CN106847271A - A kind of data processing method and device for talking with interactive system - Google Patents
A kind of data processing method and device for talking with interactive system Download PDFInfo
- Publication number
- CN106847271A CN106847271A CN201611139722.0A CN201611139722A CN106847271A CN 106847271 A CN106847271 A CN 106847271A CN 201611139722 A CN201611139722 A CN 201611139722A CN 106847271 A CN106847271 A CN 106847271A
- Authority
- CN
- China
- Prior art keywords
- result
- dialogue
- data
- model
- generation
- 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
- 230000002452 interceptive effect Effects 0.000 title claims abstract description 45
- 238000003672 processing method Methods 0.000 title claims abstract description 11
- 230000003993 interaction Effects 0.000 claims abstract description 53
- 238000000034 method Methods 0.000 claims abstract description 49
- 230000006870 function Effects 0.000 claims abstract description 34
- 230000002708 enhancing effect Effects 0.000 claims abstract description 21
- 238000004458 analytical method Methods 0.000 claims abstract description 18
- 230000011218 segmentation Effects 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 10
- 230000007246 mechanism Effects 0.000 claims description 9
- 230000007935 neutral effect Effects 0.000 claims description 7
- 210000005036 nerve Anatomy 0.000 claims 1
- 238000010030 laminating Methods 0.000 abstract description 3
- 210000002569 neuron Anatomy 0.000 description 15
- 230000009471 action Effects 0.000 description 12
- 230000008569 process Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 206010019233 Headaches Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 231100000869 headache Toxicity 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 238000007476 Maximum Likelihood Methods 0.000 description 2
- 239000003795 chemical substances by application Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000021760 high fever Diseases 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000012892 rational function Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L2015/225—Feedback of the input speech
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Human Computer Interaction (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Physiology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Machine Translation (AREA)
Abstract
A kind of data processing method and device for talking with interactive system, wherein, the method includes：Interaction data obtaining step, obtains the dialogue interaction data of user input；Dialog model result generation step, is parsed to dialogue interaction data, and analysis result is input in default dialogue generation model, obtains dialog model result, wherein, dialogue generation model is built based on NRM and enhancing learning algorithm；Feedback data exports step, generates corresponding feedback data according to dialog model result and exports.Compared to existing method, this method can solve the problem that the more conservative problem of the feedback result of the dialogue generation model generation of the coding layer decoding layer framework for being currently based on MLE object functions, so also allow for ultimately generating and exporting the feedback data more hommization to user and the actual interaction scenarios of laminating, so as to improve the Consumer's Experience of dialogue interactive system.
Description
Technical field
The present invention relates to human-computer interaction technique field, specifically, it is related at a kind of data for talking with interactive system
Reason method and device.
Background technology
With continuing to develop for science and technology, the introducing of information technology, computer technology and artificial intelligence technology, machine
Industrial circle is progressively walked out in the research of people, gradually extend to the neck such as medical treatment, health care, family, amusement and service industry
Domain.And people for the requirement of robot also conform to the principle of simplicity the multiple mechanical action of substance be promoted to anthropomorphic question and answer, independence and with
The intelligent robot that other robot is interacted, man-machine interaction also just turns into the key factor for determining intelligent robot development.
Dialogue interactive system plays act foot light as the nucleus module in man-machine interactive system in man-machine interactive system
The role of weight.In daily interactive process, the dialogue between user and robot is interacted and occupies most of ratio
Example.Therefore, the quality of dialogue interactive system is directly connected to the Consumer's Experience of intelligent robot product, and then influences going for user
Stay, the dialogue interaction capabilities for how lifting intelligent robot are then problem demanding prompt solutions.
The content of the invention
To solve the above problems, the invention provides a kind of data processing method for talking with interactive system, it includes：
Interaction data obtaining step, obtains the dialogue interaction data of user input；
Dialog model result generation step, parses to the dialogue interaction data, analysis result is input to default
In dialogue generation model, dialog model result is obtained, wherein, the dialogue generation model is calculated based on NRM and enhancing study
What method built；
Feedback data exports step, generates corresponding feedback data according to the dialog model result and exports.
According to one embodiment of present invention, the default dialogue generation model includes coding layer and decoding layer, wherein, institute
The whole vector tables for stating the word segmentation result generation word segmentation result of the coding layer for being obtained according to parsing reach and part vector table
Reach, and whole vector tables are reached and part vector table is up to splicing, obtain splicing result, the decoding layer is used to be based on
Notice mechanism generates the dialog model result according to the splicing result.
According to one embodiment of present invention, the decoding layer is based on the neutral net of enhancing learning algorithm and engages in the dialogue mould
The generation of type result.
According to one embodiment of present invention, in the enhancing learning algorithm, using MMI functions as reward function.
According to one embodiment of present invention, methods described also includes：
Modifying model step, obtains input data of the user for the feedback data, and according to the input number
It is modified according to the default dialogue generation model.
Present invention also offers a kind of data processing equipment for talking with interactive system, it includes：
Interaction data acquisition module, its dialogue interaction data for being used to obtain user input；
Dialog model result-generation module, it is used to parse the dialogue interaction data, analysis result is input into
To in default dialogue generation model, dialog model result is obtained, wherein, the dialogue generation model is based on NRM and enhancing
What learning algorithm built；
Feedback data output module, it is used to generate corresponding feedback data according to the dialog model result and export.
According to one embodiment of present invention, the default dialogue generation model includes coding layer and decoding layer, wherein, institute
Coding layer is stated to be configured to be reached and partly vectorial according to the whole vector tables for parsing the word segmentation result generation word segmentation result for obtaining
Expression, and whole vector tables are reached and part vector table is up to splicing, splicing result is obtained, the decoding layer is configured to
The dialog model result is generated according to the splicing result based on notice mechanism.
According to one embodiment of present invention, the decoding layer be configured to strengthen learning algorithm neutral net carry out
The generation of feedback data.
According to one embodiment of present invention, in the enhancing learning algorithm, using MMI functions as reward function.
According to one embodiment of present invention, described device also includes：
Modifying model module, it is used to obtain input data of the user for the feedback data, and according to described
Input data is modified to the default dialogue generation model.
Compared to existing method, the data processing method for talking with interactive system provided by the present invention can be based on use
The enquirement at family, generates more reasonably answer, meanwhile, this method can also solve to be currently based on the coding layer-solution of MLE object functions
The more conservative problem of feedback result of the dialogue generation model generation of code layer framework, so also allows for ultimately generating and exporting
Feedback data more hommization and the actual interaction scenarios of laminating to user, so as to improve the user's body of dialogue interactive system
Test.
Additionally, this method can also by object function it is self-defined come cause dialogue interactive system and user between
Dialogue interaction according to more have interest, diversity, interactivity and continuation, this be favorably improved dialogue interaction be
User's viscosity of system.
Other features and advantages of the present invention will be illustrated in the following description, also, the partly change from specification
Obtain it is clear that or being understood by implementing the present invention.The purpose of the present invention and other advantages can be by specification, rights
Specifically noted structure is realized and obtained in claim and accompanying drawing.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing wanted needed for technology description to do simple introduction：
Fig. 1 is according to an embodiment of the invention to realize that flow is shown for talk with the data processing method of interactive system
It is intended to；
Fig. 2 is that dialog model result generation step according to an embodiment of the invention implements schematic flow sheet；
Fig. 3 is the structural representation of dialogue generation model according to an embodiment of the invention；
Fig. 4 is the structural representation for talking with the data processing equipment of interactive system according to an embodiment of the invention
Figure.
Specific embodiment
Describe embodiments of the present invention in detail below with reference to drawings and Examples, how the present invention is applied whereby
Technological means solves technical problem, and reaches the implementation process of technique effect and can fully understand and implement according to this.Need explanation
As long as not constituting conflict, each embodiment in the present invention and each feature in each embodiment can be combined with each other,
The technical scheme for being formed is within protection scope of the present invention.
Meanwhile, in the following description, many details are elaborated for illustrative purposes, to provide to of the invention real
Apply the thorough understanding of example.It will be apparent, however, to one skilled in the art, that the present invention can be without tool here
Body details or described ad hoc fashion are implemented.
In addition, can be in the such as one group department of computer science of computer executable instructions the step of the flow of accompanying drawing is illustrated
Performed in system, and, although logical order is shown in flow charts, but in some cases, can be with different from herein
Order perform shown or described step.
The data object of dialogue interactive system treatment mainly includes customer problem and answer.According to the affiliated number of customer problem
According to field, dialogue interactive system is commonly divided into the question answering system towards defined domain, the question answering system towards open field and face
To the dialogue interactive system of common problem collection.Originated according to the different pieces of information of answer, dialogue interactive system can be divided into based on knot
The dialogue interactive system of structure data, the dialogue interactive system based on free text and the dialogue based on question and answer pair interaction system
System.
Additionally, being divided according to the generation feedback mechanism of answer, it is right based on retrieval type that dialogue interactive system is further divided into
Words interactive system and the dialogue interactive system based on production.Wherein, retrieval type dialogue interactive system is from the language for having existed
The answer information with problem information best match, the degree of accuracy of the answer information that retrieval type dialogue interactive system is obtained are searched in material
It is higher but adaptability is poor.In contrast, the answer information of production dialogue interactive system is by substantial amounts of language material and machine
Learning algorithm parsing obtain, production dialogue interactive system obtained by answer information have good system suitability but at present
Method there is no to ensure accuracy rate higher.
For the above mentioned problem in the presence of prior art, a kind of number for talking with interactive system is present embodiments provided
According to processing method.
In order to clearly illustrate the reality of the data processing method for talking with interactive system that the present embodiment is provided
Existing principle, implementation process and advantage, further say below in conjunction with schematic flow sheet is realized shown in Fig. 1 to the method
It is bright.
As shown in figure 1, the data processing method for talking with interactive system that the present embodiment is provided is first in step
The dialogue interaction data of user input is obtained in S101.It is pointed out that according to actual needs, the method is in step S101
Accessed dialogue interaction data both can be that the voice collected using audio collecting device (such as microphone etc.) is handed over
Mutual data, or the text interaction data collected using text collection equipment (such as keyboard etc.), the present invention are not limited
In this.
After above-mentioned dialogue interaction data is obtained, the method can be solved to above-mentioned dialogue interaction data in step s 102
Analysis, so as to obtain analysis result.Finally, can be input into for analysis result resulting in step S102 in step s 103 by the method
To in default dialogue generation model, so as to generate corresponding dialog model according to above-mentioned analysis result by the dialogue generation model
As a result.
In order to overcome existing retrieval type dialog generation system to generate the problem of answer information poor accuracy, the present embodiment
In, the method employs the default dialogue generation model built based on NRM and enhancing learning algorithm in step s 103.
Fig. 2 is parsed and corresponding using generation model generation is talked with showing the present embodiment to dialogue interaction data
Dialog model result implement schematic flow sheet, Fig. 3 shows the structural representation for talking with generation model in the present embodiment
Figure, is further elaborated below in conjunction with Fig. 2 and Fig. 3.
As shown in Fig. 2 when the method is parsed to above-mentioned dialogue interaction data, preferably in step s 201 to above-mentioned
Dialogue interaction data carries out word segmentation processing, so as to obtain talking with the word segmentation result of interaction data.It is pointed out that in the present invention
Different embodiments in, according to actual needs, the method to dialogue interaction data carry out word segmentation processing when, can both use base
In the mode of string matching, it would however also be possible to employ the mode based on statistics and machine learning, or it is using other reasonable sides
Formula, the invention is not restricted to this.
After the analysis result for obtaining talking with interaction data, the method can be in step S202 according to gained in step S201
To word segmentation result come utilize default dialogue generation model generate whole vector tables of above-mentioned word segmentation result up to and part vector table
Reach.
In the present embodiment, the dialogue generation model that the method is used includes coding layer and decoding layer.Wherein, coding layer
RNN models are employed with decoding layer.Different from traditional FNN (Feed-forward Neural Networks, feed-forward
Neutral net) model, RNN models introduce directed circulation, its problem that can process forward-backward correlation between those inputs.
It, again to output layer, is between layers full connection that traditional neural network model is from input layer to hidden layer,
Node between every layer is connectionless.But this common neutral net is for many problems but helpless.For example, you
What the next word for predicting sentence is, generally requires the word used above, because front and rear word is simultaneously in a sentence
It is not independent.
In recent years, deep neural network is furtherd investigate and achieved in the tasks such as image classification, speech recognition
Prominent effect, shows excellent expression learning ability.At the same time, language learning is represented by deep neural network
It has been increasingly becoming a new research tendency.
However, flexible and changeable and semantic information complicated abstract due to human language so that deep neural network mould
Type shows that application in study is faced than in the bigger challenge of image, voice in language table.First, compared to voice and image, language
Speech is tone artifacts, is entirely that the notation for being produced by brain and being processed, is human civilization wisdom in human civilization process
Height embody, the variability and flexibility ratio of language are considerably beyond image and voice signal.Second, image and voice have clearly
Mathematical notation, such as gray level image is numerical matrix mathematically, and its represent minimum particle size element have determination
Physical significance, the value of each point of image pixel represents certain gray scale color-values.
Therefore the method that the present embodiment is provided employs NRM combination RNN neutral nets to be analyzed, and why is RNN
Neural network is referred to as circulated, because the current output of sequence is also relevant with output above.Specifically the form of expression is
Network can be remembered and be applied in the calculating of current output to information above, i.e., the node between hidden layer is no longer without even
Connect but have connection, and the input of hidden layer not only exports also defeated including last moment hidden layer including input layer
Go out.In theory, RNN can be processed the sequence data of any length.
As shown in figure 3, in the present embodiment, the input data of each neuron is to divide in the input layer Input X of coding layer
Each participle in word result, such as xi represents i-th participle in word segmentation result.It is each in the hidden layer Layer H of coding layer
Individual neuron is preferably a nonlinear function (such as logistic function, general using LSTM/GRU cores), right
For i-th neuron hi in hidden layer, the input of the neuron is hi-1 and xi.In the present embodiment, coding layer it is hidden
Tibetan layer is output as the whole vector tables to analysis result and reaches.
In the present embodiment, based on NRM, coding layer employs mixed mechanism, specifically, similar with hidden layer Layer H,
Each neuron is preferably a nonlinear function in neuronal layers Layer G, comes for i-th neuron gi therein
Say, the input of the neuron is xi and gi-1.Neuronal layers Layer G are output as a part vector to analysis result
Expression.
From foregoing description as can be seen that the present embodiment in, coding layer by using mixed mechanism come according to step S201
Analysis result obtain corresponding vector table and reach, this mode can not only realize the dialogue interaction data being input into user
All hold, be also just so the feedback for finally being generated while the detailed information of dialogue interaction data can also be sufficiently reserved
The degree of accuracy of data and relevance grade provide the foundation.
As shown in Figures 2 and 3, in the present embodiment, reached and part vector in the whole vector tables for obtaining above-mentioned analysis result
After expression, the method can reach to whole vector tables of above-mentioned analysis result in step S203 and part vector table is up to spelling
Connect, so as to obtain splicing result.
After splicing result is obtained, the method can be come according to resulting in step S203 in step S204 by decoding layer
Splicing result generates dialog model result.Specifically, in the present embodiment, based on NRM, decoding layer is preferably by based on notice
The mode of mechanism to generate corresponding dialog model result according to splicing result resulting in step S203.
To sum up, NRM uses a kind of mixed mechanism in coded portion, so that the sequence that coding obtains intermediate representation can not only
The overall assurance of user's sentence information is enough realized, while the detailed information of sentence can also be sufficiently reserved.And adopted in decoded portion
With notice mechanism, so that the complex interactive mode in the grasp question answering process that generation model can be relatively easy.
Further, in the present embodiment, decoding layer generates dialogue using the circulation neural network based on enhancing learning algorithm
Model result.Enhancing study is a kind of learning method different from supervised learning, unsupervised learning, and it can solve the problem that similar playing chess
In to do the problem that a series of decision-making just can determine that final result, and this exactly supervised learning and unsupervised learning institute solution be never
.
Enhancing study is made up of intelligent body agent and the parts of environment environment two, it comprises several big elements, i.e.,
State set state (i.e. all states being likely to occur, corresponding to the location sets of chess piece in the problem of playing chess), action collection action
(i.e. all possible action, corresponding to the scheme of beginning of chess piece in the problem of playing chess), decision function are (i.e. in certain state state
Generate the process of some action action) and reward function reward function (i.e. for rewarding the target letter of action
Number).So, the whole decision process of intelligent body agent is just analogous to " state-(action)-state-(action)-state-... "
Flow.
For existing enhancing learning algorithm, it is generally used as target letter using the negative logarithm of maximum likelihood function
Number.Maximal possibility estimation (MLE) knowledge-chosen makes data i.e. (i.e. decoding layer the exports elongated sentence) probability of occurrence observed
The value of maximum parameter (parameter i.e. in coding layer-decoding layer).However, existing this enhancing learning algorithm is easily caused institute
The answer information for being directed to problem information for obtaining is too conservative.That is, in the case where a problem information is given, using maximum seemingly
The negative logarithm of right function is tended to generate a too conservative answer information as the enhancing learning algorithm of object function, and this is answered
Case information is not easy mistake semantically occur, but simultaneously also without actual meaning.
For example, for the problem information of " headache of coming to school is caught a cold ", using the negative logarithm of maximum likelihood function as mesh
The enhancing learning algorithm of scalar functions would generally be obtained such as " I is closed the border in practice, and we giocoso play again after a while ".
Regarding to the issue above, the method that the present embodiment is provided is in step S204 based on enhancing study according to step S203
In resulting result when generating corresponding dialog model result, its object function employs maximum mutual information (MMI) function.With
Based on unlike MLE object functions, the enhancing learning algorithm based on MMI functions can set up input sentence and output sentence it
Between mutual information.
As shown in figure 3, in the present embodiment, used as an intelligent body, the parameter in decoding layer is used as decision-making for decoding layer
, used as a state, the output data of each neuron is used as action for each neuron in policy, Layer Y
Action, reward function is MMI functions.It is pointed out that in other the present embodiment of the invention, reward function can also
It is the linear combination of more multiple objective function to increase, and realizes talking with the difference in functionality purpose of generation model with this.
In the present embodiment, decoding layer preferably includes three layers of neuronal structure, wherein, each god of input layer Layer C
The correspondence vector in above-mentioned splicing result can be received through unit.It is pointed out that in the present embodiment, input layer Layer C's is every
The vector that individual neuron is received is differed.
Layer H as decoding layer hidden layer, the input of its i-th neuron is (si-1, yi-1, ci), is output as
si.Wherein, si-1 represents the i-th -1 output of neuron in hidden layer, and yi-1 represents in output layer the defeated of the i-th -1 neuron
Go out, ci represents i-th output of neuron in input layer.
Output Y as decoding layer output layer, the input of its i-th neuron is (si, yi-1, ci), is output as
yi.The output data of the output layer dialog model result that namely entirely dialogue generation model is generated according to analysis result,
The result can be represented with (y1, y2 ... yT ').Wherein, yi is i-th word in dialog model result.It is defeated in the present embodiment
Go out layer preferably by softmax nonlinear functions to realize.Certainly, in other embodiments of the invention, output layer may be used also
Realized with other rational functions, the invention is not restricted to this.
Again as shown in figure 1, in the present embodiment, after dialog model result is obtained, the method can in step S104 root
Corresponding feedback data is generated according to the dialog model result obtained by step S103 and is exported.It is pointed out that in this hair
In bright different embodiments, according to actual needs, feedback data of the method generated in step S104 can be using difference
Appearance form (such as text, voice, limb action or image etc.), the invention is not restricted to this.
For example, for the dialogue interaction data " headache of coming to school is caught a cold " of user input, the method is in step S104
The voice feedback data of input " I is also headache " can be generated according to dialog model result resulting in step S103.
Again for example, for the dialogue interaction data " annual New Year's Day all carrys out high fever " of user input, existing data processing side
Method would generally obtain such as " annual New Year's Day all goes to a hospital to see a doctor ", and the method that the present embodiment is provided is added by coding layer
Hidden layer Layer H and/or Layer G, so that dialogue interactive system can more accurately parse above-mentioned dialogue interaction
The semanteme of data, and then such as " best wishes for a speedy recovery, healthy " can be generated.
It is pointed out that in the present embodiment, in order that interactive system must be talked with can preferably be interacted with user,
The method can also be modified to above-mentioned default dialogue generation model.Specifically, as shown in figure 1, the method can be in step
Input data of the user for above-mentioned feedback data is obtained in S105, and is entered data to default dialogue generation model according to this
It is modified.
As can be seen that compared to existing method from foregoing description, the present embodiment provided for talking with interactive system
Data processing method enable to the elongated sentence can preferably to be expressed by neutral net.Meanwhile, this method can also be solved
The feedback result for being certainly currently based on the dialogue generation model generation of the coding layer-decoding layer framework of MLE object functions is more guarded
Problem, so also allow for ultimately generating and export to the actual interaction field of feedback data more hommization and laminating of user
Scape, so as to improve the Consumer's Experience of dialogue interactive system.
Additionally, this method can also by object function it is self-defined come cause dialogue interactive system and user between
Dialogue interaction according to more have interest, diversity, interactivity and continuation, this be favorably improved dialogue interaction be
User's viscosity of system.
Present invention also offers a kind of data processing equipment for talking with interactive system, Fig. 4 is shown in the present embodiment
The structural representation of the data processing equipment.
As shown in figure 4, being preferably included for talking with the data processing equipment of interactive system of being provided of the present embodiment：Hand over
Mutual data acquisition module 401, dialog model generation module 402, feedback data output module 403 and Modifying model module 404.
Wherein, interaction data acquisition module 401 is used to obtain the dialogue interaction data of user input.It is pointed out that according to reality
Need the difference of the form of the dialogue interaction data for obtaining, interaction data acquisition module 401 using different circuit or can set
It is standby to realize, the invention is not restricted to this.For example, interaction data acquisition module 401 can be by audio collecting device (such as Mike
Wind etc.) realize, gathering corresponding interactive voice data；Interaction data acquisition module 401 can be with text collection equipment (example
Such as keyboard) realize, to gather corresponding text interaction data.
After dialogue interaction data is obtained, interaction data acquisition module 401 can be transmitted to right above-mentioned dialogue interaction data
Words model result generation module 402.Dialog model result-generation module 402 can be solved to the dialogue interaction data for receiving
Analysis, and the default dialogue generation module corresponding dialog model result of generation utilize according to analysis result.
The dialog model result that dialog model result-generation module 402 can generate itself is transmitted to feedback data and exports mould
Block 403, with by feedback data output module 403 is according to the corresponding feedback data of above-mentioned dialog model result generation and exports.
In the present embodiment, in order that interactive system must be talked with can preferably be interacted with user, data processing dress
Putting can also be modified using Modifying model module 404 to above-mentioned default dialogue generation model.Specifically, Modifying model module
404 can obtain input data of the user for above-mentioned feedback data, and be entered data to default dialogue generation model according to this
It is modified.
It is pointed out that in the present embodiment, dialog model result-generation module 402, feedback data output module 403 with
And Modifying model module 404 realizes its each concrete principle and flow of function and step S102 in above-mentioned Fig. 1 to step
Content that S105 is illustrated is similar to, thus herein no longer dialog model result-generation module 402, feedback data output module 403 with
And the related content of Modifying model module 404 is repeated.
It should be understood that disclosed embodiment of this invention is not limited to ad hoc structure disclosed herein or treatment step
Suddenly, the equivalent substitute of these features that those of ordinary skill in the related art are understood should be extended to.It should also be understood that
It is that term as used herein is only used for describing the purpose of specific embodiment, and is not intended to limit.
" one embodiment " or " embodiment " mentioned in specification means special characteristic, the structure for describing in conjunction with the embodiments
Or characteristic is included at least one embodiment of the present invention.Therefore, the phrase " reality that specification various places throughout occurs
Apply example " or " embodiment " same embodiment might not be referred both to.
Although above-mentioned example is used to illustrate principle of the present invention in one or more applications, for the technology of this area
For personnel, in the case of without departing substantially from principle of the invention and thought, hence it is evident that can in form, the details of usage and implementation
It is upper various modifications may be made and without paying creative work.Therefore, the present invention is defined by the appended claims.
Claims (10)
1. a kind of data processing method for talking with interactive system, it is characterised in that including：
Interaction data obtaining step, obtains the dialogue interaction data of user input；
Dialog model result generation step, is parsed to the dialogue interaction data, and analysis result is input into default dialogue
In generation model, dialog model result is obtained, wherein, the dialogue generation model is based on NRM and enhancing learning algorithm structure
Build；
Feedback data exports step, generates corresponding feedback data according to the dialog model result and exports.
2. the method for claim 1, it is characterised in that the default dialogue generation model includes coding layer and decoding
Layer, wherein, the coding layer be used for according to the word segmentation result that obtains of parsing generate whole vector tables of the word segmentation result up to and
Part vector table reaches, and whole vector tables is reached and part vector table is up to splicing, and obtains splicing result, the decoding
Layer according to the splicing result based on notice mechanism for generating the dialog model result.
3. method as claimed in claim 2, it is characterised in that the neutral net that the decoding layer is based on enhancing learning algorithm is entered
The generation of row dialog model result.
4. method as claimed in claim 3, it is characterised in that in the enhancing learning algorithm, using MMI functions as reward
Function.
5. the method as any one of Claims 1 to 4, it is characterised in that methods described also includes：
Modifying model step, obtains input data of the user for the feedback data, and according to the input data pair
The default dialogue generation model is modified.
6. a kind of data processing equipment for talking with interactive system, it is characterised in that including：
Interaction data acquisition module, its dialogue interaction data for being used to obtain user input；
Dialog model result-generation module, it is used to parse the dialogue interaction data, analysis result is input to pre-
If in dialogue generation model, obtaining dialog model result, wherein, the dialogue generation model is based on NRM and enhancing study
What algorithm built；
Feedback data output module, it is used to generate corresponding feedback data according to the dialog model result and export.
7. the device that such as claim 6 is stated, it is characterised in that the default dialogue generation model includes coding layer and decoding layer,
Wherein, the coding layer is configured to be reached and portion according to the whole vector tables for parsing the word segmentation result generation word segmentation result for obtaining
Point vector table reaches, and whole vector tables is reached and part vector table is up to splicing, and obtains splicing result, the decoding layer
It is configured to notice mechanism and the dialog model result is generated according to the splicing result.
8. device as claimed in claim 7, it is characterised in that the decoding layer is configured to strengthen the nerve of learning algorithm
Network carries out the generation of feedback data.
9. device as claimed in claim 8, it is characterised in that in the enhancing learning algorithm, using MMI functions as reward
Function.
10. the device as any one of claim 6~9, it is characterised in that described device also includes：
Modifying model module, it is used to obtain input data of the user for the feedback data, and according to the input
Data are modified to the default dialogue generation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611139722.0A CN106847271A (en) | 2016-12-12 | 2016-12-12 | A kind of data processing method and device for talking with interactive system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611139722.0A CN106847271A (en) | 2016-12-12 | 2016-12-12 | A kind of data processing method and device for talking with interactive system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106847271A true CN106847271A (en) | 2017-06-13 |
Family
ID=59140027
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611139722.0A Pending CN106847271A (en) | 2016-12-12 | 2016-12-12 | A kind of data processing method and device for talking with interactive system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106847271A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644641A (en) * | 2017-07-28 | 2018-01-30 | 深圳前海微众银行股份有限公司 | Session operational scenarios recognition methods, terminal and computer-readable recording medium |
| CN107766506A (en) * | 2017-10-20 | 2018-03-06 | 哈尔滨工业大学 | A kind of more wheel dialog model construction methods based on stratification notice mechanism |
| CN108520741A (en) * | 2018-04-12 | 2018-09-11 | 科大讯飞股份有限公司 | A kind of whispering voice restoration methods, device, equipment and readable storage medium storing program for executing |
| CN108897723A (en) * | 2018-06-29 | 2018-11-27 | 北京百度网讯科技有限公司 | The recognition methods of scene dialog text, device and terminal |
| CN109933773A (en) * | 2017-12-15 | 2019-06-25 | 上海擎语信息科技有限公司 | A kind of multiple semantic sentence analysis system and method |
| CN109992657A (en) * | 2019-04-03 | 2019-07-09 | 浙江大学 | A kind of interactive problem generation method based on reinforcing Dynamic Inference |
| CN110019734A (en) * | 2017-12-28 | 2019-07-16 | 顺丰科技有限公司 | A kind of intention in utterance recognition methods, device, equipment and storage medium |
| CN111625628A (en) * | 2019-02-12 | 2020-09-04 | 阿里巴巴集团控股有限公司 | Information processing method and device, storage medium and processor |
| CN111862970A (en) * | 2020-06-05 | 2020-10-30 | 珠海高凌信息科技股份有限公司 | False propaganda treatment application method and device based on intelligent voice robot |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102915731A (en) * | 2012-10-10 | 2013-02-06 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing personalized speeches |
| CN103885924A (en) * | 2013-11-21 | 2014-06-25 | 北京航空航天大学 | Field-adaptive automatic open class subtitle generating system and field-adaptive automatic open class subtitle generating method |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105447207A (en) * | 2016-01-08 | 2016-03-30 | 北京光年无限科技有限公司 | Interactive questioning and answering method and system for intelligent robot |
| CN105787560A (en) * | 2016-03-18 | 2016-07-20 | 北京光年无限科技有限公司 | Dialogue data interaction processing method and device based on recurrent neural network |
| CN106056207A (en) * | 2016-05-09 | 2016-10-26 | 武汉科技大学 | Natural language-based robot deep interacting and reasoning method and device |
| CN106095834A (en) * | 2016-06-01 | 2016-11-09 | 竹间智能科技（上海）有限公司 | Intelligent dialogue method and system based on topic |
-
2016
- 2016-12-12 CN CN201611139722.0A patent/CN106847271A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102915731A (en) * | 2012-10-10 | 2013-02-06 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing personalized speeches |
| CN103885924A (en) * | 2013-11-21 | 2014-06-25 | 北京航空航天大学 | Field-adaptive automatic open class subtitle generating system and field-adaptive automatic open class subtitle generating method |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105447207A (en) * | 2016-01-08 | 2016-03-30 | 北京光年无限科技有限公司 | Interactive questioning and answering method and system for intelligent robot |
| CN105787560A (en) * | 2016-03-18 | 2016-07-20 | 北京光年无限科技有限公司 | Dialogue data interaction processing method and device based on recurrent neural network |
| CN106056207A (en) * | 2016-05-09 | 2016-10-26 | 武汉科技大学 | Natural language-based robot deep interacting and reasoning method and device |
| CN106095834A (en) * | 2016-06-01 | 2016-11-09 | 竹间智能科技（上海）有限公司 | Intelligent dialogue method and system based on topic |
Non-Patent Citations (2)
| Title |
|---|
| JIWEI LI,ECT.: ""Deep Reinforcement Learning for Dialogue Generation"", 《IN PROCEEDINGS OF ACL》 * |
| LIFENG SHANG: ""Neural Responding Machine for Short-Text Conversation"", 《IN PROCEEDINGS OF ACL》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644641B (en) * | 2017-07-28 | 2021-04-13 | 深圳前海微众银行股份有限公司 | Dialog scene recognition method, terminal and computer-readable storage medium |
| CN107644641A (en) * | 2017-07-28 | 2018-01-30 | 深圳前海微众银行股份有限公司 | Session operational scenarios recognition methods, terminal and computer-readable recording medium |
| CN107766506A (en) * | 2017-10-20 | 2018-03-06 | 哈尔滨工业大学 | A kind of more wheel dialog model construction methods based on stratification notice mechanism |
| CN109933773A (en) * | 2017-12-15 | 2019-06-25 | 上海擎语信息科技有限公司 | A kind of multiple semantic sentence analysis system and method |
| CN109933773B (en) * | 2017-12-15 | 2023-05-26 | 上海擎语信息科技有限公司 | Multiple semantic statement analysis system and method |
| CN110019734A (en) * | 2017-12-28 | 2019-07-16 | 顺丰科技有限公司 | A kind of intention in utterance recognition methods, device, equipment and storage medium |
| CN108520741A (en) * | 2018-04-12 | 2018-09-11 | 科大讯飞股份有限公司 | A kind of whispering voice restoration methods, device, equipment and readable storage medium storing program for executing |
| US11508366B2 (en) | 2018-04-12 | 2022-11-22 | Iflytek Co., Ltd. | Whispering voice recovery method, apparatus and device, and readable storage medium |
| CN108897723A (en) * | 2018-06-29 | 2018-11-27 | 北京百度网讯科技有限公司 | The recognition methods of scene dialog text, device and terminal |
| CN111625628A (en) * | 2019-02-12 | 2020-09-04 | 阿里巴巴集团控股有限公司 | Information processing method and device, storage medium and processor |
| CN111625628B (en) * | 2019-02-12 | 2023-05-02 | 阿里巴巴集团控股有限公司 | Information processing method and device, storage medium and processor |
| CN109992657B (en) * | 2019-04-03 | 2021-03-30 | 浙江大学 | Dialogue type problem generation method based on enhanced dynamic reasoning |
| CN109992657A (en) * | 2019-04-03 | 2019-07-09 | 浙江大学 | A kind of interactive problem generation method based on reinforcing Dynamic Inference |
| CN111862970A (en) * | 2020-06-05 | 2020-10-30 | 珠海高凌信息科技股份有限公司 | False propaganda treatment application method and device based on intelligent voice robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106847271A (en) | A kind of data processing method and device for talking with interactive system | |
| CN112183747B (en) | Neural network training method, neural network compression method and related equipment | |
| CN111444709B (en) | Text classification method, device, storage medium and equipment | |
| Wang et al. | TETFN: A text enhanced transformer fusion network for multimodal sentiment analysis | |
| CN109036465B (en) | Speech emotion recognition method | |
| CN108009285B (en) | Forest Ecology man-machine interaction method based on natural language processing | |
| CN112101045B (en) | Multi-mode semantic integrity recognition method and device and electronic equipment | |
| CN112131366A (en) | Method, device and storage medium for training text classification model and text classification | |
| CN107301168A (en) | Intelligent robot and its mood exchange method, system | |
| CN107578106A (en) | A kind of neutral net natural language inference method for merging semanteme of word knowledge | |
| CN106844442A (en) | Multi-modal Recognition with Recurrent Neural Network Image Description Methods based on FCN feature extractions | |
| CN105843382B (en) | A kind of man-machine interaction method and device | |
| CN108133705A (en) | Speech recognition and phonetic synthesis model training method based on paired-associate learning | |
| CN106844587B (en) | It is a kind of for talking with the data processing method and device of interactive system | |
| CN106897254A (en) | A kind of network representation learning method | |
| CN110457661A (en) | Spatial term method, apparatus, equipment and storage medium | |
| Zhou et al. | ICRC-HIT: A deep learning based comment sequence labeling system for answer selection challenge | |
| CN106934458A (en) | Multilayer automatic coding and system based on deep learning | |
| CN115393933A (en) | Video face emotion recognition method based on frame attention mechanism | |
| CN112115242A (en) | Intelligent customer service question-answering system based on naive Bayes classification algorithm | |
| Xiong et al. | DGI: recognition of textual entailment via dynamic gate matching | |
| CN107862058A (en) | Method and apparatus for generating information | |
| Mai et al. | A unimodal representation learning and recurrent decomposition fusion structure for utterance-level multimodal embedding learning | |
| CN115659987A (en) | Multi-mode named entity recognition method, device and equipment based on double channels | |
| Zhang et al. | Adapting to context-aware knowledge in natural conversation for multi-turn response selection |
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: 20170613 |