CN109992671A - Intension recognizing method, device, equipment and storage medium - Google Patents
Intension recognizing method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109992671A CN109992671A CN201910285223.XA CN201910285223A CN109992671A CN 109992671 A CN109992671 A CN 109992671A CN 201910285223 A CN201910285223 A CN 201910285223A CN 109992671 A CN109992671 A CN 109992671A
- Authority
- CN
- China
- Prior art keywords
- list entries
- vector
- input
- indicated
- network layer
- 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
Abstract
Present disclose provides a kind of intension recognizing methods, comprising: by list entries Input knowledge map layer, and based on the feature in the knowledge mapping, the primary vector for generating the list entries is indicated;The primary vector is indicated to input pre-trained first nerves network layer, is indicated with generating the secondary vector of the list entries;The secondary vector is indicated that input nervus opticus network layer obtains the third vector expression of the list entries, and indicates the sequence labelling result of the generation list entries based on the third vector;Based on the sequence labelling as a result, obtaining one or more intentions relevant to the list entries.The disclosure additionally provides a kind of intention assessment device, electronic equipment and computer readable storage medium.
Description
Technical field
This disclosure relates to a kind of intension recognizing method, intension recognizing method device, electronic equipment and computer-readable storage
Medium.
Background technique
Existing most models are all to come parsing sentence dependence, the benefit of this method using syntactic analysis tree
It is that can quickly solve the parsing of common fixation user command, the syntax tree creation of incipient stage is easier.But this
The insufficient place of method is, this syntactic analysis tree constructs extremely complex, needs a large amount of manpower interventions, and with language
Method parsing tree becomes increasingly complex, and manually can not go to maintain, only just can solve, do not exist specified in syntax tree
It orders, can not parse, without good generalization ability defined in syntax tree.
Summary of the invention
At least one of in order to solve the above-mentioned technical problem, present disclose provides a kind of intension recognizing method, it is intended to know
Other device, electronic equipment and readable storage medium storing program for executing.
According to one aspect of the disclosure, a kind of intension recognizing method includes the following steps: list entries Input knowledge
Map layer, and based on the feature in the knowledge mapping, the primary vector for generating the list entries indicates;By described first to
Amount is indicated to input pre-trained first nerves network layer, be indicated with generating the secondary vector of the list entries;By described
Two vectors indicate input nervus opticus network layer obtain the list entries third vector indicate, and based on the third to
Amount indicates to generate the sequence labelling result of the list entries;And result based on the sequence labelling obtain with it is described defeated
Enter the relevant one or more intentions of sequence.
According at least one embodiment of the disclosure, the method also includes the secondary vector is indicated input second
After neural network obtains the third vector expression of the list entries, the third vector is indicated into input third nerve network
Layer, obtains the sequence labelling result of the list entries.
According at least one embodiment of the disclosure, the method also includes by the list entries Input knowledge figure
Before composing layer, list entries is encoded, and by the list entries Input knowledge map layer after coding.
According at least one embodiment of the disclosure, the first nerves network layer is carried out using bi-directional language model
Pre-training, and operated including the fine tuning for sequence labelling task.
According at least one embodiment of the disclosure, the method also includes obtaining the label including at least one label
Set, also, the sequence labelling result includes marking each word in the list entries in the upper tag set
One of label.
According at least one embodiment of the disclosure, the nervus opticus network layer is two-way shot and long term memory network,
The secondary vector indicates to include marking each word in the list entries to go up one of label in the list of labels
Probability value.
According at least one embodiment of the disclosure, the third nerve network layer is condition random field.
According to another aspect of the present disclosure, a kind of intention assessment device, comprising: first processing module, list entries is defeated
Enter knowledge mapping layer, and based on the feature in the knowledge mapping, the primary vector for generating the list entries is indicated;At second
Module is managed, the primary vector is indicated to input pre-trained first nerves network layer, to generate the of the list entries
Two vectors indicate;And third processing module, the secondary vector is indicated that input nervus opticus network layer obtains the input
The third vector of sequence indicates, and the sequence labelling result of the generation list entries is indicated based on the third vector;With
And fourth processing module, for obtaining one or more meanings relevant to the list entries based on the sequence labelling result
Figure.
According to the another further aspect of the disclosure, a kind of electronic equipment, comprising: memory, memory store executable instruction;With
And processor, processor executes the executable instruction of memory storage, so that processor executes above-mentioned method.
According to the another aspect of the disclosure, a kind of readable storage medium storing program for executing is stored with executable instruction in readable storage medium storing program for executing,
For realizing above-mentioned method when executable instruction is executed by processor.
Detailed description of the invention
Attached drawing shows the illustrative embodiments of the disclosure, and it is bright together for explaining the principles of this disclosure,
Which includes these attached drawings to provide further understanding of the disclosure, and attached drawing is included in the description and constitutes this
Part of specification.
Fig. 1 is the schematic flow according to the sequence labelling method for intention assessment of one embodiment of the disclosure
Figure.
Fig. 2 is the schematic diagram according to the intension recognizing method of one embodiment of the disclosure.
Fig. 3 is according to the schematic block diagram of the sequence labelling device of one embodiment of the disclosure.
Fig. 4 is the schematic block diagram according to the electronic equipment of one embodiment of the disclosure.
Specific embodiment
The disclosure is described in further detail with embodiment with reference to the accompanying drawing.It is understood that this place
The specific embodiment of description is only used for explaining related content, rather than the restriction to the disclosure.It also should be noted that being
Convenient for description, part relevant to the disclosure is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the disclosure can
To be combined with each other.The disclosure is described in detail below with reference to the accompanying drawings and in conjunction with embodiment.
So-called " intention assessment " is that sentence or inquiry are assigned to corresponding intention type by the method for classification.For example,
" song for playing XYZ ", its intention is to belong to " query music ", and " inquiry Pekinese's weather " is to belong to " inquiry day
Gas ".
So-called " sequence labelling ", that is for an one-dimensional linear list entries, such as linear list entries X=x1,
X2, x3 ..., xi ..., xn are some label in the tagged set of each element in linear order, such as tally set
Y=y1, y2, y3 are closed ..., yi ..., yn.Sequence labelling is most typical natural language processing NLP task, such as Chinese point
Word, part-of-speech tagging name Entity recognition, semantic character labeling, it is intended that the problem that can come within the category such as identification.
Illustrate the process of sequence labelling with Chinese word segmentation task.It " plays the song of XYZ assuming that inputting sentence now and looks into
Ask weather " (XYZ is the name of singer, is indicated herein with XYZ), task is that correctly this sentence is segmented.Firstly, handle
Sentence regards a series of linear list entries of individual character compositions as, and the task of sequence labelling is exactly to stamp a mark to each word
Label, for participle task, can define tag set are as follows: tag set={ B, M, E, O }, wherein B, which represents this word, is
The beginning character of sentence, M represent the intermediate character that this word is sentence, and E represents the termination character that this word is sentence, and O generation
List words.There are this four labels in tag set to input sentence " song of XYZ can be played and inquire weather "
(name that XYZ is certain singer) segment, and at this moment Chinese word segmentation is converted to the sequence labelling problem to word.Assuming that utilizing this
It is open to have trained sequence labelling model, then respectively to some label in the tagged set of each word, even if this
It is that participle finishes, final basis can be obtained by following output result:
" broadcast/B puts/E X/B Y/M Z/E /O song/E simultaneously/O looks into/B inquiry/E days/B gas/E ".Pass through after arranging again, in
It is that the example just passes through sequence labelling, is just segmented into following form: " broadcasting/XYZ//song/simultaneously/inquiry/weather ".
Similarly, the sequence labelling method that the disclosure provides can also be applied in intention assessment task.It is intended to, a sentence
Intention pointed by son or purpose, such as " song for playing XYZ ", correspondence is intended to " query music ".Firstly, list entries is regarded as
Then the linear order of word composition one by one it is as follows to define tag set:
Tag set={ 1,2 }, wherein 1 corresponding query music, 2 corresponding inquiry weather.
Equally, intention mark is carried out to input sentence " play the song of XYZ and inquire weather ", which includes two kinds
It is intended to, wherein first half sentence is the first intention, that is, query music, and later half sentence is second of intention, i.e. inquiry weather.It utilizes
Disclosed method carries out sequence labelling to above-mentioned input sentence, obtain " broadcast/1 put/1X/1Y/1Z/1 /O song/1 simultaneously/2 looks into/2
Ask/2 days/2 gas/2 ".
Knowledge mapping (Knowledge Graph, abbreviation KG), target will cover the letter of all entities and entity in the world
Breath.Knowledge mapping can be expressed as a triple (sub, rel, obj).For example: the father of Xiao Ming be it is big bright, be expressed as
Triple is (Xiao Ming, father are big bright).The former is main body, and centre is relationship, and the latter is object.Subject and object is referred to as reality
Body (entity).Relationship has an attribute, irreversible, that is to say, that subject and object cannot reverse.The collection of knowledge mapping
It closes, is chained up as a figure (graph), each node is that entity, each edge are a relationships one by one, in other words
One fact (fact).Namely digraph, main body are directed toward object.
In accordance with one embodiment of the present disclosure, a kind of intension recognizing method is provided.Preferably, above-mentioned intention assessment side
Method is for more intention assessments.In the disclosure, it is intended that be identified by sequence labelling method to realize.
As shown in Figure 1, the intension recognizing method, include the following steps: S11 by list entries Input knowledge map layer, and
Based on the feature in the knowledge mapping, the primary vector for generating the list entries is indicated;S12 indicates the primary vector
Pre-trained first nerves network layer is inputted, is indicated with generating the secondary vector of the list entries;S13 by described second to
Amount indicates that input nervus opticus network layer obtains the third vector expression of the list entries, and is based on the third vector table
Show the sequence labelling result for generating the list entries;And S15 is based on the sequence labelling as a result, obtaining and the input
The relevant one or more intentions of sequence.
Before step S11, input from the user is received in input layer.In addition, by list entries Input knowledge figure
Before composing layer, list entries can also be encoded.Such as list entries can be carried out using one-hot coding method
Coding.
In step s 11, it by list entries Input knowledge map layer, and based on the feature in the knowledge mapping, generates
The primary vector of the list entries indicates.Entity and one or more categories relevant to entity are stored in knowledge mapping
Property.Firstly, obtaining at least one entity relevant to list entries by inquiry knowledge mapping, and it is related to obtain above-mentioned entity
One or more attributes, and the relevant one or more attributes of entity are encoded, to form the first of list entries
Vector indicates.
Such as the one or more entities for including in list entries can be extracted by way of naming Entity recognition,
Above-mentioned entity is inquired in knowledge mapping again, and obtains the relevant one or more attributes of entity in turn, the attribute includes
Part of speech and the relationship of other entities etc..
For another example can be to carry out sequence composed by each word obtained after word segmentation processing to source text to list entries
Column.It is the word segmentation processing of Chinese text for list entries, the participle mode based on dictionary or based on statistics can be used.For defeated
Enter the word segmentation processing that sequence is English text, the participle mode such as word can be split according to space.Later again in knowledge mapping
In inquire one or more attributes of each word in list entries.
By above-mentioned knowledge mapping feature learning process, the generalization ability of model can greatly be improved.
In step s 12, the primary vector is indicated to input pre-trained first nerves network layer, described in generating
The secondary vector of list entries indicates.For example, generate the term vector of list entries using first nerves network model, and with from step
Primary vector expression is obtained in rapid S11 to be merged, and secondary vector expression is obtained.The term vector is by that can be embedded in by word
Discrete word sequence is converted into continuous space representation sequence vector by (word embedding) processing.Wherein, first nerves
Network model is the language model of pre-training, such as can be the models such as word2vec, ELMO and BERT.
Preferably, BERT (Bidirectional Encoder Representations from can be used
Transformers) model, for the model that Google issues in October, 2018, the purpose is to be adjusted in all layers by joint
Context carry out the two-way expression of pre-training depth.The model is all achieved very in 11 NLP tasks by pre-training and accurate adjustment
Good effect, meanwhile, the Transformer used makes entire model more efficient, can more capture the dependence of long range.Its
Main feature includes that first is two-stage model, and first stage bi-directional language model pre-training, second stage is according to Downstream Jobs
Accurate adjustment (Fine-tuning) is carried out, such as constructs language model in training, using big corpus A come train language model
Increase a small amount of neural net layer on the basis of language model to complete specific tasks, such as sequence labelling, semantic classification etc., then
There is supervision ground training pattern using markd corpus B;Second is that feature extraction uses Transformer as feature extraction
Device.
The execution sequence of above-mentioned steps S11 and S12 are only exemplary.It is described in another embodiment of the present disclosure
List entries can be concurrently input into knowledge mapping layer and first nerves network, by knowledge mapping layer and first nerves
The processing of network layer, obtaining primary vector respectively indicates and the term vector of list entries and then by above-mentioned primary vector table
Show and the term vector of list entries enters fusion, to obtain secondary vector expression.
In step s 13, the secondary vector is indicated into input nervus opticus network layer, obtains the of the list entries
Tri-vector is shown, and indicates formation sequence annotation results based on the third vector.Specifically, firstly, obtaining includes at least one
The tag set of a label, also, the sequence labelling result includes that each word mark in the list entries is upper described
One of label in tag set.And the nervus opticus network layer can be as two-way shot and long term memory network (BLSTM),
It is in two-way shot and long term memory network that hidden status switch that positive LSTM is exported and exporting at various locations for reversed LSTM is hidden
State carries out opsition dependent and splices to obtain complete hidden status switch;Later after dropout is set, a linear layer is accessed, it will
Hidden state vector dimensionality reduction, the dimension of number of tags into tag set, thus the sentence characteristics automatically extracted, that is, third
Vector indicates.The third vector indicates to include the mark marked each word in the list entries in the upper list of labels
The probability value of one of label.
For example, list entries be " play XYZ song simultaneously inquire weather ", tag set to be marked={ 1,2 }, wherein
1 corresponding query music, 2 corresponding inquiry weather, then each word in available list entries marks the probability of upper " 1 " or " 2 "
In value, such as " song for playing XYZ simultaneously inquires weather " on each character label the probability distribution of " 1 " or " 2 " be (0.6,0.1),
(0.7,0.1), (0.8,0.1), (0.9,0.1), (0.6,0.1), (0.6,0.1) (0.6,0.1), (0.7,0.1) (0.1,0.6)
(0.2,0.7) (0.2,0.8), (0.1,0.8), (0.1,0.9).In one embodiment of the disclosure, directly can therefrom take most
Greatest obtains sequence labelling result.It will be understood by those skilled in the art that indicating that each word marks the upper list of labels
In the vector of probability value of one of label different dimensions can be had according to the difference of the number of label to be marked.
Alternatively, according to another embodiment of the present disclosure, in step S14, the secondary vector is indicated into input the
After two neural networks obtain the third vector expression of the list entries, the third vector is indicated into input third nerve net
Network layers, such as condition random field obtain the sequence labelling result of the list entries.Sentence can be carried out in conditional random field models
The sequence labelling of sub- grade.Before output access CRF layers of emphasis be done using label transition probability sentence level label it is pre-
It surveys, so that annotation process is no longer to each word independent sorting (such as in BLSTM layers).
Those skilled in the art are, it should be understood that the model shown in above-mentioned steps S13 and S14 for sequence labelling is only to show
Example property, the combination of other models or model for sequence mark can also be used for the nerve net in alternative steps S13 or S14
Network layers, to realize sequence labelling task.In above-mentioned embodiment of the disclosure, pass through the execution of step S11-S13 or S11-S14
Complete sequence labelling task.
In step S15, after obtaining sequence labelling result in above-mentioned sequence labelling method, it is based on the sequence labelling
Result obtain one or more intention relevant to the list entries.Preferably, it can identify related to list entries
Multiple intentions.Also, the multiple intention is associated with the corresponding part of the list entries.For example, for list entries X
=x1,x2,x3,...,xn, identify and be intended to I=i1,i2,i3,...im.Wherein, m is the number being intended to.i1In list entries
Preceding i element composition sequence x1,x2,x3,...,xiIt is associated, i2It is formed with i-th to j-th element in list entries
Sequence xi,xi+1,xi+2,...,xjIt is associated, and so on.
In another embodiment of the present disclosure, as shown in Fig. 2, realizing a kind of intension recognizing method.Firstly, in input layer
Receive the input of sequence, such as input sentence --- " put the song of XYZ and inquire weather ".Then, in knowledge mapping layer (in Fig. 2
KG layer) in, the vector of the knowledge encoding inquired in knowledge picture library to list entries is indicated.Later, at BERT layers, pass through
After the BERT model of pre-training carries out accurate adjustment, term vector is obtained.It is subsequent to obtain being intended to mark by BLSTM layers and CRF layers
As a result.As shown in Figure 2, " song of XYZ is put and inquires weather " and be marked with that " 1/ song 1/ and 2/ for putting 1/X1/Y1/Z1/ is looked into
2/ askes 2/ day 2/ gas 2/ ", wherein 1 corresponding " Cha Yinle ", 2 corresponding " Cha Tianqi ".To identify that input sentence " puts the song of XYZ
And inquire weather " it include two intentions, respectively " Cha Yinle " and " Cha Tianqi ".Further, it is possible to obtain and " Cha Yinle " phase
Corresponding is the first six word inputted in sentence, and corresponding with " Cha Tianqi " is rear five words inputted in sentence.
According to a further embodiment of the disclosure, a kind of intention assessment device 30 is provided.As shown in figure 3, the intention is known
Other device 30 includes input module 31, first processing module 32, Second processing module 33, third processing module 34, fourth process
Module 35 and the 5th processing module 36.Wherein input module 31, for receiving the input of list entries;
First processing module 32, it is raw by list entries Input knowledge map layer, and based on the feature in the knowledge mapping
It is indicated at the primary vector of the list entries;
Second processing module 33 indicates the primary vector to input pre-trained first nerves network layer, to generate
The secondary vector of the list entries indicates;
Third processing module 34, for the secondary vector to be indicated that input nervus opticus network layer obtains the input sequence
The third vector of column indicates, and the sequence labelling result of the generation list entries is indicated based on the third vector;And
Fourth processing module 35, for based on the sequence labelling result obtain it is relevant to the list entries one or
Multiple intentions;And
5th processing module 36 is obtained for receiving the input of the expression of the third vector in third processing module 34
The sequence labelling result of the list entries.
And the treatment process executed in above-mentioned each module is opposite with the respective process specifically described in the above method respectively
It answers.
In the disclosure, for the shortcoming of traditional grammar parsing tree, the sequence labelling method of the disclosure passes through inquiry
It with the information in learning knowledge map, and introduces pre-training model and it is finely adjusted, be re-used as subsequent neural network model
Input, make up the missing on the related information between prior information and sample.Turn one's knowledge to advantage map simultaneously, thus
Better effect is obtained in more intention assessment tasks, is reduced manually in the degree of participation of more intention assessments, is obtained one and more lead to
With, can safeguard, readily understood, high-precision model.
The disclosure also provides a kind of electronic equipment, as shown in figure 4, the equipment includes: communication interface 1000, memory 2000
With processor 3000.Communication interface 1000 carries out data interaction for being communicated with external device.In memory 2000
It is stored with the computer program that can be run on processor 3000.Processor 3000 is realized above-mentioned when executing the computer program
Method in embodiment.The quantity of the memory 2000 and processor 3000 can be one or more.
Memory 2000 may include high speed RAM memory, can also further include nonvolatile memory (non-
Volatile memory), a for example, at least magnetic disk storage.
If communication interface 1000, memory 2000 and the independent realization of processor 3000, communication interface 1000, memory
2000 and processor 3000 can be connected with each other by bus and complete mutual communication.The bus can be industrial standard
Architecture (ISA, Industry Standard Architecture) bus, external equipment interconnection (PCI, Peripheral
Component) bus or extended industry-standard architecture (EISA, Extended Industry Standard
Component) bus etc..The bus can be divided into address bus, data/address bus, control bus etc..For convenient for expression, the figure
In only indicated with a thick line, it is not intended that an only bus or a type of bus.
Optionally, in specific implementation, if communication interface 1000, memory 2000 and processor 3000 are integrated in one
On block chip, then communication interface 1000, memory 2000 and processor 3000 can complete mutual lead to by internal interface
Letter.
Any process described otherwise above or method description are construed as in flow chart or herein, and expression includes
Module, segment or the portion of the code of the executable instruction for the step of one or more is for realizing specific logical function or process
Point, and the range of the preferred embodiment of the disclosure includes other realization, wherein can not press shown or discussed suitable
Sequence, including according to related function by it is basic simultaneously in the way of or in the opposite order, Lai Zhihang function, this should be by the disclosure
Embodiment person of ordinary skill in the field understood.Processor executes each method as described above and processing.
For example, the method implementation in the disclosure may be implemented as computer software programs, being tangibly embodied in machine can
Read medium, such as memory.In some embodiments, some or all of of computer software programs can be via memory
And/or communication interface and be loaded into and/or install.When computer software programs are loaded into memory and are executed by processor,
One or more steps in method as described above can be executed.Alternatively, in other embodiments, processor can lead to
It crosses other any modes (for example, by means of firmware) appropriate and is configured as executing one of above method.
Expression or logic and/or step described otherwise above herein in flow charts, may be embodied in any
In computer-readable medium, for instruction execution system, device or equipment (such as computer based system, including processor
System or other can be from instruction execution system, device or equipment instruction fetch and the system executed instruction) use, or combine these
Instruction execution system, device or equipment and use.
For the purpose of this specification, " computer readable storage medium ", which can be, any may include, store, communicating, propagating
Or transfer program uses for instruction execution system, device or equipment or in conjunction with these instruction execution systems, device or equipment
Device.The more specific example (non-exhaustive list) of computer-readable medium include the following: there are one or more wirings
Electrical connection section (electronic device), portable computer diskette box (magnetic device), random access memory (RAM), read-only memory
(ROM), erasable edit read-only storage (EPROM or flash memory), fiber device and portable read-only memory
(CDROM).In addition, computer-readable medium can even is that the paper that can print described program on it or other suitable Jie
Matter, because can then be edited, be interpreted or when necessary with other for example by carrying out optical scanner to paper or other media
Suitable method is handled electronically to obtain described program, is then stored in computer storage.
It should be appreciated that each section of the disclosure can be realized with hardware, software or their combination.In above-mentioned embodiment party
In formula, multiple steps or method can carry out reality in memory and by the software that suitable instruction execution system executes with storage
It is existing.It, and in another embodiment, can be in following technology well known in the art for example, if realized with hardware
Any one or their combination are realized: having a discrete logic for realizing the logic gates of logic function to data-signal
Circuit, the specific integrated circuit with suitable combinational logic gate circuit, programmable gate array (PGA), field-programmable gate array
Arrange (FPGA) etc..
Those skilled in the art are understood that realize all or part of the steps of above embodiment method
It is that relevant hardware can be instructed to complete by program, the program can store in a kind of computer readable storage medium
In, which when being executed, includes the steps that one or a combination set of method implementation.
In addition, can integrate in a processing module in each functional unit in each embodiment of the disclosure, it can also
To be that each unit physically exists alone, can also be integrated in two or more units in a module.It is above-mentioned integrated
Module both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module
If in the form of software function module realize and when sold or used as an independent product, also can store one calculating
In machine readable storage medium storing program for executing.The storage medium can be read-only memory, disk or CD etc..
In the description of this specification, reference term " an embodiment/mode ", " some embodiment/modes ",
The description of " example ", " specific example " or " some examples " etc. means the embodiment/mode or example is combined to describe specific
Feature, structure, material or feature are contained at least one embodiment/mode or example of the application.In this specification
In, schematic expression of the above terms are necessarily directed to identical embodiment/mode or example.Moreover, description
Particular features, structures, materials, or characteristics can be in any one or more embodiment/modes or example in an appropriate manner
In conjunction with.In addition, without conflicting with each other, those skilled in the art can be by different implementations described in this specification
Mode/mode or example and different embodiments/mode or exemplary feature are combined.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present application, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
It will be understood by those of skill in the art that above embodiment is used for the purpose of clearly demonstrating the disclosure, and simultaneously
Non- be defined to the scope of the present disclosure.For those skilled in the art, may be used also on the basis of disclosed above
To make other variations or modification, and these variations or modification are still in the scope of the present disclosure.
Claims (10)
1. a kind of intension recognizing method characterized by comprising
By list entries Input knowledge map layer, and based on the feature in the knowledge mapping, the of the list entries is generated
One vector indicates;
The primary vector is indicated to input pre-trained first nerves network layer, with generate the second of the list entries to
Amount indicates;
The secondary vector is indicated that input nervus opticus network layer obtains the third vector expression of the list entries, and base
The sequence labelling result of the generation list entries is indicated in the third vector;And
Based on the sequence labelling as a result, obtaining one or more intentions relevant to the list entries.
2. it is described defeated that the secondary vector is the method for claim 1, wherein indicated that input nervus opticus network obtains
After the third vector expression for entering sequence,
The third vector is indicated into input third nerve network layer, obtains the sequence labelling result of the list entries.
3. the method as described in claim 1, which is characterized in that before by the list entries Input knowledge map layer,
List entries is encoded, and by the list entries Input knowledge map layer after coding.
4. method according to any one of claims 1 to 3, which is characterized in that
The first nerves network layer carries out pre-training, and the accurate adjustment including being directed to sequence labelling task using bi-directional language model
Operation.
5. method according to any one of claims 1 to 4, which is characterized in that further include obtaining including at least one label
Tag set, also, the sequence labelling result includes that each word in the list entries is marked the upper tally set
One of label in conjunction.
6. the method as described in any one of claims 1 to 5, which is characterized in that
The nervus opticus network layer is two-way shot and long term memory network, and the secondary vector expression includes by the list entries
In each word mark the probability value of one of label in the upper list of labels.
7. method according to claim 2, which is characterized in that the third nerve network layer is condition random field.
8. a kind of intention assessment device characterized by comprising
First processing module is used for list entries Input knowledge map layer, and based on the feature in the knowledge mapping, is generated
The primary vector of the list entries indicates;
Second processing module inputs pre-trained first nerves network layer for indicating the primary vector, to generate
The secondary vector for stating list entries indicates;
Third processing module, for the secondary vector to be indicated that input nervus opticus network layer obtains the of the list entries
Tri-vector is shown, and the sequence labelling result of the generation list entries is indicated based on the third vector;And
Fourth processing module, for obtaining one or more meanings relevant to the list entries based on the sequence labelling result
Figure.
9. a kind of electronic equipment characterized by comprising
Memory, the memory store executable instruction;And
Processor, the processor executes the executable instruction of the memory storage, so that the processor executes such as right
It is required that method described in any one of 1 to 7.
10. a kind of readable storage medium storing program for executing, which is characterized in that be stored with executable instruction in the readable storage medium storing program for executing, it is described can
It executes instruction when being executed by processor for realizing the method as described in any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910285223.XA CN109992671A (en) | 2019-04-10 | 2019-04-10 | Intension recognizing method, device, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910285223.XA CN109992671A (en) | 2019-04-10 | 2019-04-10 | Intension recognizing method, device, equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109992671A true CN109992671A (en) | 2019-07-09 |
Family
ID=67132756
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910285223.XA Pending CN109992671A (en) | 2019-04-10 | 2019-04-10 | Intension recognizing method, device, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109992671A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110390100A (en) * | 2019-07-16 | 2019-10-29 | 广州小鹏汽车科技有限公司 | Processing method, the first electric terminal, the second electric terminal and processing system |
CN110489521A (en) * | 2019-07-15 | 2019-11-22 | 北京三快在线科技有限公司 | Text categories detection method, device, electronic equipment and computer-readable medium |
CN110502643A (en) * | 2019-08-28 | 2019-11-26 | 南京璇玑信息技术有限公司 | A kind of next model autocreating technology of the prediction based on BERT model |
CN111222327A (en) * | 2019-12-23 | 2020-06-02 | 东软集团股份有限公司 | Word embedding representation method, device and equipment |
CN111552821A (en) * | 2020-05-14 | 2020-08-18 | 北京华宇元典信息服务有限公司 | Legal intention searching method, legal intention searching device and electronic equipment |
CN111832282A (en) * | 2020-07-16 | 2020-10-27 | 平安科技(深圳)有限公司 | External knowledge fused BERT model fine adjustment method and device and computer equipment |
CN113360751A (en) * | 2020-03-06 | 2021-09-07 | 百度在线网络技术(北京)有限公司 | Intention recognition method, apparatus, device and medium |
WO2021208696A1 (en) * | 2020-11-19 | 2021-10-21 | 平安科技(深圳)有限公司 | User intention analysis method, apparatus, electronic device, and computer storage medium |
RU2762702C2 (en) * | 2020-04-28 | 2021-12-22 | Публичное Акционерное Общество "Сбербанк России" (Пао Сбербанк) | System and method for automated assessment of intentions and emotions of users of dialogue system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107679039A (en) * | 2017-10-17 | 2018-02-09 | 北京百度网讯科技有限公司 | The method and apparatus being intended to for determining sentence |
CN109145153A (en) * | 2018-07-02 | 2019-01-04 | 北京奇艺世纪科技有限公司 | It is intended to recognition methods and the device of classification |
CN109388793A (en) * | 2017-08-03 | 2019-02-26 | 阿里巴巴集团控股有限公司 | Entity mask method, intension recognizing method and corresponding intrument, computer storage medium |
-
2019
- 2019-04-10 CN CN201910285223.XA patent/CN109992671A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109388793A (en) * | 2017-08-03 | 2019-02-26 | 阿里巴巴集团控股有限公司 | Entity mask method, intension recognizing method and corresponding intrument, computer storage medium |
CN107679039A (en) * | 2017-10-17 | 2018-02-09 | 北京百度网讯科技有限公司 | The method and apparatus being intended to for determining sentence |
CN109145153A (en) * | 2018-07-02 | 2019-01-04 | 北京奇艺世纪科技有限公司 | It is intended to recognition methods and the device of classification |
Non-Patent Citations (1)
Title |
---|
林先辉: "面向出行领域的任务型对话系统研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110489521A (en) * | 2019-07-15 | 2019-11-22 | 北京三快在线科技有限公司 | Text categories detection method, device, electronic equipment and computer-readable medium |
CN110390100B (en) * | 2019-07-16 | 2023-10-31 | 广州小鹏汽车科技有限公司 | Processing method, first electronic terminal, second electronic terminal and processing system |
CN110390100A (en) * | 2019-07-16 | 2019-10-29 | 广州小鹏汽车科技有限公司 | Processing method, the first electric terminal, the second electric terminal and processing system |
CN110502643A (en) * | 2019-08-28 | 2019-11-26 | 南京璇玑信息技术有限公司 | A kind of next model autocreating technology of the prediction based on BERT model |
CN111222327A (en) * | 2019-12-23 | 2020-06-02 | 东软集团股份有限公司 | Word embedding representation method, device and equipment |
CN111222327B (en) * | 2019-12-23 | 2023-04-28 | 东软集团股份有限公司 | Word embedding representation method, device and equipment |
CN113360751A (en) * | 2020-03-06 | 2021-09-07 | 百度在线网络技术(北京)有限公司 | Intention recognition method, apparatus, device and medium |
RU2762702C2 (en) * | 2020-04-28 | 2021-12-22 | Публичное Акционерное Общество "Сбербанк России" (Пао Сбербанк) | System and method for automated assessment of intentions and emotions of users of dialogue system |
CN111552821B (en) * | 2020-05-14 | 2022-03-01 | 北京华宇元典信息服务有限公司 | Legal intention searching method, legal intention searching device and electronic equipment |
CN111552821A (en) * | 2020-05-14 | 2020-08-18 | 北京华宇元典信息服务有限公司 | Legal intention searching method, legal intention searching device and electronic equipment |
WO2021139266A1 (en) * | 2020-07-16 | 2021-07-15 | 平安科技(深圳)有限公司 | Fine-tuning method and apparatus for external knowledge-fusing bert model, and computer device |
CN111832282A (en) * | 2020-07-16 | 2020-10-27 | 平安科技(深圳)有限公司 | External knowledge fused BERT model fine adjustment method and device and computer equipment |
WO2021208696A1 (en) * | 2020-11-19 | 2021-10-21 | 平安科技(深圳)有限公司 | User intention analysis method, apparatus, electronic device, and computer storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109992671A (en) | Intension recognizing method, device, equipment and storage medium | |
CN110427623B (en) | Semi-structured document knowledge extraction method and device, electronic equipment and storage medium | |
CN108984683B (en) | Method, system, equipment and storage medium for extracting structured data | |
CN109063221A (en) | Query intention recognition methods and device based on mixed strategy | |
CN111552821B (en) | Legal intention searching method, legal intention searching device and electronic equipment | |
CN108595708A (en) | A kind of exception information file classification method of knowledge based collection of illustrative plates | |
CN111695345B (en) | Method and device for identifying entity in text | |
CN107015969A (en) | Can self-renewing semantic understanding System and method for | |
CN103530282A (en) | Corpus tagging method and equipment | |
CN111159385A (en) | Template-free universal intelligent question-answering method based on dynamic knowledge graph | |
CN112231472B (en) | Judicial public opinion sensitive information identification method integrated with domain term dictionary | |
CN112182229A (en) | Text classification model construction method, text classification method and device | |
CN109685056A (en) | Obtain the method and device of document information | |
CN111738016A (en) | Multi-intention recognition method and related equipment | |
CN111723569A (en) | Event extraction method and device and computer readable storage medium | |
CN111311459A (en) | Interactive question setting method and system for international Chinese teaching | |
CN114661872A (en) | Beginner-oriented API self-adaptive recommendation method and system | |
CN114580424B (en) | Labeling method and device for named entity identification of legal document | |
CN114661951A (en) | Video processing method and device, computer equipment and storage medium | |
CN112380848B (en) | Text generation method, device, equipment and storage medium | |
CN117112767A (en) | Question and answer result generation method, commercial query big model training method and device | |
CN115115432B (en) | Product information recommendation method and device based on artificial intelligence | |
CN111274354A (en) | Referee document structuring method and device | |
CN113763947B (en) | Voice intention recognition method and device, electronic equipment and storage medium | |
CN115273856A (en) | Voice recognition method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190709 |