and device based on neural network |
| CN112463982A (en) * | 2020-11-27 | 2021-03-09 | 华东师范大学 | Relationship extraction method based on explicit and implicit entity constraint |
| CN113011161A (en) * | 2020-12-29 | 2021-06-22 | 中国航天科工集团第二研究院 | Method for extracting human and pattern association relation based on deep learning and pattern matching |
| CN114218956A (en) * | 2022-01-24 | 2022-03-22 | 平安科技（深圳）有限公司 | Relation extraction method and system based on neural network and remote supervision |
| CN114757179A (en) * | 2022-04-13 | 2022-07-15 | 成都信息工程大学 | Entity relationship joint extraction method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009017464A1 (en) * | 2007-07-31 | 2009-02-05 | Agency For Science, Technology And Research | Relation extraction system |
| CN104021115A (en) * | 2014-06-13 | 2014-09-03 | 北京理工大学 | Chinese comparative sentence recognizing method and device based on neural network |
-
2016
- 2016-08-18 CN CN201610685532.2A patent/CN106354710A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009017464A1 (en) * | 2007-07-31 | 2009-02-05 | Agency For Science, Technology And Research | Relation extraction system |
| CN104021115A (en) * | 2014-06-13 | 2014-09-03 | 北京理工大学 | Chinese comparative sentence recognizing method and device based on neural network |
Non-Patent Citations (2)
| Title |
|---|
| YANKAI LIN ET AL.: "Neural Relation Extraction with Selective Attention over Instances", 《PROCEEDINGS OF THE 54TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS》 * |
| 刘丽佳 等: "基于LM算法的领域概念实体属性关系抽取", 《中文信息学报》 * |
Cited By (72)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108399180A (en) * | 2017-02-08 | 2018-08-14 | 腾讯科技（深圳）有限公司 | A kind of knowledge mapping construction method, device and server |
| CN108399180B (en) * | 2017-02-08 | 2021-11-26 | 腾讯科技（深圳）有限公司 | Knowledge graph construction method and device and server |
| CN106844738A (en) * | 2017-02-14 | 2017-06-13 | 华南理工大学 | The sorting technique of Junker relation between food materials based on neutral net |
| CN106844738B (en) * | 2017-02-14 | 2019-07-16 | 华南理工大学 | The classification method of Junker relationship between food materials neural network based |
| CN106970981A (en) * | 2017-03-28 | 2017-07-21 | 北京大学 | A kind of method that Relation extraction model is built based on transfer matrix |
| CN106970981B (en) * | 2017-03-28 | 2021-01-19 | 北京大学 | Method for constructing relation extraction model based on transfer matrix |
| CN107133211A (en) * | 2017-04-26 | 2017-09-05 | 中国人民大学 | A kind of composition methods of marking based on notice mechanism |
| CN107256228A (en) * | 2017-05-02 | 2017-10-17 | 清华大学 | Answer selection system and method based on structuring notice mechanism |
| CN107273349B (en) * | 2017-05-09 | 2019-11-22 | 清华大学 | A kind of entity relation extraction method and server based on multilingual |
| CN107273349A (en) * | 2017-05-09 | 2017-10-20 | 清华大学 | A kind of entity relation extraction method and server based on multilingual |
| CN107273800A (en) * | 2017-05-17 | 2017-10-20 | 大连理工大学 | A kind of action identification method of the convolution recurrent neural network based on attention mechanism |
| CN107220237A (en) * | 2017-05-24 | 2017-09-29 | 南京大学 | A kind of method of business entity's Relation extraction based on convolutional neural networks |
| CN107239446A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of intelligence relationship extracting method based on neutral net Yu notice mechanism |
| WO2018218707A1 (en) * | 2017-05-27 | 2018-12-06 | 中国矿业大学 | Neural network and attention mechanism-based information relation extraction method |
| CN107194422A (en) * | 2017-06-19 | 2017-09-22 | 中国人民解放军国防科学技术大学 | A kind of convolutional neural networks relation sorting technique of the forward and reverse example of combination |
| CN107392229A (en) * | 2017-06-21 | 2017-11-24 | 清华大学 | A kind of network representation method based on the Relation extraction that most gears to the needs of the society |
| CN107766506A (en) * | 2017-10-20 | 2018-03-06 | 哈尔滨工业大学 | A kind of more wheel dialog model construction methods based on stratification notice mechanism |
| CN107944559A (en) * | 2017-11-24 | 2018-04-20 | 国家计算机网络与信息安全管理中心 | A kind of entity relationship automatic identifying method and system |
| CN108563653A (en) * | 2017-12-21 | 2018-09-21 | 清华大学 | A kind of construction method and system for knowledge acquirement model in knowledge mapping |
| CN108563653B (en) * | 2017-12-21 | 2020-07-31 | 清华大学 | Method and system for constructing knowledge acquisition model in knowledge graph |
| CN108304911A (en) * | 2018-01-09 | 2018-07-20 | 中国科学院自动化研究所 | Knowledge Extraction Method and system based on Memory Neural Networks and equipment |
| CN108491680A (en) * | 2018-03-07 | 2018-09-04 | 安庆师范大学 | Drug relationship abstracting method based on residual error network and attention mechanism |
| CN108536754A (en) * | 2018-03-14 | 2018-09-14 | 四川大学 | Electronic health record entity relation extraction method based on BLSTM and attention mechanism |
| CN108733792B (en) * | 2018-05-14 | 2020-12-01 | 北京大学深圳研究生院 | Entity relation extraction method |
| CN108733792A (en) * | 2018-05-14 | 2018-11-02 | 北京大学深圳研究生院 | A kind of entity relation extraction method |
| CN108875809A (en) * | 2018-06-01 | 2018-11-23 | 大连理工大学 | The biomedical entity relationship classification method of joint attention mechanism and neural network |
| CN110852066B (en) * | 2018-07-25 | 2021-06-01 | 清华大学 | Multi-language entity relation extraction method and system based on confrontation training mechanism |
| CN110852066A (en) * | 2018-07-25 | 2020-02-28 | 清华大学 | Multi-language entity relation extraction method and system based on confrontation training mechanism |
| CN109062897A (en) * | 2018-07-26 | 2018-12-21 | 苏州大学 | Sentence alignment method based on deep neural network |
| CN109376864A (en) * | 2018-09-06 | 2019-02-22 | 电子科技大学 | A kind of knowledge mapping relation inference algorithm based on stacking neural network |
| CN109522920A (en) * | 2018-09-18 | 2019-03-26 | 义语智能科技（上海）有限公司 | Training method and equipment based on the synonymous discrimination model for combining semantic feature |
| CN109522920B (en) * | 2018-09-18 | 2020-10-13 | 义语智能科技（上海）有限公司 | Training method and device of synonymy discriminant model based on combination of semantic features |
| CN109597894B (en) * | 2018-09-30 | 2023-10-03 | 创新先进技术有限公司 | Correlation model generation method and device, and data correlation method and device |
| CN109597894A (en) * | 2018-09-30 | 2019-04-09 | 阿里巴巴集团控股有限公司 | A kind of correlation model generation method and device, a kind of data correlation method and device |
| CN109522557B (en) * | 2018-11-16 | 2021-07-16 | 中山大学 | Training method and device of text relation extraction model and readable storage medium |
| CN109522557A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Training method, device and the readable storage medium storing program for executing of text Relation extraction model |
| CN109635124A (en) * | 2018-11-30 | 2019-04-16 | 北京大学 | A kind of remote supervisory Relation extraction method of combination background knowledge |
| CN111274812B (en) * | 2018-12-03 | 2023-04-18 | 阿里巴巴集团控股有限公司 | Figure relation recognition method, equipment and storage medium |
| CN111274812A (en) * | 2018-12-03 | 2020-06-12 | 阿里巴巴集团控股有限公司 | Character relation recognition method, device and storage medium |
| CN109670050A (en) * | 2018-12-12 | 2019-04-23 | 科大讯飞股份有限公司 | A kind of entity relationship prediction technique and device |
| CN109902171A (en) * | 2019-01-30 | 2019-06-18 | 中国地质大学（武汉） | Text Relation extraction method and system based on layering knowledge mapping attention model |
| CN109992629A (en) * | 2019-02-28 | 2019-07-09 | 中国科学院计算技术研究所 | A kind of neural network Relation extraction method and system of fusion entity type constraint |
| CN109992629B (en) * | 2019-02-28 | 2021-08-06 | 中国科学院计算技术研究所 | Neural network relation extraction method and system fusing entity type constraints |
| CN110210016A (en) * | 2019-04-25 | 2019-09-06 | 中国科学院计算技术研究所 | Bilinearity neural network Deceptive news detection method and system based on style guidance |
| CN110245292A (en) * | 2019-05-28 | 2019-09-17 | 华东师范大学 | A kind of natural language Relation extraction method based on neural network filtering noise characteristic |
| CN110309516A (en) * | 2019-05-30 | 2019-10-08 | 清华大学 | Training method, device and the electronic equipment of Machine Translation Model |
| CN110196978A (en) * | 2019-06-04 | 2019-09-03 | 重庆大学 | A kind of entity relation extraction method for paying close attention to conjunctive word |
| CN111191461B (en) * | 2019-06-06 | 2021-08-03 | 北京理工大学 | Remote supervision relation extraction method based on course learning |
| CN111191461A (en) * | 2019-06-06 | 2020-05-22 | 北京理工大学 | Remote supervision relation extraction method based on course learning |
| CN110275960A (en) * | 2019-06-11 | 2019-09-24 | 中国电子科技集团公司电子科学研究院 | Representation method and system based on the knowledge mapping and text information for censuring sentence |
| CN110275960B (en) * | 2019-06-11 | 2021-09-14 | 中国电子科技集团公司电子科学研究院 | Method and system for expressing knowledge graph and text information based on named sentence |
| CN110263019A (en) * | 2019-06-18 | 2019-09-20 | 中南民族大学 | Construction method, device and the storage medium of entity relation extraction model |
| CN110263019B (en) * | 2019-06-18 | 2021-08-31 | 中南民族大学 | Method and device for constructing entity relationship extraction model and storage medium |
| CN110309512A (en) * | 2019-07-05 | 2019-10-08 | 北京邮电大学 | A kind of Chinese grammer error correction method thereof based on generation confrontation network |
| CN112328784A (en) * | 2019-08-05 | 2021-02-05 | 上海智臻智能网络科技股份有限公司 | Data information classification method and device |
| CN112328784B (en) * | 2019-08-05 | 2023-04-18 | 上海智臻智能网络科技股份有限公司 | Data information classification method and device |
| CN112347196A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Entity relation extraction method and device based on neural network |
| CN112347196B (en) * | 2019-08-06 | 2023-05-23 | 上海智臻智能网络科技股份有限公司 | Entity relation extraction method and device based on neural network |
| CN112347263B (en) * | 2019-08-06 | 2023-04-14 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN112347265B (en) * | 2019-08-06 | 2023-04-14 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN112347265A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN112347263A (en) * | 2019-08-06 | 2021-02-09 | 上海智臻智能网络科技股份有限公司 | Knowledge graph construction method |
| CN110580340A (en) * | 2019-08-29 | 2019-12-17 | 桂林电子科技大学 | neural network relation extraction method based on multi-attention machine system |
| CN111192692B (en) * | 2020-01-02 | 2023-12-08 | 上海联影智能医疗科技有限公司 | Entity relationship determination method and device, electronic equipment and storage medium |
| CN111192692A (en) * | 2020-01-02 | 2020-05-22 | 上海联影智能医疗科技有限公司 | Entity relationship determination method and device, electronic equipment and storage medium |
| CN111091007A (en) * | 2020-03-23 | 2020-05-01 | 杭州有数金融信息服务有限公司 | Method for identifying relationships among multiple enterprises based on public sentiment and enterprise portrait |
| CN111831783B (en) * | 2020-07-07 | 2023-12-08 | 北京北大软件工程股份有限公司 | Method for extracting chapter-level relation |
| CN111831783A (en) * | 2020-07-07 | 2020-10-27 | 北京北大软件工程股份有限公司 | Chapter-level relation extraction method |
| CN112463982A (en) * | 2020-11-27 | 2021-03-09 | 华东师范大学 | Relationship extraction method based on explicit and implicit entity constraint |
| CN113011161A (en) * | 2020-12-29 | 2021-06-22 | 中国航天科工集团第二研究院 | Method for extracting human and pattern association relation based on deep learning and pattern matching |
| CN114218956A (en) * | 2022-01-24 | 2022-03-22 | 平安科技（深圳）有限公司 | Relation extraction method and system based on neural network and remote supervision |
| CN114757179A (en) * | 2022-04-13 | 2022-07-15 | 成都信息工程大学 | Entity relationship joint extraction method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106354710A (en) | Neural network relation extracting method | |
| CN110083705B (en) | Multi-hop attention depth model, method, storage medium and terminal for target emotion classification | |
| CN106484664B (en) | Similarity calculating method between a kind of short text | |
| CN113239186B (en) | Graph convolution network relation extraction method based on multi-dependency relation representation mechanism | |
| CN111259987B (en) | Method for extracting event main body by multi-model fusion based on BERT | |
| CN106547735A (en) | The structure and using method of the dynamic word or word vector based on the context-aware of deep learning | |
| CN108664632A (en) | A kind of text emotion sorting algorithm based on convolutional neural networks and attention mechanism | |
| CN112507699B (en) | Remote supervision relation extraction method based on graph convolution network | |
| CN109543722A (en) | A kind of emotion trend forecasting method based on sentiment analysis model | |
| CN109558487A (en) | Document Classification Method based on the more attention networks of hierarchy | |
| CN110245229A (en) | A kind of deep learning theme sensibility classification method based on data enhancing | |
| CN106650725A (en) | Full convolutional neural network-based candidate text box generation and text detection method | |
| CN107133214A (en) | A kind of product demand preference profiles based on comment information are excavated and its method for evaluating quality | |
| CN105468713A (en) | Multi-model fused short text classification method | |
| CN107153642A (en) | A kind of analysis method based on neural network recognization text comments Sentiment orientation | |
| CN111079409B (en) | Emotion classification method utilizing context and aspect memory information | |
| CN108197294A (en) | A kind of text automatic generation method based on deep learning | |
| CN107943784A (en) | Relation extraction method based on generation confrontation network | |
| CN108536801A (en) | A kind of civil aviaton's microblogging security public sentiment sentiment analysis method based on deep learning | |
| CN113761893B (en) | Relation extraction method based on mode pre-training | |
| CN108549658A (en) | A kind of deep learning video answering method and system based on the upper attention mechanism of syntactic analysis tree | |
| CN110825850B (en) | Natural language theme classification method and device | |
| CN112559734A (en) | Presentation generation method and device, electronic equipment and computer readable storage medium | |
| CN108875034A (en) | A kind of Chinese Text Categorization based on stratification shot and long term memory network | |
| CN114462409A (en) | Audit field named entity recognition method based on countermeasure training |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170125 |
|
| WD01 | Invention patent application deemed withdrawn after publication |