CN109388805A - A kind of industrial and commercial analysis on altered project method extracted based on entity - Google Patents
A kind of industrial and commercial analysis on altered project method extracted based on entity Download PDFInfo
- Publication number
- CN109388805A CN109388805A CN201811239874.7A CN201811239874A CN109388805A CN 109388805 A CN109388805 A CN 109388805A CN 201811239874 A CN201811239874 A CN 201811239874A CN 109388805 A CN109388805 A CN 109388805A
- Authority
- CN
- China
- Prior art keywords
- entity
- industrial
- name
- commercial
- change
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 23
- 238000000034 method Methods 0.000 title claims abstract description 21
- 230000008859 change Effects 0.000 claims abstract description 32
- 238000000605 extraction Methods 0.000 claims abstract description 24
- 230000004075 alteration Effects 0.000 claims abstract description 11
- 230000007787 long-term memory Effects 0.000 claims abstract description 7
- 238000002360 preparation method Methods 0.000 claims abstract description 6
- 230000007246 mechanism Effects 0.000 claims description 26
- 239000011159 matrix material Substances 0.000 claims description 23
- 238000005457 optimization Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 5
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000002457 bidirectional effect Effects 0.000 description 3
- 230000008520 organization Effects 0.000 description 3
- 241000219780 Pueraria Species 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Include the following steps: the entity class and attribute structure that define training sample the invention discloses a kind of industrial and commercial analysis on altered project method extracted based on entity；The preparation and mark of training sample corpus；Using the combination of two-way shot and long term memory network and condition random field, training entity attribute extraction model；Extract target user before changing with the entity attribute after change；Horizontal analysis is carried out with the entity attribute after change before changing to the target user extracted, obtains the industrial and commercial alteration of the target user；The present invention uses the combination of two-way shot and long term memory network and condition random field, constructs entity attribute extraction model, extracts to Target Enterprise entity information, analyzes to realize Target Enterprise industry and commerce alteration；It avoids using traditional rule and probabilistic method and the shortcomings that rule coverage is not complete, prepares corpus heavy workload and can not analyze long text occurs.
Description
Technical field
The invention belongs to technical field of data processing, and in particular to a kind of industrial and commercial analysis on altered project side extracted based on entity
Method.
Background technique
It is provided according to Company Law of the People's Republic of China, there be can stepping on to company for information changing in enterprise during operation
Remember that organ applies for change of registration, it therefore, can be from this when we want to understand the real operation status of an enterprise or company
The industrial and commercial alteration of a enterprise or company is started with.For example, just having can when this enterprise or Top Management leave office one after another
It can illustrate that this enterprise or company just meet with change of personnel crisis, concern and early warning can be carried out to it.
The prior art mainly uses rule-based industrial and commercial analysis on altered project.It carries out industrial and commercial change using the method for pure rule
The extraction of information, but since the data source of industrial and commercial change at present is relatively more, data itself are more chaotic, none is unified
Specification, same change type may be there are many kinds of data format.This is just covered with very big requirement to rule, and I
Rule tend not to cover all samples, this allows for carrying out analysis using the method for pure rule and can generate much to ask
Topic, such as: the name or mechanism name mistake extracted, leakage isolate according to etc., very big shadow is had to last result in this way
It rings.Furthermore the complexity for being exactly this rule can be very high, because can be related to the identification of name, mechanism name, uses pure rule
Then carrying out analysis will lead to inefficiency.
Summary of the invention
In order to solve the above problems existing in the present technology, it is an object of that present invention to provide a kind of based on entity extraction
Industrial and commercial analysis on altered project method.
The technical scheme adopted by the invention is as follows:
It is a kind of based on entity extract industrial and commercial analysis on altered project method include the following steps:
Define the entity class and attribute structure of training sample；
The preparation and mark of training sample corpus；
Using the combination of two-way shot and long term memory network and condition random field, training entity attribute extraction model；
Target user before changing and in the industrial and commercial text data input entity attribute extraction model after change, is extracted
Target user before changing with the entity attribute after change；
Horizontal analysis is carried out with the entity attribute after change before changing to the target user extracted, obtains target use
The industrial and commercial alteration at family.
Further, it includes mechanism that the entity class for defining training sample and attribute structure, which include: definition entity class,
Name and name；Defined attribute field is type field, starting bit field, cut-off one of bit field and body field or more
Kind.
Further, the preparation and mark of the training sample corpus include mark mechanism name start bit label, mechanism name
Intermediate label, mechanism name stop bits label, name start bit label, name intermediate label, name stop bits label, other texts
Word label.
Further, the trained entity attribute extraction model includes the following steps:
1) training sample corpus is labeled by word, carries out one-hot coding as input text, obtains one-hot
Input text matrix [N*max_seq] after coding；
2) the input text matrix [N*max_seq] after encoding one-hot is input in Embedding layers, obtains word
Vector three-dimensional matrice [N*max_seq*embedding_size]；
3) term vector three-dimensional matrice [N*max_seq*embedding_size] is input in BiLSTM network, is obtained
About the other probability distribution emission matrix [N*max_seq*num_tag] of tag class；
4) condition random will be input to about the other probability distribution emission matrix [N*max_seq*num_tag] of tag class
In, state-transition matrix [num_tag*num_tag] is trained.
Further, the entity attribute includes name, mechanism name and job information.
Further, the industrial and commercial alteration of the target user includes:
1) if someone or mechanism exist before changing, but no longer exist after change, then defines the individual or mechanism is moved back
Chu Liao the said firm.
If 2) someone or mechanism are being not present before changing, but exist after change, then defining the individual can mechanism addition
The said firm.
3) if someone exists with after change before changing, but its job information is changed, then defines the individual
Belong to information change.
It further, further include being extracted to the entity attribute trained in the trained entity attribute extraction model step
Model carries out the step of model score and model optimization.
The invention has the benefit that
The present invention uses two-way shot and long term memory network (Bidirectional LSTM, BiLSTM) and condition random field
The combination of (conditional random fields), construct entity attribute extraction model, to Target Enterprise entity information into
Row extracts, and analyzes to realize Target Enterprise industry and commerce alteration；BiLSTM can letter between oneself learning text
Breath, it is no longer necessary to complicated Feature Engineering, and have good support to long text, it avoids using traditional rule and probability
Statistical method and there is the shortcomings that rule coverage is not complete, prepares corpus heavy workload and can not analyze long text；And add
Entering condition random field then more can be using this mutual information of text, and the result for generating it is more reliable.
Detailed description of the invention
Fig. 1 is flow chart of the present invention.
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention is further elaborated.
A kind of industrial and commercial analysis on altered project method extracted based on entity, is included the following steps:
S101, the entity class and attribute structure for defining training sample.
Entity class can be mechanism name (ORG) and name (PER).
For every a kind of entity, its standardized attribute structure is defined.In one exemplary embodiment, name/machine is defined
The attribute structure of structure name are as follows:
The preparation and mark of S102, training sample corpus.
In one exemplary embodiment, word Marking Guidelines and meaning are as follows:
B-ORG representative organization name start bit label
I-ORG representative organization name intermediate label
E-ORG representative organization name stop bits label
B-PER represents name start bit label
I-PER represents name intermediate label
E-PER represents name stop bits label
B-POS represents position start bit label
I-POS represents position intermediate label
E-POS represents position stop bits label
O represents other texts
By the above specification, the mark of each word of training sample is completed.After the completion of corpus mark, down-stream is understood that
The meaning of entity, facilitates machine to handle text in text.
S103, training entity attribute extraction model.
Using two-way shot and long term memory network (Bidirectional LSTM, BiLSTM) and condition random field
The combination of (conditional random fields) constructs entity attribute extraction model.
Two-way shot and long term memory network (Bidirectional LSTM, BiLSTM) includes preceding to LSTM and backward LSTM
Two groups of modules can obtain the associated dependence of the long range of context long-time, capture context substance feature, obtain more
Temporal correlation between multiple entity, and can from both direction shadow of the noises such as exclusive PCR entity to neural network model
It rings, excavation of the very big power-assisted to long-term dependence is extracted to the vital height such as information extraction and entity-relationship recognition
Layer semantic feature.The advantage of opposite Bayesian network, LSTM and its mutation is the long sequence relation between capable of capturing entity, but
Its inferential capability and interpretation are poor.
Condition random field (conditional random fields) is a kind of discriminate probabilistic model, is random field
One kind being usually used in mark or analytical sequence data, such as natural language text or biological sequence.Such as Markov random field, item
Part random field is that the vertex with undirected graph model, in figure represents stochastic variable, and the line between vertex represents between stochastic variable
Dependence relation, in condition random field, stochastic variable Y's is distributed as conditional probability, and given observed value is then stochastic variable
X.In principle, the graph model layout of condition random field can be any given, and general common layout is the frame of chain eliminant
Structure, no matter chain eliminant framework is all deposited in training (training), inference (inference) or decoding (decoding)
In the higher algorithm of efficiency for calculation.
The advantage of BiLSTM is can to remember contextual information, excavation of the very big power-assisted to long-term dependence, to semanteme
Understanding is very helpful, but if being directly labeled task with it, with regard to having a problem, BiLSTM belongs to timing
Model, so its output belongs to locally optimal solution just for current character.And condition random field then to the requirement of template very
Height covers the information that comprehensive template can allow model to acquire many contexts, but often has template and cover infull feelings
Condition occurs.The information of the available context of BiLSTM, but it is desirable that a solution model, and condition random field can be with
Generate globally optimal solution, but it needs the information of context, therefore, present invention combination BiLSTM and condition random field the two
Model, to construct the complete model of a mutual supplement with each other's advantages.
Training entity attribute extraction model includes the following steps:
1) training sample corpus is labeled by word, carries out one-hot coding as input text, obtains one-hot
Input text matrix [N*max_seq] after coding.[N*max_seq] matrix is used to train term vector, wherein N is represented
Batch_size i.e. batch size, max_seq represent in entire batch maximum sentence length, be used to by entire batch into
Row alignment operation.
2) the input text matrix [N*max_seq] after encoding one-hot is input in Embedding layers, is obtained
Term vector three-dimensional matrice [N*max_seq*embedding_size].[N*max_seq* embedding_size] is represented will
The input text of one-hot form is indicated in a manner of term vector, can indicate the similarity degree between word and word.Its
In, embedding_size represents the size of word vector, it represents the dimension of entire term vector, can often influence model
Overall performance.
3) term vector three-dimensional matrice [N*max_seq*embedding_size] is input in BiLSTM network, is obtained
About the other probability distribution emission matrix [N*max_seq*num_tag] of tag class.[N* max_seq*num_tag] is one
About the other probability distribution of tag class, what is respectively indicated is the probability that each word of input text is each label, wherein num_
Tag is the total number of label.
4) condition random will be input to about the other probability distribution emission matrix [N*max_seq*num_tag] of tag class
It in, trains state-transition matrix [num_tag*num_tag], is solved after convenient.State-transition matrix [num_tag*
Num_tag] represent the probability that some label is transferred to other labels.
S104, target user entity attribute extraction
Target user before changing and in the industrial and commercial text data input entity attribute extraction model after change, is extracted
Target user before changing with the entity attribute after change.Entity attribute includes name, mechanism name and job information.
Specifically, target text is inputted in entity attribute extraction model, the state-transition matrix and hair of the text are obtained
Matrix is penetrated, is solved using viterbi algorithm, final sequence is obtained.Viterbi algorithm is a kind of algorithm of Dynamic Programming, is used for
Find the most possible-Viterbi path-hidden state sequence for generating observed events sequence.
Viterbi algorithm method for solving is as follows:
It suppose there is state space S, share k state, the probability of original state i is πi, from state x to the transfer of state k
Probability is ax,k.Enabling the output observed is y1,...,yT.Generate the most possible status switch x of observation result1,...,xT
It is provided by recurrence relation:
V1,k=P (y1|k)·πk
Vt,k=maxx∈S(P(yt|k)·ax,k·Vt-1,x),
Wherein P (yt| it is k) emission matrix of the output of BiLSTM, ax,kFor the transfer matrix that condition random field trains,
V1,kRepresent the probability under k-state, Vt,kThe probability that k-state is under t moment is then represented, we are stateful to t moment institute
Probability be maximized, an optimal paths can be found, eventually find most suitable sequence label.
S105, the analysis of target user's industry and commerce alteration.
Horizontal analysis is carried out with the entity attribute after change before changing to the target user extracted, obtains target use
The industrial and commercial alteration at family.The present invention is defined as follows the industrial and commercial alteration of target user:
1) if someone or mechanism exist before changing, but no longer exist after change, then defines the individual or mechanism is moved back
Chu Liao the said firm.
If 2) someone or mechanism are being not present before changing, but exist after change, then defining the individual can mechanism addition
The said firm.
3) if someone exists with after change before changing, but its job information is changed, then defines the individual
Belong to information change.
S103 training entity attribute extraction model step in, further include to the entity attribute extraction model trained into
The step of row model score and model optimization, to guarantee that the entity attribute extraction model of training can accurately extract target
The entity attribute of text.
Model score:
The output matrix of Bi-LSTM is P, whereinRepresent word ωiIt is mapped toNon-normalized probability.For CRF
For, it is assumed that there are a shift-matrix As, thenIt representsIt is transferred toTransition probability.
Output tag sequences y corresponding for list entries X defines the score s (X, y) of each output tag sequences y
Are as follows:
Utilize Softmax function, YXFor entire status switch, we define one for each correct tag sequences y
Probability value, i.e. likelihood probability p (y | X):
Thus in training, we only need to maximize likelihood probability p (y | X), are estimated using log-likelihood:
So loss function is defined as-log (p (y | X) by us), so that it may using gradient descent method come Optimized model.
In one exemplary embodiment, the primary change of certain company is as follows:
Before changing: looking into certain (director)；Pueraria lobota (director)；(other are non-natural by Gwill Telecomm Unication.Inc
People investor)
After change: Guo (director)；Pueraria lobota (executive director)
Industrial and commercial analysis on altered project is carried out according to the method for the present invention, can obtain following result:
Using the method for the present invention, the change of position is not only shown, identifies the change shape of party yet
State.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are each under the inspiration of the present invention
The product of kind form, however, make any variation in its shape or structure, it is all to fall into the claims in the present invention confining spectrum
Interior technical solution, is within the scope of the present invention.
Claims (7)
1. a kind of industrial and commercial analysis on altered project method extracted based on entity, which comprises the steps of:
Define the entity class and attribute structure of training sample；
The preparation and mark of training sample corpus；
Using the combination of two-way shot and long term memory network and condition random field, training entity attribute extraction model；
Target user before changing and in the industrial and commercial text data input entity attribute extraction model after change, is extracted into target use
Family before changing with the entity attribute after change；
Horizontal analysis is carried out with the entity attribute after change before changing to the target user extracted, obtains the work of the target user
Quotient's alteration.
2. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the definition instruction
Practice sample entity class and attribute structure include:
Defining entity class includes mechanism name and name；
Defined attribute field is type field, starting bit field, cut-off one of bit field and body field or a variety of.
3. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the trained sample
The preparation and mark of this corpus include mark mechanism name start bit label, mechanism name intermediate label, mechanism name stop bits label, people
Name start bit label, name intermediate label, name stop bits label, other word tags.
4. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the training is real
Body attribute extraction model includes the following steps:
1) training sample corpus is labeled by word, carries out one-hot coding as input text, obtains one-hot coding
Input text matrix [N*max_seq] afterwards；
2) the input text matrix [N*max_seq] after encoding one-hot is input in Embedding layers, obtains term vector
Three-dimensional matrice [N*max_seq*embedding_size]；
3) term vector three-dimensional matrice [N*max_seq*embedding_size] is input in BiLSTM network, is obtained about mark
Sign the probability distribution emission matrix [N*max_seq*num_tag] of classification；
4) it will be input in condition random field about the other probability distribution emission matrix [N*max_seq*num_tag] of tag class,
Train state-transition matrix [num_tag*num_tag].
5. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the entity category
Property includes name, mechanism name and job information.
6. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the target is used
The industrial and commercial alteration at family includes:
1) if someone or mechanism exist before changing, but no longer exist after change, then defines the individual or mechanism exits
The said firm；
If 2) someone or mechanism are being not present before changing, but exist after change, then define the individual can mechanism joined this
Company；
3) if someone exists with after change before changing, but its job information is changed, then defines the individual and belong to
Information change.
7. the industrial and commercial analysis on altered project method according to claim 1 extracted based on entity, which is characterized in that the training is real
It further include that model score and model optimization are carried out to the entity attribute extraction model trained in body attribute extraction model step
Step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811239874.7A CN109388805A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial and commercial analysis on altered project method extracted based on entity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811239874.7A CN109388805A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial and commercial analysis on altered project method extracted based on entity |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109388805A true CN109388805A (en) | 2019-02-26 |
Family
ID=65427756
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811239874.7A Pending CN109388805A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial and commercial analysis on altered project method extracted based on entity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109388805A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191130A (en) * | 2019-12-30 | 2020-05-22 | 泰康保险集团股份有限公司 | Information extraction method, device, equipment and computer readable storage medium |
| CN113297238A (en) * | 2021-04-07 | 2021-08-24 | 北京金堤征信服务有限公司 | Method and device for information mining based on historical change records |
| CN113627139A (en) * | 2021-08-11 | 2021-11-09 | 平安国际智慧城市科技股份有限公司 | Enterprise reporting form generation method, device, equipment and storage medium |
| WO2021232595A1 (en) * | 2020-05-22 | 2021-11-25 | 平安国际智慧城市科技股份有限公司 | Enterprise state supervision method, apparatus, and device, and computer readable storage medium |
| CN113901834A (en) * | 2021-10-14 | 2022-01-07 | 盐城金堤科技有限公司 | Text display method and device, computer storage medium and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207936A (en) * | 2010-03-30 | 2011-10-05 | 国际商业机器公司 | Method and system for indicating content change of electronic document |
| CN105740446A (en) * | 2016-02-02 | 2016-07-06 | 河南九博科技股份有限公司 | Enterprise information integration method and apparatus used for honesty and credit evaluation in recruitment website |
| CN106469200A (en) * | 2016-08-31 | 2017-03-01 | 国信优易数据有限公司 | There are the address location change method and system that but industry and commerce is not put on record in time in a kind of prediction enterprise |
| CN107506343A (en) * | 2017-07-27 | 2017-12-22 | 北京金堤科技有限公司 | The processing method and platform of a kind of information editing |
| CN108399240A (en) * | 2018-02-28 | 2018-08-14 | 北京金堤科技有限公司 | Enterprise's modification information data digging method and system |
-
2018
- 2018-10-23 CN CN201811239874.7A patent/CN109388805A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207936A (en) * | 2010-03-30 | 2011-10-05 | 国际商业机器公司 | Method and system for indicating content change of electronic document |
| CN105740446A (en) * | 2016-02-02 | 2016-07-06 | 河南九博科技股份有限公司 | Enterprise information integration method and apparatus used for honesty and credit evaluation in recruitment website |
| CN106469200A (en) * | 2016-08-31 | 2017-03-01 | 国信优易数据有限公司 | There are the address location change method and system that but industry and commerce is not put on record in time in a kind of prediction enterprise |
| CN107506343A (en) * | 2017-07-27 | 2017-12-22 | 北京金堤科技有限公司 | The processing method and platform of a kind of information editing |
| CN108399240A (en) * | 2018-02-28 | 2018-08-14 | 北京金堤科技有限公司 | Enterprise's modification information data digging method and system |
Non-Patent Citations (1)
| Title |
|---|
| GUILLAUME LAMPLE 等: "Neural Architectures for Named Entity Recognition", 《网页在线公开：HTTPS://ARXIV.ORG/ABS/1603.01360V3》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191130A (en) * | 2019-12-30 | 2020-05-22 | 泰康保险集团股份有限公司 | Information extraction method, device, equipment and computer readable storage medium |
| WO2021232595A1 (en) * | 2020-05-22 | 2021-11-25 | 平安国际智慧城市科技股份有限公司 | Enterprise state supervision method, apparatus, and device, and computer readable storage medium |
| CN113297238A (en) * | 2021-04-07 | 2021-08-24 | 北京金堤征信服务有限公司 | Method and device for information mining based on historical change records |
| CN113297238B (en) * | 2021-04-07 | 2023-10-20 | 北京金堤征信服务有限公司 | Method and device for mining information based on history change record |
| CN113627139A (en) * | 2021-08-11 | 2021-11-09 | 平安国际智慧城市科技股份有限公司 | Enterprise reporting form generation method, device, equipment and storage medium |
| CN113901834A (en) * | 2021-10-14 | 2022-01-07 | 盐城金堤科技有限公司 | Text display method and device, computer storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109388805A (en) | A kind of industrial and commercial analysis on altered project method extracted based on entity | |
| Han et al. | Neural knowledge acquisition via mutual attention between knowledge graph and text | |
| CN110598203B (en) | Method and device for extracting entity information of military design document combined with dictionary | |
| CN109857990B (en) | Financial bulletin information extraction method based on document structure and deep learning | |
| CN107729312B (en) | Multi-granularity word segmentation method and system based on sequence labeling modeling | |
| CN107766483A (en) | The interactive answering method and system of a kind of knowledge based collection of illustrative plates | |
| CN111767732B (en) | Document content understanding method and system based on graph attention model | |
| WO2021208696A1 (en) | User intention analysis method, apparatus, electronic device, and computer storage medium | |
| CN106250915A (en) | A kind of automatic image marking method merging depth characteristic and semantic neighborhood | |
| CN110196980A (en) | A kind of field migration based on convolutional network in Chinese word segmentation task | |
| CN109446523A (en) | Entity attribute extraction model based on BiLSTM and condition random field | |
| Dhingra et al. | Linguistic knowledge as memory for recurrent neural networks | |
| CN104809105B (en) | Recognition methods and the system of event argument and argument roles based on maximum entropy | |
| CN113672718B (en) | Dialogue intention recognition method and system based on feature matching and field self-adaption | |
| CN108829823A (en) | A kind of file classification method | |
| CN113051914A (en) | Enterprise hidden label extraction method and device based on multi-feature dynamic portrait | |
| CN113434688B (en) | Data processing method and device for public opinion classification model training | |
| CN114580424B (en) | Labeling method and device for named entity identification of legal document | |
| CN109919175A (en) | A kind of more classification methods of entity of combination attribute information | |
| CN113468887A (en) | Student information relation extraction method and system based on boundary and segment classification | |
| CN113962224A (en) | Named entity recognition method and device, equipment, medium and product thereof | |
| Paul et al. | A modern approach for sign language interpretation using convolutional neural network | |
| CN111209362A (en) | Address data analysis method based on deep learning | |
| JP2022151838A (en) | Extraction of open information from low resource language | |
| CN112989811B (en) | History book reading auxiliary system based on BiLSTM-CRF and control method thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190226 |
|
| RJ01 | Rejection of invention patent application after publication |