CN107341146A - The semantic resolution system of transportable spoken language and its implementation based on semantic groove internal structure - Google Patents
The semantic resolution system of transportable spoken language and its implementation based on semantic groove internal structure Download PDFInfo
- Publication number
- CN107341146A CN107341146A CN201710483733.9A CN201710483733A CN107341146A CN 107341146 A CN107341146 A CN 107341146A CN 201710483733 A CN201710483733 A CN 201710483733A CN 107341146 A CN107341146 A CN 107341146A
- Authority
- CN
- China
- Prior art keywords
- semantic
- atomic concepts
- model
- sequence
- atomic
- 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
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000013526 transfer learning Methods 0.000 claims abstract description 22
- 238000013528 artificial neural network Methods 0.000 claims abstract description 11
- 230000000306 recurrent effect Effects 0.000 claims abstract description 11
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 17
- 238000000034 method Methods 0.000 claims description 11
- 230000001419 dependent effect Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 6
- 238000013508 migration Methods 0.000 claims description 5
- 230000005012 migration Effects 0.000 claims description 5
- 230000002457 bidirectional effect Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 230000007935 neutral effect Effects 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 230000003134 recirculating effect Effects 0.000 claims description 2
- 238000012217 deletion Methods 0.000 claims 1
- 230000037430 deletion Effects 0.000 claims 1
- 210000003739 neck Anatomy 0.000 claims 1
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising 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
- G06F40/30—Semantic analysis
-
- 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
A kind of semantic resolution system of transportable spoken language and its implementation based on semantic groove internal structure, including：Source domain model training module containing Recognition with Recurrent Neural Network, target domain transfer learning module and parsing module containing the model based on atomic concepts sequence, source domain model training module gathers source domain data and is exported the semantic understanding model for training obtained source domain to target domain transfer learning module according to the atomic concepts sequence definition of source domain, target domain transfer learning module is carried out second training according to the sample data and atomic concepts sequence definition of target domain and is carried out the optimization of transfer learning to the spoken semantic understanding model of resulting target domain using single field or multi-field mode；Parsing module is according to the input by sentence of the spoken semantic understanding model analyzing user after optimization and obtains semantic results.The present invention can support the spoken semantic understanding that field migrates.
Description
Technical field
The present invention relates to a kind of technology in phonetic entry field, it is specifically a kind of based on semantic groove internal structure can
The spoken semantic resolution system of migration and its implementation.
Background technology
The huge advance of development and speech recognition technology recently as development of Mobile Internet technology, such as Microsoft Research
The speech recognition accuracy to be matched in excellence or beauty with professional person is achieved in specific set of data, the demand of man-machine spoken interaction becomes more next
It is bigger.Spoken semantic understanding serves the key for helping machine to understand user view as the ring after immediately speech recognition
Property effect.
The content of the invention
The present invention is directed to the rule and labeled data that prior art depends on a large amount of manual compilings, easily produces conflict and hardly possible
Data between maintenance, different dialogue field are often moved because the definition of semantic groove is inconsistent without reusable, for field
The defects of analytic ability of shifting is not strong, propose a kind of semantic resolution system of transportable spoken language based on semantic groove internal structure and its
Implementation method, by reasonably representing the semantic slot structure of the relation between semantic groove, with reference to bidirectional circulating neural network model,
Semantic slot structure is modeled, it would be preferable to support the spoken semantic understanding of field migration.
The present invention is achieved by the following technical solutions：
The present invention relates to a kind of transportable semantic resolution system of spoken language based on semantic groove internal structure, including：Containing following
The source domain model training module of ring neutral net, the target domain transfer learning mould containing the model based on atomic concepts sequence
Block and parsing module, wherein：Source domain model training module gathers source domain data and according to the atomic concepts sequence of source domain
Row definition exports the semantic understanding model for training obtained source domain to target domain transfer learning module, target domain migration
Study module carries out second training according to the sample data and atomic concepts sequence definition of target domain and uses single field or more
Field mode carries out the optimization of transfer learning to the spoken semantic understanding model of resulting target domain；Parsing module is according to optimization
The input by sentence of spoken semantic understanding model analyzing user afterwards simultaneously obtains semantic results.
The present invention relates to the implementation method of said system, including：
Step 1) carries out sequence labelling using Recognition with Recurrent Neural Network to input sentence；
Semantic groove is expressed as atomic concepts sequence by step 2)；
Step 3) is modeled by the expression of semantic groove based on atomic concepts sequence to spoken semantic understanding, obtain towards
The spoken semantic understanding model of the target domain of fine granularity semantic meaning representation, and spoken semanteme is parsed using the model；
The described model based on atomic concepts sequence includes：Atomic concepts stand alone type and dependent formula, wherein：Atom is general
It is separate between level in sequence to read in free standing model, layer adjacent in sequence in atomic concepts dependent formula model
It is order dependent (i.e. prediction result of the atomic concepts of last layer level dependent on the atomic concepts of next level) between level.
Step 4) carries out transfer learning using single field or multi-field mode, i.e., on the basis of the modeling of step 3, passes through group
Knit enough source domain data and a small amount of target domain data carry out the parameter learning of model so that target domain is in only a small amount of number
Good semantic understanding performance can be also obtained in the case of.
Technique effect
Compared with prior art, the present invention is modeled when in face of not marking field using atomic concepts, can allow different languages
The shared atomic concepts for having common factor of adopted groove, and the present invention can utilize existing field (source using the learning strategy of field migration
Field) data auxiliary mark field model training, so as to significantly reduce the amount of labour of manual compiling.
Brief description of the drawings
Fig. 1 a are traditional modeling pattern, i.e., semantic groove as a single classification and Fig. 1 b for based on atomic concepts
Modeling pattern, i.e. Independent modeling schematic diagram；
Fig. 2 is the modeling pattern schematic diagram based on atomic concepts；
Fig. 3 is system module figure.
Embodiment
As shown in figure 3, the transportable semantic resolution system of spoken language based on semantic groove internal structure that the present embodiment is related to, bag
Include：Source domain model training module containing Recognition with Recurrent Neural Network, the target domain containing the model based on atomic concepts sequence
Transfer learning module and parsing module, wherein：Source domain model training module gathers source domain data and according to source domain
Atomic concepts sequence definition exports the semantic understanding model for training obtained source domain to target domain transfer learning module, mesh
Mark field transfer learning module carries out second training according to a small amount of sample data and atomic concepts sequence definition of target domain
And the optimization of transfer learning is carried out to the spoken semantic understanding model of resulting target domain using single field or multi-field mode；
Parsing module is according to the input by sentence of the spoken semantic understanding model analyzing user after optimization and obtains semantic results.
Described Recognition with Recurrent Neural Network includes：
1. recirculating network (RNN)：ht=f (Uxt+Wht-1), wherein：F is nonlinear activation function, such as sigmoid,
Tanh or ReLU etc.；xtIt is the input (word) of t, U is the weight matrix of input layer, ht-1And htIt is t-1 moment and t respectively
The hidden layer vector at moment, W are circulation layer matrixs.
2. long mnemon (LSTM) in short-term：it=σ (Wi·[ht-1,xt]+bi),ft=σ (Wf·[ht-1,xt]+bf),ot
=σ (Wo·[ht-1,xt]+bo),Ct=ft*Ct-1+it*tanh(WC·[ht-1,xt]+bC),ht=ot*tanh(Ct), wherein：σ is
Sigmoid functions, Wi、Wf、Wo、WCIt is input layer matrix, itIt is input gate steering amount, ftIt is to forget gate vector, otIt is defeated
Steering of going out amount, CtIt is that mnemon is vectorial, bi、bf、bo、bCIt is shift term.
3. door control unit (GRU)：rt=σ (Wr·[ht-1,xt]), zt=σ (Wz·[ht-1,xt]), Wherein：σ is sigmoid functions, and t represents t, xt、ht-1And
htIt is defined as above, Wr、Wz、WhIt is input layer matrix, rtIt is Memory-Gate steering amount, ztIt is to forget gate vector,It is implicit shape
State vector.
Said system realizes transportable spoken semantic parsing especially by the following manner：
Step 1) carries out sequence labelling using Recognition with Recurrent Neural Network.
Semantic groove is expressed as atomic concepts sequence by step 2), is comprised the following steps that：
2.1) atomic concepts set C={ c are built1,c2,…,cV, wherein：V is the quantity of atomic concepts.
Described atomic concepts is not subdivisible minimum semantic primitive, common semantic primitive have " city ", " date ",
" time ", " name ", " song title " etc..
2.2) for some dialogue field, its semantic groove is represented using atomic concepts sequence, is specially：
I) to semantic groove siAtomic concepts is divided into, in the event of the new atomic concepts not having in C, is then added
Into C.
Ii) by semantic groove siIt is newly defined as atomic concepts sequence [ci,1,ci,2,…,ci,M], wherein：M is the sequence
Length, ci,mRepresent semantic groove siM-th of atomic concepts, 1≤m≤M.
Iii) as semantic groove siSubstantive atomic concepts number less than M, the atomic concepts null of sky will be filled.
Described atomic concepts sequence meets：
A) atomic concepts of different dimensions does not have common factor in sequence；
B) ranking criteria of atomic concepts sequence be more independent of context concept more come before.
Such as：For the atomic concepts of semantic groove " date of birth ", " date " should be come before " birth ", because " going out
It is raw " dependent on the verb beyond the date (such as " being born in ", " birth ").
Step 3) is modeled by the expression of semantic groove based on atomic concepts sequence to spoken semantic understanding, obtain towards
The spoken semantic understanding model of fine granularity semantic meaning representation, and spoken semanteme is parsed using the model, specifically include：
3.1) when different dimensions are separate in atomic concepts sequence：The atomic concepts of each dimension in sequence is pre-
The prediction task independent as one is surveyed, sequence length M, just there is M task；Bidirectional circulating nerve is shared between different task
The input layer of network and implicit layer parameter, and have the exclusive input layer of a task respectively.
3.2) when different dimensions are mutually not only immediately in atomic concepts sequence：By the atomic concepts of each dimension in sequence
An atomic concepts to preceding dimension is regarded in prediction, which as, the prediction task of dependence.
To M bidirectional circulating neural network model of M task design in the present embodiment, the model in m dimensions it is defeated
Enter the prediction result (m except inputting sentence, also the atomic concepts of m-1 dimensions including user>1).
Step 4) carries out transfer learning using single field or multi-field mode, wherein：
Described single field transfer learning refers to：
4.1) according to the semantic slot definition of source domain, the semantic analytic modell analytical model Model_org based on atomic concepts is built；
4.2) parameter of the Model_org models is updated using the data of source domain, such as：The term vector square of mode input
The weight matrix of battle array, Recognition with Recurrent Neural Network weight matrix and output layer.
4.3) according to the semantic slot definition of target domain, the atomic concepts on Model_org is increased and deletes behaviour
Make and (specially increase and delete input, output layer weight vector corresponding to corresponding atomic concepts), obtain target domain model
Structure and initial parameter, such as：Input word vector matrix and output layer weight matrix after adaptive adjustment.
4.4) parameter of the model after being adjusted in step 4.3) is updated using a small amount of sample data of target domain, such as：
The weight matrix of the term vector matrix of mode input, Recognition with Recurrent Neural Network weight matrix and output layer.
Described multi-field transfer learning refers to：
4.a) according to source domain and the semantic slot definition of target domain, the semantic analytic modell analytical model based on atomic concepts is built
Model_comb；
4.b) upset at random using the data of source domain and the sample data of target domain, obtain combined training data；
Model M odel_comb parameter 4.c) is updated using combined training data, such as：The term vector matrix of mode input,
The weight matrix of Recognition with Recurrent Neural Network weight matrix and output layer.
In actual production, if multiple target domains, alternative needs to carry out source domain data using multipass
Training, it is very time-consuming, and preferred option only needs to use a source domain data, can save the transfer learning of target domain
Time.
Test initial data：DSTC2&3(http://camdial.org/~mh521/dstc/).DSTC2 is source domain
(data of 2118 dialogues), DSTC3 are target domain (seed data of 11 dialogues, the test data of 1688 words).
The parameter setting of Recognition with Recurrent Neural Network：Input word vector dimension 100 is tieed up, and hidden layer vector dimension 100 is tieed up, parameter renewal
Mode is stochastic gradient descent algorithm, and greatest iteration round is 100 wheels.
Performance indications：The harmonic-mean (F-score) of semantic groove prediction,Wherein：P is semantic groove prediction
Accuracy rate, R are the recall rates of semantic groove prediction, and F values are the higher the better.
Experimental result data：
From experimental result data as can be seen that traditional semantic understanding model performance based on the modeling of semantic groove can not show a candle to base
In the semantic understanding model of atomic concepts established model.Either simply with the situation of a small amount of target domain seed data or same
When used a large amount of source domain data in the case of, atomic concepts sequence proposed by the present invention and its model migrate task in field
On all achieve more preferable performance.Meanwhile the dependent formula atomic concepts Series Modeling inside the present invention is better than freestanding modeling,
The level dependence for embodying atomic concepts is physical presence, and the rising space for also having reacted the present invention is also very big.
Above-mentioned specific implementation can by those skilled in the art on the premise of without departing substantially from the principle of the invention and objective with difference
Mode local directed complete set is carried out to it, protection scope of the present invention is defined by claims and not by above-mentioned specific implementation institute
Limit, each implementation in the range of it is by the constraint of the present invention.
Claims (9)
- A kind of 1. transportable semantic resolution system of spoken language based on semantic groove internal structure, it is characterised in that including：Contain circulation The source domain model training module of neutral net, the target domain transfer learning module containing the model based on atomic concepts sequence And parsing module, wherein：Source domain model training module gathers source domain data and according to the atomic concepts sequence of source domain The semantic understanding model for training obtained source domain is exported to target domain transfer learning module, target domain migration and learned by definition Practise module and second training is carried out and using single field or more necks according to the sample data and atomic concepts sequence definition of target domain Domain mode carries out the optimization of transfer learning to the spoken semantic understanding model of resulting target domain；After parsing module is according to optimization Spoken semantic understanding model analyzing user input by sentence and obtain semantic results.
- 2. system according to claim 1, it is characterized in that, described Recognition with Recurrent Neural Network includes：Recirculating network：ht=f (Uxt+Wht-1), wherein：F is nonlinear activation function, such as sigmoid, tanh or ReLU Deng；xtFor the input (word) of t, U is the weight matrix of input layer, ht-1And htRespectively the hidden layer of t-1 moment and t to Amount, W are circulation layer matrix；Long mnemon in short-term：it=σ (Wi·[ht-1,xt]+bi),ft=σ (Wf·[ht-1,xt]+bf),ot=σ (Wo·[ht-1, xt]+bo),Ct=ft*Ct-1+it*tanh(WC·[ht-1,xt]+bC),ht=ot*tanh(Ct), wherein：σ is sigmoid functions, Wi、Wf、Wo、WCIt is input layer matrix, itFor input gate steering amount, ftTo forget gate vector, otFor out gate steering amount, Ct For mnemon vector, bi、bf、bo、bCIt is shift term；Door control unit： Wherein：σ is sigmoid functions, Wr、Wz、WhIt is input layer matrix, rtFor Memory-Gate Steering amount, ztTo forget gate vector,For hidden state vector.
- A kind of 3. transportable spoken semantic parsing implementation method of system according to claim 1 or claim 2, it is characterised in that bag Include：Step 1) carries out sequence labelling using Recognition with Recurrent Neural Network to input sentence；Semantic groove is expressed as atomic concepts sequence by step 2)；Step 3) is modeled by the semantic groove expression based on atomic concepts sequence to spoken semantic understanding, is obtained towards thin The spoken semantic understanding model of the target domain of granularity semantic meaning representation, and spoken semanteme is parsed using the model；Step 4) carries out transfer learning using single field or multi-field mode.
- 4. according to the method for claim 3, it is characterized in that, the model based on atomic concepts sequence includes：Atom Concept stand alone type and dependent formula, wherein：It is separate, atom between level in sequence in atomic concepts free standing model Prediction result of the atomic concepts of last layer level dependent on the atomic concepts of next level in sequence in concept dependent formula model.
- 5. the method according to claim 11, it is characterized in that, it is described that semantic groove is expressed as atomic concepts sequence, specifically Refer to：2.1) atomic concepts set C={ c are built1,c2,…,cV, wherein：V is the quantity of atomic concepts, and atomic concepts is can not The minimum semantic primitive divided again；2.2) for some dialogue field, its semantic groove is represented using atomic concepts sequence, is specially：I) to semantic groove siAtomic concepts is divided into, in the event of the new atomic concepts not having in C, is then added to C In；Ii) by semantic groove siIt is newly defined as atomic concepts sequence [ci,1,ci,2,…,ci,M], wherein：M is the length of the sequence, ci,mRepresent semantic groove siM-th of atomic concepts, 1≤m≤M；Iii) as semantic groove siSubstantive atomic concepts number less than M, the atomic concepts null of sky will be filled.
- 6. according to any described method in claim 3~5, it is characterized in that, described atomic concepts sequence meets：A) atomic concepts of different dimensions does not have common factor in sequence；B) ranking criteria of atomic concepts sequence be more independent of context concept more come before.
- 7. according to the method for claim 3, it is characterized in that, described step 3), specifically include：3.1) when different dimensions are separate in atomic concepts sequence：The atomic concepts of each dimension in sequence is predicted and made For an independent prediction task, sequence length M, just there is M task；Bidirectional circulating neutral net is shared between different task Input layer and implicit layer parameter, and have the exclusive input layer of a task respectively；3.2) when different dimensions are mutually not only immediately in atomic concepts sequence：The atomic concepts of each dimension in sequence is predicted Regarding an atomic concepts to preceding dimension as has the prediction task of dependence.
- 8. according to the method for claim 3, it is characterized in that, described single field transfer learning refers to：4.1) according to the semantic slot definition of source domain, the semantic analytic modell analytical model Model_org based on atomic concepts is built；4.2) parameter of the Model_org models is updated using the data of source domain；4.3) according to the semantic slot definition of target domain, the atomic concepts on Model_org is increased and deletion action, obtained To the structure and initial parameter of target domain model；4.4) using the parameter of the model after being adjusted in the sample data renewal step c of target domain.
- 9. according to the method for claim 3, it is characterized in that, described multi-field transfer learning refers to：4.a) according to source domain and the semantic slot definition of target domain, the semantic analytic modell analytical model Model_ based on atomic concepts is built comb；4.b) upset at random using the data of source domain and the sample data of target domain, obtain combined training data；4.c) using combined training data renewal model M odel_comb parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710483733.9A CN107341146B (en) | 2017-06-23 | 2017-06-23 | Migratable spoken language semantic analysis system based on semantic groove internal structure and implementation method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710483733.9A CN107341146B (en) | 2017-06-23 | 2017-06-23 | Migratable spoken language semantic analysis system based on semantic groove internal structure and implementation method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107341146A true CN107341146A (en) | 2017-11-10 |
| CN107341146B CN107341146B (en) | 2020-08-04 |
Family
ID=60221182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710483733.9A Active CN107341146B (en) | 2017-06-23 | 2017-06-23 | Migratable spoken language semantic analysis system based on semantic groove internal structure and implementation method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107341146B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090520A (en) * | 2018-01-08 | 2018-05-29 | 北京中关村科金技术有限公司 | Training method, system, device and the readable storage medium storing program for executing of intention assessment model |
| CN108491380A (en) * | 2018-03-12 | 2018-09-04 | 苏州思必驰信息科技有限公司 | Confrontation multitask training method for speech understanding |
| CN109189921A (en) * | 2018-08-07 | 2019-01-11 | 阿里巴巴集团控股有限公司 | Comment on the training method and device of assessment models |
| CN109508457A (en) * | 2018-10-31 | 2019-03-22 | 浙江大学 | A kind of transfer learning method reading series model based on machine |
| CN109597993A (en) * | 2018-11-30 | 2019-04-09 | 深圳前海微众银行股份有限公司 | Sentence analysis processing method, device, equipment and computer readable storage medium |
| CN110674648A (en) * | 2019-09-29 | 2020-01-10 | 厦门大学 | Neural network machine translation model based on iterative bidirectional migration |
| CN111488622A (en) * | 2019-01-25 | 2020-08-04 | 深信服科技股份有限公司 | Method and device for detecting webpage tampering behavior and related components |
| CN111600734A (en) * | 2019-02-21 | 2020-08-28 | 烽火通信科技股份有限公司 | Network fault processing model construction method, fault processing method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103778407A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Gesture recognition algorithm based on conditional random fields under transfer learning framework |
| US20160224892A1 (en) * | 2015-01-29 | 2016-08-04 | Panasonic Intellectual Property Management Co., Ltd. | Transfer learning apparatus, transfer learning system, transfer learning method, and recording medium |
-
2017
- 2017-06-23 CN CN201710483733.9A patent/CN107341146B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103778407A (en) * | 2012-10-23 | 2014-05-07 | 南开大学 | Gesture recognition algorithm based on conditional random fields under transfer learning framework |
| US20160224892A1 (en) * | 2015-01-29 | 2016-08-04 | Panasonic Intellectual Property Management Co., Ltd. | Transfer learning apparatus, transfer learning system, transfer learning method, and recording medium |
Non-Patent Citations (2)
| Title |
|---|
| SU ZHU ET AL.: "Concept Transfer Learning for Adaptive Language Understanding", 《ARXIV:1706.00927V1》 * |
| 张朝阳: "RNN和LSTM", 《RNN和LSTM - 张朝阳 - 博客园（HTTPS://WWW.CNBLOGS.COM/ZHANGCHAOYANG/ARTICLES/6684906.HTML）》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090520A (en) * | 2018-01-08 | 2018-05-29 | 北京中关村科金技术有限公司 | Training method, system, device and the readable storage medium storing program for executing of intention assessment model |
| CN108491380A (en) * | 2018-03-12 | 2018-09-04 | 苏州思必驰信息科技有限公司 | Confrontation multitask training method for speech understanding |
| CN108491380B (en) * | 2018-03-12 | 2021-11-23 | 思必驰科技股份有限公司 | Anti-multitask training method for spoken language understanding |
| CN109189921A (en) * | 2018-08-07 | 2019-01-11 | 阿里巴巴集团控股有限公司 | Comment on the training method and device of assessment models |
| CN109189921B (en) * | 2018-08-07 | 2021-09-07 | 创新先进技术有限公司 | Comment evaluation model training method and device |
| CN109508457A (en) * | 2018-10-31 | 2019-03-22 | 浙江大学 | A kind of transfer learning method reading series model based on machine |
| CN109597993A (en) * | 2018-11-30 | 2019-04-09 | 深圳前海微众银行股份有限公司 | Sentence analysis processing method, device, equipment and computer readable storage medium |
| WO2020107765A1 (en) * | 2018-11-30 | 2020-06-04 | 深圳前海微众银行股份有限公司 | Statement analysis processing method, apparatus and device, and computer-readable storage medium |
| CN111488622A (en) * | 2019-01-25 | 2020-08-04 | 深信服科技股份有限公司 | Method and device for detecting webpage tampering behavior and related components |
| CN111600734A (en) * | 2019-02-21 | 2020-08-28 | 烽火通信科技股份有限公司 | Network fault processing model construction method, fault processing method and system |
| CN110674648A (en) * | 2019-09-29 | 2020-01-10 | 厦门大学 | Neural network machine translation model based on iterative bidirectional migration |
| CN110674648B (en) * | 2019-09-29 | 2021-04-27 | 厦门大学 | Neural network machine translation model based on iterative bidirectional migration |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107341146B (en) | 2020-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107341146A (en) | The semantic resolution system of transportable spoken language and its implementation based on semantic groove internal structure | |
| CN111985245B (en) | Relationship extraction method and system based on attention cycle gating graph convolution network | |
| CN110209770B (en) | Named entity identification method based on strategy value network and tree search enhancement | |
| CN106650813B (en) | A kind of image understanding method based on depth residual error network and LSTM | |
| Xiao et al. | Research progress of RNN language model | |
| CN113487088A (en) | Traffic prediction method and device based on dynamic space-time diagram convolution attention model | |
| CN104598611B (en) | The method and system being ranked up to search entry | |
| CN106980683A (en) | Blog text snippet generation method based on deep learning | |
| CN111291556B (en) | Chinese entity relation extraction method based on character and word feature fusion of entity meaning item | |
| CN106910497A (en) | A kind of Chinese word pronunciation Forecasting Methodology and device | |
| CN110083700A (en) | A kind of enterprise's public sentiment sensibility classification method and system based on convolutional neural networks | |
| CN106156003A (en) | A kind of question sentence understanding method in question answering system | |
| CN110298043B (en) | Vehicle named entity identification method and system | |
| CN108197294A (en) | A kind of text automatic generation method based on deep learning | |
| CN110232122A (en) | A kind of Chinese Question Classification method based on text error correction and neural network | |
| CN110930008B (en) | Mine disaster event detection method based on convolutional neural network | |
| CN106652999A (en) | System and method for voice recognition | |
| CN116415654A (en) | Data processing method and related equipment | |
| CN109992773A (en) | Term vector training method, system, equipment and medium based on multi-task learning | |
| CN110162789A (en) | A kind of vocabulary sign method and device based on the Chinese phonetic alphabet | |
| CN110502640A (en) | A kind of extracting method of the concept meaning of a word development grain based on construction | |
| CN109597988A (en) | The former prediction technique of vocabulary justice, device and electronic equipment across language | |
| CN106844327A (en) | Text code method and system | |
| CN109918649A (en) | A kind of suicide Risk Identification Method based on microblogging text | |
| CN117094325B (en) | Named entity identification method in rice pest field |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200623 Address after: Room 223, old administration building, 800 Dongchuan Road, Minhang District, Shanghai, 200240 Applicant after: Shanghai Jiaotong University Intellectual Property Management Co.,Ltd. Applicant after: AI SPEECH Co.,Ltd. Address before: 200240 Dongchuan Road, Shanghai, No. 800, No. Applicant before: SHANGHAI JIAO TONG University Applicant before: AI SPEECH Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201020 Address after: 215123 14 Tengfei Innovation Park, 388 Xinping street, Suzhou Industrial Park, Suzhou, Jiangsu. Patentee after: AI SPEECH Co.,Ltd. Address before: Room 223, old administration building, 800 Dongchuan Road, Minhang District, Shanghai, 200240 Patentee before: Shanghai Jiaotong University Intellectual Property Management Co.,Ltd. Patentee before: AI SPEECH Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 215123 14 Tengfei Innovation Park, 388 Xinping street, Suzhou Industrial Park, Suzhou, Jiangsu. Patentee after: Sipic Technology Co.,Ltd. Address before: 215123 14 Tengfei Innovation Park, 388 Xinping street, Suzhou Industrial Park, Suzhou, Jiangsu. Patentee before: AI SPEECH Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A transferable oral semantic parsing system based on the internal structure of semantic slots and its implementation method Effective date of registration: 20230726 Granted publication date: 20200804 Pledgee: CITIC Bank Limited by Share Ltd. Suzhou branch Pledgor: Sipic Technology Co.,Ltd. Registration number: Y2023980049433 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |