CN108804536A - Human-computer dialogue and strategy-generating method, equipment, system and storage medium - Google Patents
Human-computer dialogue and strategy-generating method, equipment, system and storage medium Download PDFInfo
- Publication number
- CN108804536A CN108804536A CN201810421418.8A CN201810421418A CN108804536A CN 108804536 A CN108804536 A CN 108804536A CN 201810421418 A CN201810421418 A CN 201810421418A CN 108804536 A CN108804536 A CN 108804536A
- Authority
- CN
- China
- Prior art keywords
- slot
- dialogue
- value
- human
- state machine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Machine Translation (AREA)
Abstract
A kind of human-computer dialogue of the embodiment of the present application offer and strategy-generating method, equipment, system and storage medium.In the embodiment of the present application, slot is filled and is combined with finite state machine, multigroup slot-value pair in session operational scenarios with dialogue meaning and multigroup slot-value are generated in the form of slot is filled first to corresponding dialogue state, it is then based on multigroup slot-value pair and corresponding dialogue state builds finite state machine model, in this process, it is realized using slot filling flexible, the advantages such as simple, the realization of dialogue state in session operational scenarios can be simplified, and then the structure difficulty of finite state machine model can be reduced, so that the management that may finally be engaged in the dialogue in the form of finite state machine, be conducive to give full play to the advantage of finite state machine in various session operational scenarios, it realizes simpler, neatly dialogue management.
Description
Technical field
This application involves field of artificial intelligence more particularly to a kind of human-computer dialogue and strategy-generating method, equipment, it is
System and storage medium.
Background technology
With the development of artificial intelligence, there is interactive system, interactive system, which is one kind, to be carried out with people
The computer system of coherent dialogue.Interactive system includes mainly five funtion parts:Speech recognition, language understanding, dialogue
Management, language generation and phonetic synthesis.Dialogue management is the Core Feature of interactive system, and which control user and systems
Entire dialog procedure, decides the everything of system, thus the design degree of perfection of dialogue management be related to it is entire man-machine right
The performance of telephone system.
In the prior art, relatively simple interactive system generally uses finite state machine to realize dialogue management, i.e.,
The dialogue state in session operational scenarios and the row such as the transfer between these dialogue states and action are indicated by finite state machine
For.In conjunction with the characteristics of finite state machine, using finite state machine realize dialogue management when, can flexible expansion dialogue state, but
Be built with the increase of conversation tasks complexity finite state machine difficulty it is bigger, this causes finite state machine in complexity
Application in conversation tasks is relatively fewer.
Invention content
A kind of human-computer dialogue of many aspects offer and strategy-generating method, equipment, system and the storage medium of the application, is used
To reduce realization difficulty of the finite state machine in session operational scenarios, utilization rate of the finite state machine in session operational scenarios is improved.
The embodiment of the present application provides a kind of Dialogue management strategy generation method, including:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize limited shape
The form of state machine engages in the dialogue management to the human-computer dialogue process in the session operational scenarios.
The embodiment of the present application also provides a kind of interactive method, including:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue
In data obtain can trigger finite state machine engage in the dialogue state transfer input information;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
The embodiment of the present application also provides a kind of man-machine dialogue equipment, including:Memory and processor;
Memory, for storing computer program;
The processor, for executing the computer program, for:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize limited shape
The form of state machine engages in the dialogue management to the human-computer dialogue process in the session operational scenarios.
The embodiment of the present application also provides a kind of computer readable storage medium of storage computer instruction, when the computer
When instruction is executed by one or more processors, it includes action below to cause one or more of processor execution:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize limited shape
The form of state machine engages in the dialogue management to the human-computer dialogue process in the session operational scenarios.
The embodiment of the present application also provides a kind of man-machine dialogue equipment, including:Memory and processor;
The memory, for storing computer program;
The processor, for executing the computer program, for:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue
In data obtain can trigger finite state machine engage in the dialogue state transfer input information;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
The embodiment of the present application also provides a kind of computer readable storage medium of storage computer instruction, which is characterized in that
When the computer instruction is executed by one or more processors, it includes following to cause one or more of processor execution
Action:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue
In data obtain can trigger finite state machine engage in the dialogue state transfer input information;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
The embodiment of the present application also provides a kind of interactive system, including:Server and terminal device;
The terminal device, for receiving human-computer dialogue data input by user in session operational scenarios, by the human-computer dialogue
Data are sent to the server, and receive the corresponding reply data of the human-computer dialogue data of the server return simultaneously
It exports to the user;
The server, the human-computer dialogue data sent for receiving the terminal device, according in the session operational scenarios
Each slot-value fills out slot language material and cancellation slot language material to corresponding, is obtained from the human-computer dialogue data and can trigger finite state machine
The input information of the state that engages in the dialogue transfer;The finite state machine is controlled according to the input information to jump from current dialogue states
Go to Next dialog states;According to the related data of the Next dialog states, it is described man-machine right to be returned to the terminal device
Talk about the reply data of data.
In the embodiment of the present application, slot is filled and is combined with finite state machine, first the generation pair in the form of slot is filled
Multigroup slot-value pair in scene with dialogue meaning and multigroup slot-value are talked about to corresponding dialogue state, is then based on multigroup
Slot-value pair and corresponding dialogue state build finite state machine model, in this process, are realized using slot filling flexible, simple
Etc. advantages, the realization of dialogue state in session operational scenarios can be simplified, and then the structure difficulty of finite state machine model can be reduced so that
May finally be engaged in the dialogue management in the form of finite state machine, be conducive to give full play to finite state in various session operational scenarios
The advantage of machine realizes simpler, neatly dialogue management.
Description of the drawings
Attached drawing described herein is used for providing further understanding of the present application, constitutes part of this application, this Shen
Illustrative embodiments and their description please do not constitute the improper restriction to the application for explaining the application.In the accompanying drawings:
Fig. 1 is a kind of structural schematic diagram for interactive system that one exemplary embodiment of the application provides;
Fig. 2 is a kind of state diagram for finite state machine that one exemplary embodiment of the application provides;
Fig. 3 is a kind of flow signal for Dialogue management strategy generation method that the application another exemplary embodiment provides
Figure;
Fig. 4 is a kind of flow diagram for interactive method that the application another exemplary embodiment provides;
Fig. 5 a are that a kind of robot of accompanying and attending to of the corresponding family of the application scenarios 1 that provide of the application another exemplary embodiment is chatted
The structural schematic diagram of its system;
Fig. 5 b are the human-computer dialogue processing procedure suitable for each application scenarios that the application another exemplary embodiment provides
Rough schematic view;
Fig. 5 c are that the corresponding another family of application scenarios 1 that the application another exemplary embodiment provides accompanies and attends to robot
The structural schematic diagram of chat system;
Fig. 5 d are the corresponding a kind of bank self-help operation system of application scenarios 2 that the application another exemplary embodiment provides
Structural schematic diagram;
Fig. 5 e are the corresponding a kind of knot of network seat reservation system of application scenarios 3 that the application another exemplary embodiment provides
Structure schematic diagram;
Fig. 6 a are a kind of structural representation for Dialogue management strategy generating means that the application another exemplary embodiment provides
Figure;
Fig. 6 b are a kind of structural schematic diagram for man-machine dialogue equipment that the application another exemplary embodiment provides;
Fig. 7 a are a kind of structural schematic diagram for human-computer dialogue device that the application another exemplary embodiment provides;
Fig. 7 b are the structural schematic diagram for another man-machine dialogue equipment that the application another exemplary embodiment provides.
Specific implementation mode
To keep the purpose, technical scheme and advantage of the application clearer, below in conjunction with the application specific embodiment and
Technical scheme is clearly and completely described in corresponding attached drawing.Obviously, described embodiment is only the application one
Section Example, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not doing
Go out the every other embodiment obtained under the premise of creative work, shall fall in the protection scope of this application.
For it is existing in session operational scenarios using finite state machine engage in the dialogue management problem faced, the application some
In embodiment, slot is filled and is combined with finite state machine, generated in the form of slot is filled has dialogue first in session operational scenarios
Multigroup slot-the value pair and multigroup slot-value of meaning are then based on multigroup slot-value pair and corresponding right to corresponding dialogue state
Speech phase builds finite state machine model, in this process, realizes the advantages such as flexible, simple using slot filling, can simplify dialogue
The realization of dialogue state in scene, and then the structure difficulty of finite state machine model can be reduced so that it may finally be with limited shape
The form of state machine engages in the dialogue management, is conducive to the advantage that finite state machine is given full play in various session operational scenarios, realizes more
Add simple, neatly dialogue management.
Below in conjunction with attached drawing, the technical solution that each embodiment of the application provides is described in detail.
Fig. 1 is a kind of structural schematic diagram for interactive system that one exemplary embodiment of the application provides.Such as Fig. 1 institutes
Show, which includes:Server 10a and terminal device 10b.The server 10a and terminal presented in Fig. 1 is set
Standby 10b is exemplary illustration, is not limited the way of realization of the two.
In the present embodiment, it between server 10a and terminal device 10b can be wired or wireless network connection.It is optional
Ground, server 10a can be by mobile networks and terminal device 10b communication connections, and correspondingly, the network formats of mobile network can
Think 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+),
Any one in WiMax etc..Optionally, server 10a can also by modes such as bluetooth, WiFi, infrared ray, internets and
Terminal device 10b communication connections.
In the present embodiment, server 10a is mainly responsible for speech recognition, language understanding, dialogue in man-machine dialog procedure
The functions such as management, language generation, phonetic synthesis, and terminal device 10b is coordinated to realize human-computer dialogue.Server 10a can be one
Platform can also be Duo Tai.The way of realization of the present embodiment not Limited service device 10a.For example, in some optional embodiments
In, server 10a can be the server apparatus such as General Server, Cloud Server, cloud host, virtual center.Wherein, server
The composition of 10a equipment includes mainly that processor, hard disk, memory, system bus etc. are similar with general computer architecture.
In the present embodiment, terminal device 10b refers to user oriented, and the electronics that interactive voice can be carried out with user is set
It is standby.In some optional embodiments, terminal device 10b can be the intelligent hand for being equipped with various interactive voice class application software
Machine, tablet computer, PC, wearable device, intelligent sound etc..In other alternative embodiments, terminal device 10b can be with
It is self-help registration/payment machine of various types of voice interactive self-service terminal, self-service machine, such as hospital, the self-help drawing money of bank
Machine, the automatic ticket taking machine etc. in the scenes such as subway, station or airport.In other application scenarios, terminal device 10b can be
Some support interactive voices intelligence machine, such as can be support interactive voice family accompany and attend to class robot, chat machine
People, sweeping robot, navigation/follow robot, offer order the robot etc. of service.
No matter the physical aspect of terminal device 10b, in general, terminal device 10b generally include at least one place
Manage unit and at least one processor.Configuration and type of the quantity of processing unit and memory depending on terminal device 10b.It deposits
Reservoir may include volatibility, such as RAM, can also include non-volatile, such as read-only memory (Read-Only
Memory, ROM), flash memory etc., or can also include two kinds of simultaneously.Operating system is typically stored in memory
(Operating System, OS), one or more application software, such as interactive voice class software, can also have program stored therein
Data etc..Other than processing unit and memory, some terminal devices 10b also will include network card chip, IO buses, audio and video
The basic configuration such as component.Optionally, according to the way of realization of terminal device 10b, terminal device 10b can also include some peripheries
Equipment, such as keyboard, mouse, input pen, printer etc..These peripheral equipments are well known in the art, herein not
It repeats.
In the present embodiment, server 10a and terminal device 10b can be deployed in various session operational scenarios, be responsible for completing phase
Answer the human-computer dialogue process in scene.For example, server 10a and terminal device 10b can be deployed in hospital's scene, it is responsible for realization
Human-computer dialogue process during voice self-help registration.In another example server 10a and terminal device 10b can be deployed to station,
In the scenes such as iron or airport, it is responsible for realizing the human-computer dialogue process during the self-service ticket booking of voice.In another example server 10a and
Terminal device 10b can be deployed in bank's scene, be responsible for realizing the human-computer dialogue process during voice self-help drawing money.
Either in which kind of session operational scenarios, based on server 10a and terminal device 10b realize it is interactive substantially
Process is as follows:
User may be used natural language and be interacted with terminal device 10b, to express the need of oneself to terminal device 10b
It asks or is intended to.For example, user can to terminal device 10b input " I to withdraw the money 2000 ", " I will order the train ticket in Shanghai ",
Man-machine dialogue datas such as " I will hang number of paediatrics doctor Wang ".Wherein, human-computer dialogue data can be user with natural language side
The voice data of formula input, can also be the non-speech datas such as text data.Terminal device 10b receives user in the session operational scenarios
The man-machine dialogue data is sent to server 10a by the human-computer dialogue data of input.Server 10a receiving terminal apparatus 10b hairs
The human-computer dialogue data sent identify the corresponding user view of the man-machine dialogue data, provide response corresponding with the user view
Data, and the reply data is returned into terminal device 10b.Terminal device 10b receives the reply data that server 10a is returned,
Reply data is exported to user, a wheel human-computer dialogue process is so far completed.
Optionally, if above-mentioned human-computer dialogue data are voice data, server 10a can specifically be directed to the human-computer dialogue
Data carry out a series of processing such as speech recognition, language understanding, dialogue management, language generation and phonetic synthesis, are finally somebody's turn to do
The corresponding reply data of human-computer dialogue data.Wherein, speech recognition (ASR) refers to converting primary voice data input by user
For the process of text data.Language understanding refers to that will identify that the text data come is converted to the semantic expressiveness that machine is appreciated that
Process.Dialogue management refers to that anything should be taken to act based on dialogue state judgement, provides the process of which kind of answer data, letter
It is single to understand, it is exactly that server 10a needs to determine what meaning oneself should be expressed in the semantic expressiveness gone out from language understanding.Language
Generation refers to the process of needing the meaning expressed to be transformed into text data server 10a.Phonetic synthesis refers to by text data
Be converted to the process of voice data.
It is worth noting that if human-computer dialogue data input by user are text datas, without carrying out speech recognition.It can
Selection of land can not also carry out phonetic synthesis.That is, speech recognition and phonetic synthesis are two in human-computer dialogue processing procedure
It is a can selection operation.
In the operations such as above-mentioned speech recognition, language understanding, dialogue management, language generation and phonetic synthesis, dialogue management
It is the Core Feature of interactive system, which control the entire dialog procedures of user and interactive system 10, decide people
The everything of machine conversational system 10, the design degree of perfection of dialogue management are related to the performance of entire interactive system 10.
Therefore in the present embodiment, the realization process of dialogue management is paid close attention to.About speech recognition, language understanding, language generation and language
Sound synthesizes, and server 10a may be used various technologies and realize, the present embodiment limits not to this.
In the present embodiment, server 10a realizes dialogue management using finite state machine, i.e., by finite state machine come table
Show the dialogue state in session operational scenarios and manages the behaviors such as transfer and action between these dialogue states.Pair in the present embodiment
Talking about scene may be fairly simple, it is also possible to more complicated.If session operational scenarios are more complicated, such as dialog turns are relatively more, again
Or dialogue state is relatively more, then can be bigger using the realization difficulty of existing way structure finite state machine, limitation is limited
The use of state machine.
In order to solve the problems, such as that finite state machine realizes that difficulty is larger, in the present embodiment, in the structure of finite state machine
Engagement groove is filled in the process, is filled using slot and is generated multigroup slot-value pair in session operational scenarios with dialogue meaning and multigroup slot-
Value realizes the advantages such as flexible, simple to corresponding dialogue state, using slot filling, simplifies the reality of dialogue state in session operational scenarios
It is existing, and then the structure difficulty of finite state machine model can be reduced.Wherein, the finite state machine building process that engagement groove is filled is such as
Under:
First, it based on the semantic understanding to session operational scenarios, is determined suitable for the multiple of session operational scenarios in the form of slot is filled
Semantic slot (slot) and the corresponding candidate slot value (value) of multiple semantic slots.Semantic slot refers to that resolve to text data can quilt
The expression way for the semantic expressiveness that machine understands.Candidate slot value refers to the possible value of semantic slot, and each semanteme slot may correspond to
Multiple and different candidate slot values.
Wherein, according to the difference of session operational scenarios, semantic slot candidate slot value corresponding with semantic slot would also vary from.Example
Such as, by taking " ordering flight " scene as an example, semantic slot may include " city of setting out ", " departure time ", " purpose city " etc., Yi Jiyu
The corresponding candidate slot value of adopted slot " city of setting out " may include " Beijing ", " Shanghai " etc., the corresponding time of semantic slot " departure time "
It may include " at 8 points in the morning ", " at 2 points in afternoon " etc. to select slot value, and the corresponding candidate slot value of semantic slot " purpose city " may include
" Harbin ", " Wuhan ", " Shenzhen " etc..In another example by taking " withdrawal " scene as an example, semantic slot may include " withdrawal ", " amount of money ",
" medium " etc., and the corresponding candidate slot value of semantic slot " withdrawal " may include " null ", " confirmation ", " cancellation " etc., semantic slot
" amount of money " corresponding candidate slot value may include " 20,000 or less ", " 20,000 or more five ten thousand or less ", " 50,000 or more " etc., semantic slot
" medium " corresponding candidate slot value may include " bank card ", " bankbook " etc..
Since the corresponding candidate slot value of each semanteme slot may be multiple, by the corresponding candidate slot value progress of multiple semanteme slots
Combination can obtain multigroup slot-value pair with dialogue meaning.Wherein, a corresponding candidate slot value of a semantic slot
A slot-value can be formed to (slot-value pair).Every group includes the corresponding slot-value pair of multiple semantic slots, and
The candidate slot value of slot-value centering in different groups is not exactly the same.For the ease of understanding the concept of each group slot-value centering " group ",
By in " withdrawal " scene semantic slot and candidate slot value for illustrate.Assuming that the semantic slot in " withdrawal " scene includes " taking
Money ", " amount of money ", " medium ", the corresponding candidate slot value of semantic slot " withdrawal " includes " null " and " confirmation ", and semantic slot " amount of money " is right
The candidate slot value answered includes " 20,000 or less " and " 20,000 or more ", and the corresponding candidate slot value of semantic slot " medium " may include " silver
Row card " and " bankbook " is then combined the corresponding candidate slot values of these semantic slots, can be obtained it is as shown in table 1 below with pair
Talk about multigroup slot-value pair of meaning.
Table 1
In table 1 above, " withdrawal ", " amount of money ", the corresponding one group of slot-value pair of expression per a line of the row of " medium " three.In conjunction with
Table 1 is it is found that the candidate slot value in different groups is not exactly the same.Moreover, from table 1 it follows that the language of each group of slot-value centering
The candidate Cao Zhi of adopted slot and semantic slot combines and can indicate specific semanteme, and semantic also different expressed by different groups.
Based on above-mentioned, semanteme that can be according to multigroup slot-value to respectively indicating generates every group of slot-value to corresponding dialogue shape
State, to obtain multiple dialogue states, and multiple dialogue states and multigroup slot-value have correspondence between.Such as 1 institute of table
Show, last row indicates each group slot-value to corresponding dialogue state.As shown in Table 1, namely its corresponding one group of dialogue state
The form of expression of the slot-value to the semanteme embodied.
Multiple dialogue states and multigroup slot-value in obtaining session operational scenarios, can be according to multiple dialogue states to later
With multigroup slot-value to building finite state machine model, to utilize the form of finite state machine to man-machine right in session operational scenarios
Words process engages in the dialogue management.Here finite state machine model is mainly used for describing multiple dialogue states and multiple dialogue shapes
The information such as transfer, jump condition, corresponding action between state.Optionally, finite state machine model can be that a kind of static state is retouched
Document is stated, such as can be the configuration file of finite state machine, but not limited to this.Certainly, which can also
Take other form way of realization.
In the present embodiment, the dialogue state for needing to manage is generated for finite state machine in the way of slot filling, utilized
The advantages such as flexible, simple are realized in slot filling, simplify the realization of dialogue state in session operational scenarios, and then can reduce finite state machine
The structure difficulty of model is realized simpler, neatly convenient for giving full play to the advantage of finite state machine in various words scenes
Dialogue management.
In some optional embodiments, multiple dialogue states are mapped as in finite state machine model by server 10a
Multiple state nodes;Two-way side is added between any two state node in multiple state nodes;And according to any two
Difference of the corresponding two groups of slot-values of state node between, generate any two state node between shift when jump condition,
To build finite state machine model.
It is alternatively possible to intuitively indicate the corresponding finite state machine of finite state machine model by state diagram.Then with table 1
For shown " withdrawal " scene, a kind of corresponding state diagram of finite state machine model is as shown in Figure 2.In fig. 2, with dialogue state
It is transferred to by " 20,000 or less withdrawal " for " 20,000 or less bank card withdrawal ", then " is withdrawing the money 20,000 to withdraw the money down toward (to) bank card
20000 or less " under conditions of, and meet the transfer for the state that can engage in the dialogue when " medium " is " bank card ".
Further, when needing to extend dialogue state, new dialogue state can be generated and the dialogue state is corresponding
One group of slot-value pair.Then, add new state node in finite state machine model, and new state node with it is each
It adds two-way side between stateful node, and the corresponding two groups of slot-values of state node is had with each according to new state node
Difference between generates jump condition when being shifted between new state node and each existing state node, to realize
The extension of dialogue state.
In conjunction with Fig. 2, under " 20,000 or less bank card withdrawal " this dialogue state, if server 10a is directed to last round of dialogue
The reply data provided is " please arrive ATM self-help drawing moneys ", " sales counter can be gone to handle at this point, user says?", this be in Fig. 2 not
Existing dialogue state.If engaged in the dialogue management using slot filling mode, such case be difficult re-define some semantic slot, and
The embodiment of the present application is engaged in the dialogue management using finite state machine, and can increasing by one in finite state machine model, " sales counter is done
The dialogue state of reason ", and adaptively define trigger condition and corresponding language material.It can be seen that being based on finite state machine
Dialogue state can be neatly extended, and also need to only update finite state machine model when in use.
Further, after obtaining finite state machine model, it is based on the finite state machine model, server 10a can profit
It is engaged in the dialogue management to each human-computer dialogue process in session operational scenarios with the form of finite state machine.Server 10a has in utilization
The form of limit state machine engages in the dialogue to the human-computer dialogue process in session operational scenarios before management, needs to during human-computer dialogue
Human-computer dialogue data carry out language understanding, be converted into can be by the semantic expressiveness of machine recognition.Language understanding commonly relies on
Corpus in session operational scenarios.
Can language understanding correctly be carried out to human-computer dialogue data for the ease of server 10a, the present embodiment is further
Engagement groove filling forms form the corpus in session operational scenarios, in order to human-computer dialogue data that human-computer dialogue process is summarized into
Row language understanding provides required input information for finite state machine.
In the present embodiment, corpus is built to corresponding language material based on each slot-value in session operational scenarios.In the present embodiment
In, it not only to obtain each slot-value and fill out slot language material to corresponding, it is also necessary to cancel slot language material to increasing for each slot-value.These slots-
Value can form the corpus of the session operational scenarios to the corresponding slot language material and cancellation slot language material filled out.Wherein, it is to meet slot-to fill out slot language material
Value has the language material of affirmative meaning to what is required;And cancel slot language material and do not meet slot-value to requiring, there is Negation
Language material.Traditional slot padding scheme only fills out slot language material, does not have and cancels slot language material, in the present embodiment, increases and cancels slot
Language material, can make between each dialogue state can mutual phase transfer, to form a finite state machine model connected entirely.
By taking " withdrawal " scene described in table 1 as an example, each slot-value pair and each slot-value fill out slot to corresponding in the session operational scenarios
Language material and cancellation slot language material, as shown in table 2 below:
Table 2
Only it is that exemplary give fills out slot language material and cancel slot language material, it will be appreciated by those skilled in the art that filling out in table 2
Slot language material and cancellation slot language material are not limited to shown in table 2.
In the present embodiment, from the language material in the dimension management of dialogs scene of slot-value pair, rather than from each dialogue state it
Between transfer relationship carry out the language material in management of dialogs scene, the management dimension of language material is relatively rarely very much, simpler in management to answer, easily
In realization, be conducive to be further simplified dialogue management, reduce the cost of implementation of dialogue management.
On the basis of above-mentioned corpus and finite state machine model, server 10a and terminal device 10b are matched can be with
It is engaged in the dialogue management to human-computer dialogue process using the form of finite state machine.Human-computer dialogue process based on finite state machine is such as
Under:
User inputs human-computer dialogue data to terminal device 10b.Terminal device 10b receives human-computer dialogue number input by user
According to the man-machine dialogue data is sent to server 10a.
The human-computer dialogue data that server 10a receiving terminal apparatus 10b is sent.If the man-machine dialogue data is voice number
According to the man-machine dialogue data is converted to text type by server 10a by speech recognition technology by sound-type, is then based on
Corpus in session operational scenarios, i.e., in the session operational scenarios each slot-value to it is corresponding fill out slot language material and cancel slot language material to the text
The human-computer dialogue data of type carry out language understanding, therefrom obtain and can trigger finite state machine and engage in the dialogue the input of state transfer
Information.If the man-machine dialogue data is text data, server 10a can be directly based upon the corpus in session operational scenarios, i.e., this is right
Each slot-value carries out language understanding to corresponding slot language material and the cancellation slot language material filled out to the man-machine dialogue data in words scene, therefrom
The input information that the state that the triggerable finite state machine of acquisition engages in the dialogue shifts.
For example, filling out slot language material shown in the state diagram in conjunction with shown in Fig. 2 and table 2 and cancelling slot language material, it is assumed that in " withdrawal 20,000
Below " under this dialogue state, if user says sentences such as " 300/1,000/... ", server 10a can recognize " medium " and be
" bank card ", to obtain triggerable finite state machine, from " 20,000 or less withdrawal ", this dialogue state is transferred to " bank card withdrawal
20000 or less " input information of this dialogue state.Assume again under " 20,000 or less withdrawal " this dialogue state, if user says
When sentences such as " amount of money wrong/wrong/... ", server 10a can recognize " amount of money " and be reset, to obtain can trigger it is limited
State machine is transferred to the input information of " withdrawal " this dialogue state from " 20,000 or less withdrawal " this dialogue state.Wherein, language
The realization of the methods of keyword, regular expression, grader may be used in understanding process.
In some optional embodiments, for the ease of more quickly and conveniently right based on the corpus in session operational scenarios
Human-computer dialogue data carry out language understanding.Previously according to each slot-value in corpus slot language material and cancellation slot can be filled out to corresponding
Language material training language understands model, which has for being obtained from the human-computer dialogue data during human-computer dialogue
Limit the input information needed for state machine.
In one embodiment, slot language material and cancellation slot language material can be filled out to corresponding according to each slot-value in corpus
Training first language understands model.First language understands that model for extracting slot-value pair that human-computer dialogue data include, is extracted
Slot-value to can be used as above-mentioned input information.Correspondingly, server 10a can be with foundation human-computer dialogue data run first language
Model is understood, to obtain slot-value that human-computer dialogue data include to as the input information needed for finite state machine.In the reality
It applies in mode, finite state machine needs input information being converted to identifiable jump condition.
In another embodiment, slot language material and cancellation slot language can be filled out to corresponding according to each slot-value in corpus
Material and the correspondence training second language between multigroup slot-value pair and multiple dialogue states understand model.Second language is managed
Solution model from human-computer dialogue data for obtaining a jump condition as the input information needed for finite state machine.For example, can
According to the correspondence between multigroup slot-value pair and multiple dialogue states, to establish each slot-value and fill out slot language material to corresponding and take
Correspondence between the jump condition to disappear in slot language material and finite state machine carries out model training according to the correspondence, this
Sample can obtain the second language understanding that the jump condition needed for finite state machine can be obtained directly from human-computer dialogue data
Model.Correspondingly, server 10a can understand model according to human-computer dialogue data run second language, to finite state machine
In jump condition as the input information needed for finite state machine.In the optional embodiment, without being done to input information
Conversion, finite state machine can Direct Recognition.
After obtaining input information, server 10a can control finite state machine from current session according to the input information
State transition determines that human-computer dialogue data are corresponding to Next dialog states, and then according to the related data of the Next dialog states
Reply data, and the reply data is sent to terminal device 10b.
Optionally, the related data of Next dialog states may include the state description, corresponding dynamic of Next dialog states
The data such as work.These related datas can express which kind of response server 10a will carry out, and being then based on these related datas can be with
Determine reply data corresponding with the man-machine dialogue data.Optionally, server 10a can obtain answer number from corpus
According to, or reply data can be automatically generated.
Terminal device 10b can receive the reply data that server 10a is returned, which is exported to user.It is optional
Reply data can be played to user by ground, terminal device 10b by voice mode, alternatively, can also be by showing that screen will
Reply data is shown to user.
In the present embodiment, slot is filled and is combined with finite state machine, reduce the structure difficulty of finite state machine model,
Allow to the management that engages in the dialogue in the form of finite state machine, is conducive to give full play to finite state in various session operational scenarios
The advantage of machine realizes simpler, neatly dialogue management.
The embodiment of the present application also provides certain methods embodiment other than providing above-mentioned interactive system.These sides
Method embodiment respectively describes the generating process of finite state machine model and the human-computer dialogue process based on finite state machine.
Fig. 3 is a kind of flow signal for Dialogue management strategy generation method that the application another exemplary embodiment provides
Figure.As shown in figure 3, this method includes:
301, it based on the semantic understanding to session operational scenarios, determines suitable for multiple semantic slots of the session operational scenarios and described
The corresponding candidate slot value of multiple semanteme slots.
302, the corresponding candidate slot value of the multiple semanteme slot is combined, there is the multigroup of dialogue meaning to obtain
Slot-value pair, every group includes the corresponding slot-value pair of the multiple semantic slot.
303, the semanteme respectively indicated according to multigroup slot-value is generated with multigroup slot-value to corresponding multiple right
Speech phase.
304, had to utilize to building finite state machine model according to the multiple dialogue state and multigroup slot-value
The form for limiting state machine engages in the dialogue management to the human-computer dialogue process in the session operational scenarios.
About the detailed description of step 301-304, reference can be made to the description in above system embodiment.
In the present embodiment, engagement groove is filled in the building process of finite state machine, is filled using slot and is generated dialogue field
Have the multigroup slot-value pair for talking with meaning and multigroup slot-value to corresponding dialogue state in scape, utilizes slot to fill and realize spirit
The advantages such as living, simple, simplify the realization of dialogue state in session operational scenarios, and then the structure that can reduce finite state machine model is difficult
Degree realizes simpler, neatly dialogue management convenient for giving full play to the advantage of finite state machine in various words scenes.
In some optional embodiments, a kind of embodiment of above-mentioned steps 304 includes:Multiple dialogue states are mapped
For multiple state nodes in finite state machine model;It is added between any two state node in multiple state nodes two-way
Side;And the difference according to the corresponding two groups of slot-values of any two state node between, generate any two state node it
Between shift when jump condition, to build finite state machine model.
In some optional embodiments, after building finite state machine model, dialogue state is if desired extended, then may be used
To generate new dialogue state and the corresponding one group of slot-value pair of the dialogue state;Then, it is added in finite state machine model
New state node, and two-way side is added between new state node and each existing state node, and according to new state
Node and each difference for having the corresponding two groups of slot-values of state node between, generate new state node and have with each
Jump condition when being shifted between state node, to realize the extension of dialogue state.It can be seen that can based on finite state machine
Neatly to extend dialogue state, and finite state machine model also need to be only updated when in use.
Further, after obtaining finite state machine model, it is based on the finite state machine model, finite state can be utilized
The form of machine engages in the dialogue management to each human-computer dialogue process in session operational scenarios.In the form using finite state machine to dialogue
Human-computer dialogue process in scene engages in the dialogue before management, needs to carry out language to the human-computer dialogue data during human-computer dialogue
Speech understands, be converted into can be by the semantic expressiveness of machine recognition.Language understanding commonly relies on the corpus in session operational scenarios.Base
In this, each slot-value in session operational scenarios can also be obtained and fill out slot language material and cancellation slot language material to corresponding, to form corpus;Root
Model is understood to corresponding slot language material and the cancellation slot language material training language filled out according to each slot-value in corpus, the language understanding model
For obtaining the input information needed for finite state machine from human-computer dialogue data.
In one embodiment, slot language material and cancellation slot language material can be filled out to corresponding according to each slot-value in corpus
Training first language understands model.First language understands that model for extracting slot-value pair that human-computer dialogue data include, is extracted
Slot-value to can be used as above-mentioned input information.In this embodiment, finite state machine needs to be converted to input information and can know
Other jump condition.
In another embodiment, slot language material and cancellation slot language can be filled out to corresponding according to each slot-value in corpus
Material and the correspondence training second language between multigroup slot-value pair and multiple dialogue states understand model.Second language is managed
Solution model from human-computer dialogue data for obtaining a jump condition as the input information needed for finite state machine.For example, can
According to the correspondence between multigroup slot-value pair and multiple dialogue states, to establish each slot-value and fill out slot language material to corresponding and take
Correspondence between the jump condition to disappear in slot language material and finite state machine carries out model training according to the correspondence, this
Sample can obtain the second language understanding that the jump condition needed for finite state machine can be obtained directly from human-computer dialogue data
Model.
Optionally, after constructing finite state machine model using method shown in Fig. 3, method shown in Fig. 4 may be used
Human-computer dialogue is carried out based on finite state machine.It should be noted that finite state machine during human-computer dialogue shown in Fig. 4 can be with
It is built using mode shown in Fig. 3, but is not limited to mode shown in Fig. 3.
Fig. 4 is a kind of flow diagram for interactive method that the application another exemplary embodiment provides.Such as Fig. 4 institutes
Show, this method includes:
401, the human-computer dialogue data in session operational scenarios are obtained.
402, slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in session operational scenarios, from human-computer dialogue data
The input information that the state that the middle triggerable finite state machine of acquisition engages in the dialogue shifts.
403, finite state machine is controlled according to input information and jumps to Next dialog states from current dialogue states.
404, according to the related data of Next dialog states, the reply data of human-computer dialogue data is exported.
In an optional embodiment, the embodiment of above-mentioned steps 402 includes:
Model is understood according to human-computer dialogue data run first language, to obtain slot-value pair that human-computer dialogue data include
As input information;Or
Understand model according to human-computer dialogue data run second language, using obtain the jump condition in finite state machine as
Input information;
Wherein, first language model or second language model are to fill out slot language to corresponding according to each slot-value in session operational scenarios
What material and the training in advance of cancellation slot language material obtained.
In the present embodiment, engaged in the dialogue management in the form of finite state machine, is conducive to fill in various session operational scenarios
The advantage of finite state machine is waved in distribution, realizes simpler, neatly dialogue management.
It is worth noting that in application scenes, method logic shown in Fig. 3 and Fig. 4 can be disposed people shown in Fig. 1
Server end in machine conversational system, is executed by server, and but it is not limited to this.For example, with the development of terminal technology, terminal
The function of equipment is stronger and stronger, and method logic can also be deployed in terminal shown in Fig. 3 and Fig. 4, is taken without disposing
Business device, this is conducive to the realization framework for simplifying interactive system.Embodiment and one is disposed below in conjunction with above two
A little concrete application scenes illustrate the technical solution of the embodiment of the present application.
Application scenarios 1:
In home scenarios, family can be configured and accompanied and attended to robot.Robot is accompanied and attended to by family can be old instead of adult's nurse
People or child can therefrom free adult.Family accompany and attend to robot can accompany old people people or child game, read, chat,
Old man is reminded to take medicine.By taking chat scenario as an example, robot is accompanied and attended to by family can be as chatting object, according to the chat field of setting
Scape is chatted with user.In the present embodiment, robot upper part administration of accompanying and attending to of family has and is responsible for dialogue state management in chat scenario
Finite state machine model, the finite state machine model are built using the method in above-described embodiment.
When user need chat when, can by the modes such as voice, touch-control or physical button by family accompany and attend to robot from
Suspend mode or standby mode wake up, subsequently into chat process.As shown in Figure 5 a, the user robot 50a that can accompany and attend to family is said
In short, for example, " thering is New cinema to show within nearest one week ".The family robot 50a that accompanies and attends to receives voice data input by user,
Then human-computer dialogue processing is carried out for " having New cinema to show recently " that user says according to dialog process flow shown in Fig. 5 b,
And final output answer.Wherein, dialog process process shown in Fig. 5 b includes:Speech recognition, language understanding, dialogue management, language
It generates and several steps such as phonetic synthesis.Wherein, in dialogue management component, based on the finite state machine model built in advance,
Utilize the transfer and action between the format management dialogue state of finite state machine.As shown in Figure 5 b, in finite state machine model
Dialogue state may include slot filling definition dialogue state and the later stage extension dialogue state.
Wherein, family accompany and attend to the answer that robot 50a is provided may be with the relevant recent messages of film, such as recently on
The information of several films reflected, shows and the information such as more fiery brief introduction, the protagonist of several films recently, the foreign countries shown recently
Act large stretch of information, etc..Alternatively, if the problem of user have exceeded setting chat scenario scope or language material it is insufficient,
The family robot 50a that accompanies and attends to can also provide the answers such as " not knowing ", " unclear ".
Optionally, it accompanies and attends to robot to simplify family, it can be by dialog process function distributing shown in Fig. 5 b a to cloud
Server is realized.Based on this, the corresponding another interactive system of the scene that can be applied 1, as shown in Figure 5 c, including:
Family accompanies and attends to robot 50c and Cloud Server 50d, is deployed on Cloud Server 50d using the method structure in above-described embodiment
The finite state machine model built out.
In the system shown in Fig. 5 c, when user needs chat, it can be incited somebody to action by modes such as voice, touch-control or physical buttons
Robot 50c accompanies and attends to from suspend mode or standby mode wake-up, subsequently into chat process in family.Accompany user chat process include:
User says in short, such as " having New cinema to show within nearest one week ".The words can be sent to by the family robot 50c that accompanies and attends to
Cloud Server 50d.Cloud Server 50d " has New cinema in nearest one week according to dialog process flow shown in Fig. 5 b for what user said
Show " human-computer dialogue processing is carried out, and answer is finally obtained, such as " information for the external action sheet shown recently ", and
Answer is returned to family to accompany and attend to robot 50c, answer is played to user by the family robot 50c that accompanies and attends to.
Engagement groove is filled in the building process of finite state machine so that the dialogue management scheme based on finite state machine can
To apply in scene is accompanied and attended to by robot, engaged in the dialogue management in the form of finite state machine, is conducive to accompany and attend to field in robot
It gives full play to the advantage of finite state machine in scape, realizes simpler, neatly dialogue management so that effect is accompanied and attended to more by robot
Ideal, and then improve user's impression.
Application scenarios 2:
Withdrawal is handled in order to facilitate user and carries out checking that the business such as inquiry, existing each bank all use self-service business
System, as fig 5d, bank self-help operation system include bank server 50e and the self-service withdrawal for being deployed in many places
Machine 50f, self-help inquiry apparatus 50g etc..These self-service automatic teller machine 50f, self-help inquiry apparatus 50g have good in interactive function, can be with
It is interacted with user, further combined with the dialog process service that bank server 50e is provided, user's withdrawal, inquiry etc. can be met
Business demand.The finite state for being responsible for dialogue state management in bank self-help business scenario is deployed on bank server 50e
Machine model, the finite state machine model are built using the method in above-described embodiment.
In the bank self-help operation system shown in Fig. 5 d, user can be to self-service automatic teller machine 50f or self-help inquiry apparatus 50g
Say the business demand of oneself.By taking withdrawal as an example, user can say withdrawal demand to self-service automatic teller machine 50f, such as " take
Money ".The withdrawal demand " withdrawal " of user is sent to bank server 50e by self-service automatic teller machine 50f.Bank server 50e is pressed
Human-computer dialogue processing is carried out for " withdrawal " that user says according to dialog process flow shown in Fig. 5 b, and obtains answer, such as " may I ask
Withdraw funds are how many ", answer " may I ask withdraw funds are how many " is returned into self-service automatic teller machine 50f.Self-service automatic teller machine
Answer " may I ask withdraw funds are how many " is played to user by 50f.
User continues to say withdraw funds, such as " taking 3,000 " to self-service automatic teller machine 50f;Self-service automatic teller machine 50f will
The withdraw funds " taking 3,000 " of user are sent to bank server 50e, and bank server 50e continues at according to dialogue shown in Fig. 5 b
Reason flow carries out human-computer dialogue processing for " taking 3,000 " that user says, and obtains answer, such as " please input withdrawal password ", will
Answer " please input withdrawal password " returns to self-service automatic teller machine 50f.Entire withdrawal process can be limited according to what is built in advance
State machine model executes successively, until failure of withdrawing the money successfully or withdraw the money.In figure 5d, further part does not show that.
Engagement groove is filled in the building process of finite state machine so that the dialogue management scheme based on finite state machine can
To apply in bank self-help business scenario, engaged in the dialogue management in the form of finite state machine, is conducive in bank self-help industry
The advantage of finite state machine is given full play in business scene, realizes simpler, neatly dialogue management so that bank self-help business
More efficient, human-computer interaction it is more smooth, be conducive to improve user's impression.
Application scenarios 3:
With the development of Internet technology, user, which stays indoors, can enjoy various services.By taking network booking as an example, if with
Family needs to go on business, travel or family goes home to see one's folks holiday, can greatly be saved directly by cyber ordering ticket, air ticket etc.
It saves time.
Network seat reservation system as depicted in fig. 5e includes:The ticket-booking service device 50h of user terminal 50g and passenger-traffic system;With
Family terminal 50g is established by internet and ticket-booking service device 50h and is communicated to connect.It is deployed on ticket-booking service device 50h and is purchased in network
It is responsible for the finite state machine model of dialogue management in ticket scene, which is using the method in above-described embodiment
Structure.
When user needs ticket booking, the ticket booking software installed on user terminal 50g is opened;Then, it is proposed to the ticket booking software
The booking demands ofdifferent classes of oneself, such as " ordering the air ticket for going to Shanghai ".Optionally, user can be manually entered the booking demands ofdifferent classes of oneself,
The booking demands ofdifferent classes of oneself can also be said by voice mode.The booking demands ofdifferent classes of user " are ordered to open and go to Shanghai by the ticket booking software
Air ticket " is sent to ticket-booking service device 50h.Ticket-booking service device 50h is directed to the ticket booking of user according to dialog process flow shown in Fig. 5 b
Demand " ordering the air ticket for going to Shanghai " carries out human-computer dialogue processing, and obtains answer, such as " may I ask the air ticket for determining several points ", will answer
Case " may I ask the air ticket for determining several points " returns to ticket booking software.Ticket booking software plays the answer air ticket of several points " may I ask determine " or display
To user.
User continues to say the ticket booking time of oneself, such as " tomorrow morning at 10 " to ticket booking software;Booking tickets software will
The ticket booking time " tomorrow morning at 10 " that user requires is sent to ticket-booking service device 50h.Ticket-booking service device 50h continues according to figure
The ticket booking time " tomorrow morning at 10 " that dialog process flow shown in 5b is required for user carries out human-computer dialogue processing, and obtains
Answer, such as " may I ask where departure place is " are obtained, answer " may I ask where departure place is " is returned into ticket booking software.Ticket booking
Answer " may I ask where departure place is " is played or is shown to user by software.Entire ticket booking process can be according to building in advance
Finite state machine model executes successively, until failure of booking tickets successfully or book tickets.In Fig. 5 e, further part does not show that.
Engagement groove is filled in the building process of finite state machine so that the dialogue management scheme based on finite state machine can
To apply in network booking business scenario, engaged in the dialogue management in the form of finite state machine, is conducive in network booking industry
The advantage of finite state machine is given full play in business scene, realizes simpler, neatly dialogue management so that the effect of network booking
Rate higher, human-computer interaction are more smooth, are conducive to improve user's impression.
It should be noted that the executive agent of each step of above-described embodiment institute providing method may each be same equipment,
Alternatively, this method is also by distinct device as executive agent.For example, the executive agent of step 401 to step 403 can be equipment
A;For another example, step 401 and 402 executive agent can be device A, the executive agent of step 403 can be equipment B;Etc..
In addition, in some flows of description in above-described embodiment and attached drawing, contains and occur according to particular order
Multiple operations, but it should be clearly understood that these operations can not execute or parallel according to its sequence what appears in this article
It executes, the serial number such as 401,402 etc. of operation is only used for distinguishing each different operation, serial number itself does not represent any
Execute sequence.In addition, these flows may include more or fewer operations, and these operations can execute in order or
It is parallel to execute.It should be noted that the descriptions such as herein " first ", " second ", be for distinguish different message, equipment,
Module etc. does not represent sequencing, does not also limit " first " and " second " and is different type.
Fig. 6 a are a kind of structural representation for Dialogue management strategy generating means that the application another exemplary embodiment provides
Figure.As shown in Figure 6 a, which includes:Determining module 61, acquisition module 62, generation module 63 and structure module 64.
Determining module 61, for based on the semantic understanding to session operational scenarios, determining multiple semantemes suitable for session operational scenarios
Slot and the corresponding candidate slot value of multiple semantic slots;
Acquisition module 62 has dialogue meaning for being combined to the corresponding candidate slot value of multiple semanteme slots to obtain
Multigroup slot-value pair, every group includes multiple semantic corresponding slot-values pair of slot;
Generation module 63, the semanteme for respectively being indicated according to multigroup slot-value are generated with multigroup slot-value to corresponding more
A dialogue state;
Module 64 is built, is used for according to multiple dialogue states and multigroup slot-value to building finite state machine model, to utilize
The form of finite state machine engages in the dialogue management to the human-computer dialogue process in session operational scenarios.
In some optional embodiments, structure module 64 is specifically used for when building finite state machine model:It will be multiple
Dialogue state is mapped as multiple state nodes in finite state machine model;Any two state node in multiple state nodes
Between add two-way side;And the difference according to the corresponding two groups of slot-values of any two state node between, generate arbitrary two
Jump condition when being shifted between a state node, to build finite state machine model.
In some optional embodiments, acquisition module 62 is additionally operable to:Each slot-value in session operational scenarios is obtained to fill out to corresponding
Slot language material and cancellation slot language material, to form corpus;According to each slot-value in corpus slot language material and cancellation slot language are filled out to corresponding
Material training language understands model, and the language understanding model from human-computer dialogue data for obtaining the input needed for finite state machine
Information.
Further, acquisition module 62 is specifically used for when training language understands model:
Model is understood to corresponding slot language material and the cancellation slot language material training first language filled out according to each slot-value in corpus,
First language understands model for extracting slot-value that human-computer dialogue data include to as input information;Or
According to each slot-value in corpus to it is corresponding fill out slot language material and cancel slot language material and multigroup slot-value pair with it is multiple
Correspondence training second language between dialogue state understands that model, second language understand that model is used for from human-computer dialogue data
One jump condition of middle acquisition is as input information.
In some optional embodiments, structure module 64 is additionally operable to:According to new dialogue state and new dialogue shape
The corresponding one group of slot-value pair of state, increases new state node in finite state machine model;New state node with it is each
It adds two-way side between stateful node, and the corresponding two groups of slot-values of state node is had with each according to new state node
Difference between generates jump condition when being shifted between new state node and each existing state node.
In some optional embodiments, acquisition module 62 is additionally operable to:Obtain the human-computer dialogue number in the session operational scenarios
According to.Correspondingly, which further includes:Language understanding module, dialogue management module and language generation mould
Block.
Language understanding module, for filling out slot language material and cancellation slot language to corresponding according to each slot-value in the session operational scenarios
Material, from the human-computer dialogue data obtain can trigger finite state machine engage in the dialogue state transfer input information.
Dialogue management module is redirected for controlling the finite state machine according to the input information from current dialogue states
To Next dialog states.
Language generation module exports the human-computer dialogue data for the related data according to the Next dialog states
Reply data.
Further, Dialogue management strategy generating means can also include:Sound identification module and voice synthetic module.Voice
Identification module for human-computer dialogue data to be converted to text data, and is supplied to language understanding model.Voice synthetic module,
Reply data for generating language generation module is converted to voice data.
The foregoing describe the built-in function of Dialogue management strategy generating means and structures, and as shown in Figure 6 b, in practice, this is right
Words management generating means can realize as man-machine dialogue equipment, including:Memory 601 and processor 602.
Memory 601 for storing computer program, and can be configured as storing various other data to support man-machine
Operation on conversational device.The example of these data includes for any application program operated on man-machine dialogue equipment or side
The instruction of method, contact data, telephone book data, message, picture, video etc..
Memory 601 can realize by any kind of volatibility or non-volatile memory device or combination thereof,
Such as static RAM (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable is read-only
Memory (EPROM), programmable read only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk
Or CD.
Processor 602 is coupled with memory 601, for executing the computer program in memory 601, for:
Based on the semantic understanding to session operational scenarios, multiple semantic slots suitable for session operational scenarios and multiple semantic slots pair are determined
The candidate slot value answered;
The corresponding candidate slot value of multiple semanteme slots is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of multiple semantic slots;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to multiple dialogue states and multigroup slot-value to building finite state machine model, to utilize the shape of finite state machine
Formula engages in the dialogue management to the human-computer dialogue process in session operational scenarios.
In some optional embodiments, processor 602 is specifically used for when building finite state machine model:It will be multiple
Dialogue state is mapped as multiple state nodes in finite state machine model;Any two state node in multiple state nodes
Between add two-way side;And the difference according to the corresponding two groups of slot-values of any two state node between, generate arbitrary two
Jump condition when being shifted between a state node, to build finite state machine model.
In some optional embodiments, processor 602 is additionally operable to:Each slot-value in session operational scenarios is obtained to fill out to corresponding
Slot language material and cancellation slot language material, to form corpus;According to each slot-value in corpus slot language material and cancellation slot language are filled out to corresponding
Material training language understands model, and the language understanding model from human-computer dialogue data for obtaining the input needed for finite state machine
Information.
Further, processor 602 is specifically used for when training language understands model:
Model is understood to corresponding slot language material and the cancellation slot language material training first language filled out according to each slot-value in corpus,
First language understands model for extracting slot-value that human-computer dialogue data include to as input information;Or
According to each slot-value in corpus to it is corresponding fill out slot language material and cancel slot language material and multigroup slot-value pair with it is multiple
Correspondence training second language between dialogue state understands that model, second language understand that model is used for from human-computer dialogue data
One jump condition of middle acquisition is as input information.
In some optional embodiments, processor 602 is additionally operable to:According to new dialogue state and new dialogue state
Corresponding one group of slot-value pair, increases new state node in finite state machine model;Have with each in new state node
It adds two-way side between state node, and the corresponding two groups of slot-values pair of state node is had with each according to new state node
Between difference, generate jump condition when being shifted between new state node and each existing state node.
In some optional embodiments, processor 602 is additionally operable to:Obtain the human-computer dialogue number in the session operational scenarios
According to;Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue data
The input information that the state that the triggerable finite state machine of acquisition engages in the dialogue shifts;The limited shape is controlled according to the input information
State machine jumps to Next dialog states from current dialogue states;According to the related data of the Next dialog states, described in output
The reply data of human-computer dialogue data.
Further, processor 602 is additionally operable to:Human-computer dialogue data are converted into text data, and reply data is turned
It is changed to voice data.
Further, as shown in Figure 6 b, which further includes:Communication component 603, power supply module 604 etc. are other
Component.Members are only schematically provided in Fig. 6 b, are not meant to that man-machine dialogue equipment only includes component shown in Fig. 6 b.
In application scenes, man-machine dialogue equipment shown in Fig. 6 b can be server, such as can be regular service
The server apparatus such as device, Cloud Server, cloud host, virtual center.
In other application scenarios, man-machine dialogue equipment shown in Fig. 6 b can be terminal device, such as can be installation
There are smart mobile phone, tablet computer, PC, wearable device, the intelligent sound etc. of various interactive voice class application software, or
Can be self-help registration/payment machine of various types of voice interactive self-service terminal, self-service machine, such as hospital, bank it is self-service
Cash dispenser, the automatic ticket taking machine etc. in the scenes such as subway, station or airport;Or can be that the family of interactive voice is supported to accompany and attend to
Class robot, chat robots, sweeping robot, navigation/follow robot, offer order the robot etc. of service.
Correspondingly, the embodiment of the present application also provides a kind of computer readable storage medium being stored with computer program, when
When the computer instruction is executed by one or more processors, it includes below dynamic to cause one or more of processor execution
Make:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize limited shape
The form of state machine engages in the dialogue management to the human-computer dialogue process in the session operational scenarios.
In addition to above-mentioned action, one or more processors can also be performed in aforementioned other embodiments and can be held by server
Capable other actions.
Fig. 7 a are a kind of structural schematic diagram for human-computer dialogue device that the application another exemplary embodiment provides.Such as Fig. 7 a
Shown, which includes:Acquisition module 71, language understanding module 72, dialogue management module 73 and language synthesis module 74.
Acquisition module 71, for obtaining the human-computer dialogue data in session operational scenarios;
Language understanding module 72, for filling out slot language material and cancellation slot to corresponding according to each slot-value in the session operational scenarios
Language material, from the human-computer dialogue data obtain can trigger finite state machine engage in the dialogue state transfer input information;
Dialogue management module 73 is jumped for controlling the finite state machine according to the input information from current dialogue states
Go to Next dialog states;
Language synthesis module 74 exports the human-computer dialogue number for the related data according to the Next dialog states
According to reply data.
In an optional embodiment, language understanding module 72 is specifically used for:
Understand model according to the human-computer dialogue data run first language, includes to obtain the human-computer dialogue data
Slot-value is to as the input information;Or
Model is understood according to the human-computer dialogue data run second language, to obtain the transfer in the finite state machine
Condition is as the input information;
Wherein, the first language model or the second language model are according to each slot-value pair in the session operational scenarios
It is corresponding to fill out slot language material and cancel what the training in advance of slot language material obtained.
In an optional embodiment, which further includes:Build module.The structure module, is used for:
Based on the semantic understanding to the session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and described
The corresponding candidate slot value of multiple semanteme slots;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model.
The foregoing describe the built-in function of human-computer dialogue device and structures, as shown in Figure 7b, in practice, human-computer dialogue dress
It sets and can be achieved as man-machine dialogue equipment, including:Memory 701 and processor 702.
Memory 701 for storing computer program, and can be configured as storing various other data to support man-machine
Operation on conversational device.The example of these data includes for any application program operated on man-machine dialogue equipment or side
The instruction of method, contact data, telephone book data, message, picture, video etc..
Memory 701 can realize by any kind of volatibility or non-volatile memory device or combination thereof,
Such as static RAM (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable is read-only
Memory (EPROM), programmable read only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk
Or CD.
Processor 702 is coupled with memory 701, for executing the computer program in memory 701, for:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue
In data obtain can trigger finite state machine engage in the dialogue state transfer input information;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
In an optional embodiment, processor 702 is specifically used for:
Understand model according to the human-computer dialogue data run first language, includes to obtain the human-computer dialogue data
Slot-value is to as the input information;Or
Model is understood according to the human-computer dialogue data run second language, to obtain the transfer in the finite state machine
Condition is as the input information;
Wherein, the first language model or the second language model are according to each slot-value pair in the session operational scenarios
It is corresponding to fill out slot language material and cancel what the training in advance of slot language material obtained.
In an optional embodiment, processor 702 is additionally operable to:
Based on the semantic understanding to the session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and described
The corresponding candidate slot value of multiple semanteme slots;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value with dialogue meaning
Right, every group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue shapes
State;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model.
Further, as shown in Figure 7b, which further includes:Communication component 703, display 704, power supply module
705, other components such as audio component 706.Members are only schematically provided in Fig. 7 b, are not meant to man-machine dialogue equipment only
Including component shown in Fig. 7 b.
In application scenes, man-machine dialogue equipment shown in Fig. 7 b can be server, such as can be regular service
The server apparatus such as device, Cloud Server, cloud host, virtual center.
In other application scenarios, man-machine dialogue equipment shown in Fig. 7 b can be terminal device, such as can be installation
There are smart mobile phone, tablet computer, PC, wearable device, the intelligent sound etc. of various interactive voice class application software, or
Can be self-help registration/payment machine of various types of voice interactive self-service terminal, self-service machine, such as hospital, bank it is self-service
Cash dispenser, the automatic ticket taking machine etc. in the scenes such as subway, station or airport;Or can be that the family of interactive voice is supported to accompany and attend to
Class robot, chat robots, sweeping robot, navigation/follow robot, offer order the robot etc. of service.
Correspondingly, the embodiment of the present application also provides a kind of computer readable storage medium being stored with computer program, when
When the computer instruction is executed by one or more processors, it includes below dynamic to cause one or more of processor execution
Make:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue
In data obtain can trigger finite state machine engage in the dialogue state transfer input information;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
In addition to above-mentioned action, one or more processors can also be performed can be by terminal device in aforementioned other embodiments
The other actions executed.
Where communication component in above-mentioned Fig. 6 b and Fig. 7 b is configured to facilitate communication component between equipment and other equipment
The communication of wired or wireless way.Equipment where communication component can access the wireless network based on communication standard, such as WiFi, 2G
Or 3G or combination thereof.In one exemplary embodiment, communication component receives via broadcast channel and comes from external broadcasting pipe
The broadcast singal or broadcast related information of reason system.In one exemplary embodiment, the communication component further includes that near field is logical
(NFC) module is believed, to promote short range communication.For example, radio frequency identification (RFID) technology, infrared data association can be based in NFC module
Meeting (IrDA) technology, ultra wide band (UWB) technology, bluetooth (BT) technology and other technologies are realized.
Display in above-mentioned Fig. 7 b includes screen, and screen may include liquid crystal display (LCD) and touch panel
(TP).If screen includes touch panel, screen may be implemented as touch screen, to receive input signal from the user.It touches
It includes one or more touch sensors to sense the gesture on touch, slide, and touch panel to touch panel.The touch sensing
Device can not only sense the boundary of a touch or slide action, but also detect it is associated with the touch or slide operation continue when
Between and pressure.
Power supply module in above-mentioned Fig. 6 b and Fig. 7 b, the various assemblies of equipment provide electric power where power supply module.Power supply
Component may include power-supply management system, one or more power supplys and other with generated for equipment where power supply module, management and
Distribute the associated component of electric power.
Audio component in above-mentioned Fig. 7 b can be configured as output and/or input audio signal.For example, audio component packet
A microphone (MIC) is included, the equipment where audio component is in operation mode, as call model, logging mode and voice are known
When other pattern, microphone is configured as receiving external audio signal.The received audio signal can be further stored in and deposit
Reservoir is sent via communication component.In some embodiments, audio component further includes a loud speaker, for exporting audio letter
Number.
It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, the present invention can be used in one or more wherein include computer usable program code computer
The computer program production implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The present invention be with reference to according to the method for the embodiment of the present invention, the flow of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be realized by computer program instructions every first-class in flowchart and/or the block diagram
The combination of flow and/or box in journey and/or box and flowchart and/or the block diagram.These computer programs can be provided
Instruct the processor of all-purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine so that the instruction executed by computer or the processor of other programmable data processing devices is generated for real
The device for the function of being specified in present one flow of flow chart or one box of multiple flows and/or block diagram or multiple boxes.
These computer program instructions, which may also be stored in, can guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works so that instruction generation stored in the computer readable memory includes referring to
Enable the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one box of block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device so that count
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, in computer or
The instruction executed on other programmable devices is provided for realizing in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in a box or multiple boxes.
In a typical configuration, computing device includes one or more processors (CPU), input/output interface, net
Network interface and memory.
Memory may include computer-readable medium in volatile memory, random access memory (RAM) and/or
The forms such as Nonvolatile memory, such as read-only memory (ROM) or flash memory (flash RAM).Memory is computer-readable medium
Example.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology realizes information storage.Information can be computer-readable instruction, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase transition internal memory (PRAM), static RAM (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), fast flash memory bank or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storages, magnetic tape cassette, tape magnetic disk storage or other magnetic storage apparatus
Or any other non-transmission medium, it can be used for storage and can be accessed by a computing device information.As defined in this article, it calculates
Machine readable medium does not include temporary computer readable media (transitory media), such as data-signal and carrier wave of modulation.
It should also be noted that, the terms "include", "comprise" or its any other variant are intended to nonexcludability
Including so that process, method, commodity or equipment including a series of elements include not only those elements, but also wrap
Include other elements that are not explicitly listed, or further include for this process, method, commodity or equipment intrinsic want
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that wanted including described
There is also other identical elements in the process of element, method, commodity or equipment.
Above is only an example of the present application, it is not intended to limit this application.For those skilled in the art
For, the application can have various modifications and variations.It is all within spirit herein and principle made by any modification, equivalent
Replace, improve etc., it should be included within the scope of claims hereof.
Claims (22)
1. a kind of Dialogue management strategy generation method, which is characterized in that including:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple semanteme
The corresponding candidate slot value of slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize finite state machine
Form engage in the dialogue management to the human-computer dialogue process in the session operational scenarios.
2. according to the method described in claim 1, it is characterized in that, described according to the multiple dialogue state and described multigroup
Slot-value to build finite state machine model, including:
The multiple state nodes the multiple dialogue state being mapped as in the finite state machine model;
Two-way side is added between any two state node in the multiple state node;And
According to difference of the corresponding two groups of slot-values of any two state node between, any two state is generated
Jump condition when being shifted between node, to build the finite state machine model.
3. according to the method described in claim 1, it is characterized in that, further including:
It obtains each slot-value in the session operational scenarios and fills out slot language material and cancellation slot language material to corresponding, to form corpus;
Model is understood to corresponding slot language material and the cancellation slot language material training language filled out according to each slot-value in the corpus, it is described
Language understanding model is used to obtain the input information needed for the finite state machine from human-computer dialogue data.
4. according to the method described in claim 3, it is characterized in that, it is described according to each slot-value in the corpus to corresponding
It fills out slot language material and cancels slot language material training language and understand model, including:
Model is understood to corresponding slot language material and the cancellation slot language material training first language filled out according to each slot-value in the corpus,
The first language understands model for extracting slot-value that the human-computer dialogue data include to as the input information;Or
Person
According to each slot-value in the corpus to it is corresponding fill out slot language material and cancel slot language material and multigroup slot-value pair with
Correspondence training second language between the multiple dialogue state understands model, the second language understand model for from
A jump condition is obtained in the human-computer dialogue data as the input information.
5. according to claim 1-4 any one of them methods, which is characterized in that according to the multiple dialogue state and described
After multigroup slot-value is to structure finite state machine model, further include:
According to new dialogue state and the corresponding one group of slot-value pair of the new dialogue state, in the finite state machine mould
Increase new state node in type;
Two-way side is added in the new state node and each between existing state node, and according to the new state node
Difference of the two groups of slot-values corresponding with each existing state node between generates the new state node and has with each
Jump condition when being shifted between state node.
6. a kind of interactive method, which is characterized in that including:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue data
The input information that the state that the middle triggerable finite state machine of acquisition engages in the dialogue shifts;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
7. according to the method described in claim 6, it is characterized in that, it is described according to each slot-value in the session operational scenarios to correspondence
Fill out slot language material and cancel slot language material, from the human-computer dialogue data obtain can trigger finite state machine engage in the dialogue state turn
The input information of shifting, including:
Model is understood according to the human-computer dialogue data run first language, to obtain the slot-that the human-computer dialogue data include
Value is to as the input information;Or
Model is understood according to the human-computer dialogue data run second language, to obtain the jump condition in the finite state machine
As the input information;
Wherein, the first language model or the second language model are according to each slot-value in the session operational scenarios to correspondence
Fill out slot language material and cancel what the training in advance of slot language material obtained.
8. the method described according to claim 6 or 7, which is characterized in that controlling the limited shape according to the input information
Before state machine jumps to Next dialog states from current dialogue states, further include:
Based on the semantic understanding to the session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize finite state machine
Form engage in the dialogue management to the human-computer dialogue process in the session operational scenarios.
9. a kind of man-machine dialogue equipment, which is characterized in that including:Memory and processor;
Memory, for storing computer program;
The processor, for executing the computer program, for:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple semanteme
The corresponding candidate slot value of slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize finite state machine
Form engage in the dialogue management to the human-computer dialogue process in the session operational scenarios.
10. man-machine dialogue equipment according to claim 9, which is characterized in that the processor is specifically used for:
The multiple state nodes the multiple dialogue state being mapped as in the finite state machine model;
Two-way side is added between any two state node in the multiple state node;And
According to difference of the corresponding two groups of slot-values of any two state node between, any two state is generated
Jump condition when being shifted between node, to build the finite state machine model.
11. man-machine dialogue equipment according to claim 9, which is characterized in that the processor is additionally operable to:
It obtains each slot-value in the session operational scenarios and fills out slot language material and cancellation slot language material to corresponding, to form corpus;
Model is understood to corresponding slot language material and the cancellation slot language material training language filled out according to each slot-value in the corpus, it is described
Language understanding model is used to obtain the input information needed for the finite state machine from human-computer dialogue data.
12. man-machine dialogue equipment according to claim 11, which is characterized in that the processor is specifically used for:
Model is understood to corresponding slot language material and the cancellation slot language material training first language filled out according to each slot-value in the corpus,
The first language understands model for extracting slot-value that the human-computer dialogue data include to as the input information;Or
Person
According to each slot-value in the corpus to it is corresponding fill out slot language material and cancel slot language material and multigroup slot-value pair with
Correspondence training second language between the multiple dialogue state understands model, the second language understand model for from
A jump condition is obtained in the human-computer dialogue data as the input information.
13. according to claim 9-12 any one of them man-machine dialogue equipments, which is characterized in that the processor is additionally operable to:
According to new dialogue state and the corresponding one group of slot-value pair of the new dialogue state, in the finite state machine mould
Increase new state node in type;
Two-way side is added in the new state node and each between existing state node, and according to the new state node
Difference of the two groups of slot-values corresponding with each existing state node between generates the new state node and has with each
Jump condition when being shifted between state node.
14. according to claim 9-12 any one of them man-machine dialogue equipments, which is characterized in that the processor is additionally operable to:
Obtain the human-computer dialogue data in the session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue data
The input information that the state that the middle triggerable finite state machine of acquisition engages in the dialogue shifts;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
15. a kind of computer readable storage medium of storage computer instruction, which is characterized in that when the computer instruction is by one
When a or multiple processors execute, it includes action below to cause one or more of processor execution:
Based on the semantic understanding to session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple semanteme
The corresponding candidate slot value of slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize finite state machine
Form engage in the dialogue management to the human-computer dialogue process in the session operational scenarios.
16. a kind of man-machine dialogue equipment, which is characterized in that including:Memory and processor;
The memory, for storing computer program;
The processor, for executing the computer program, for:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue data
The input information that the state that the middle triggerable finite state machine of acquisition engages in the dialogue shifts;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
17. man-machine dialogue equipment according to claim 16, which is characterized in that the processor is specifically used for:
Model is understood according to the human-computer dialogue data run first language, to obtain the slot-that the human-computer dialogue data include
Value is to as the input information;Or
Model is understood according to the human-computer dialogue data run second language, to obtain the jump condition in the finite state machine
As the input information;
Wherein, the first language model or the second language model are according to each slot-value in the session operational scenarios to correspondence
Fill out slot language material and cancel what the training in advance of slot language material obtained.
18. man-machine dialogue equipment according to claim 16, which is characterized in that the processor is additionally operable to:
Based on the semantic understanding to the session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model, to utilize finite state machine
Form engage in the dialogue management to the human-computer dialogue process in the session operational scenarios.
19. according to claim 16-18 any one of them terminal devices, which is characterized in that the terminal device includes following
It is at least one:
Intelligent robot, self-service machine, self-aided terminal, intelligent terminal and Self-help vending machine.
20. a kind of computer readable storage medium of storage computer instruction, which is characterized in that when the computer instruction is by one
When a or multiple processors execute, it includes action below to cause one or more of processor execution:
Obtain the human-computer dialogue data in session operational scenarios;
Slot language material and cancellation slot language material are filled out to corresponding according to each slot-value in the session operational scenarios, from the human-computer dialogue data
The input information that the state that the middle triggerable finite state machine of acquisition engages in the dialogue shifts;
The finite state machine, which is controlled, according to the input information jumps to Next dialog states from current dialogue states;
According to the related data of the Next dialog states, the reply data of the human-computer dialogue data is exported.
21. a kind of interactive system, which is characterized in that including:Server and terminal device;
The terminal device, for receiving human-computer dialogue data input by user in session operational scenarios, by the human-computer dialogue data
It is sent to the server, and receives the corresponding reply data of the human-computer dialogue data of the server return and exports
To the user;
The server, the human-computer dialogue data sent for receiving the terminal device, according to each slot-in the session operational scenarios
Value fills out slot language material and cancels slot language material to corresponding, is obtained from the human-computer dialogue data and can trigger finite state machine and carry out pair
The input information of speech phase transfer;The finite state machine is controlled according to the input information to jump to down from current dialogue states
One dialogue state;According to the related data of the Next dialog states, the human-computer dialogue data are returned to the terminal device
Reply data.
22. system according to claim 21, which is characterized in that the server is additionally operable to:
Based on the semantic understanding to the session operational scenarios, determine suitable for multiple semantic slots of the session operational scenarios and the multiple
The corresponding candidate slot value of semantic slot;
The corresponding candidate slot value of the multiple semanteme slot is combined, to obtain multigroup slot-value pair with dialogue meaning, often
Group includes the corresponding slot-value pair of the multiple semantic slot;
According to the semanteme that multigroup slot-value respectively indicates, generate with multigroup slot-value to corresponding multiple dialogue states;
According to the multiple dialogue state and multigroup slot-value to building finite state machine model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810421418.8A CN108804536B (en) | 2018-05-04 | 2018-05-04 | Man-machine conversation and strategy generation method, equipment, system and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810421418.8A CN108804536B (en) | 2018-05-04 | 2018-05-04 | Man-machine conversation and strategy generation method, equipment, system and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108804536A true CN108804536A (en) | 2018-11-13 |
CN108804536B CN108804536B (en) | 2022-10-04 |
Family
ID=64093376
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810421418.8A Active CN108804536B (en) | 2018-05-04 | 2018-05-04 | Man-machine conversation and strategy generation method, equipment, system and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108804536B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109514586A (en) * | 2019-01-30 | 2019-03-26 | 第四范式(北京)技术有限公司 | Realize the method and system of intelligent customer service robot |
CN109637527A (en) * | 2018-12-13 | 2019-04-16 | 苏州思必驰信息科技有限公司 | The semantic analytic method and system of conversation sentence |
CN110096579A (en) * | 2019-04-23 | 2019-08-06 | 南京硅基智能科技有限公司 | A kind of more wheel dialogue methods |
CN110096583A (en) * | 2019-05-09 | 2019-08-06 | 苏州思必驰信息科技有限公司 | Multi-field dialog management system and its construction method |
CN110377716A (en) * | 2019-07-23 | 2019-10-25 | 百度在线网络技术(北京)有限公司 | Exchange method, device and the computer readable storage medium of dialogue |
CN110704595A (en) * | 2019-09-27 | 2020-01-17 | 百度在线网络技术(北京)有限公司 | Dialogue processing method and device, electronic equipment and readable storage medium |
CN111046674A (en) * | 2019-12-20 | 2020-04-21 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111104502A (en) * | 2019-12-24 | 2020-05-05 | 携程计算机技术(上海)有限公司 | Dialogue management method, system, electronic device and storage medium for outbound system |
CN111241236A (en) * | 2019-12-30 | 2020-06-05 | 新大陆数字技术股份有限公司 | Task-oriented question-answering method, system, electronic equipment and readable storage medium |
CN111611357A (en) * | 2019-02-25 | 2020-09-01 | 北京嘀嘀无限科技发展有限公司 | Configuration method of man-machine conversation system, multi-turn conversation configuration platform and electronic equipment |
CN111666387A (en) * | 2020-04-13 | 2020-09-15 | 北京百度网讯科技有限公司 | Dialog generation method and device, electronic equipment and storage medium |
CN112437002A (en) * | 2020-11-23 | 2021-03-02 | 彩讯科技股份有限公司 | Food ordering method, system, equipment and storage medium based on RCS message |
WO2021218061A1 (en) * | 2020-04-28 | 2021-11-04 | 平安科技(深圳)有限公司 | Smart robot deployment method, apparatus, device, and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101517533A (en) * | 2006-09-15 | 2009-08-26 | 微软公司 | Transformation of modular finite state transducers |
CN105630764A (en) * | 2015-12-22 | 2016-06-01 | 百度在线网络技术(北京)有限公司 | Address resolution method of finite-state machine and address resolution device of finite-state machine |
CN106383872A (en) * | 2016-09-06 | 2017-02-08 | 北京百度网讯科技有限公司 | Artificial intelligence-based information processing method and apparatus |
-
2018
- 2018-05-04 CN CN201810421418.8A patent/CN108804536B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101517533A (en) * | 2006-09-15 | 2009-08-26 | 微软公司 | Transformation of modular finite state transducers |
CN105630764A (en) * | 2015-12-22 | 2016-06-01 | 百度在线网络技术(北京)有限公司 | Address resolution method of finite-state machine and address resolution device of finite-state machine |
CN106383872A (en) * | 2016-09-06 | 2017-02-08 | 北京百度网讯科技有限公司 | Artificial intelligence-based information processing method and apparatus |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109637527B (en) * | 2018-12-13 | 2021-08-31 | 思必驰科技股份有限公司 | Semantic analysis method and system for dialogue statement |
CN109637527A (en) * | 2018-12-13 | 2019-04-16 | 苏州思必驰信息科技有限公司 | The semantic analytic method and system of conversation sentence |
CN109514586A (en) * | 2019-01-30 | 2019-03-26 | 第四范式(北京)技术有限公司 | Realize the method and system of intelligent customer service robot |
CN111611357B (en) * | 2019-02-25 | 2023-08-15 | 北京嘀嘀无限科技发展有限公司 | Configuration method of man-machine conversation system, multi-round conversation configuration platform and electronic equipment |
CN111611357A (en) * | 2019-02-25 | 2020-09-01 | 北京嘀嘀无限科技发展有限公司 | Configuration method of man-machine conversation system, multi-turn conversation configuration platform and electronic equipment |
CN110096579A (en) * | 2019-04-23 | 2019-08-06 | 南京硅基智能科技有限公司 | A kind of more wheel dialogue methods |
CN110096583A (en) * | 2019-05-09 | 2019-08-06 | 苏州思必驰信息科技有限公司 | Multi-field dialog management system and its construction method |
CN110377716A (en) * | 2019-07-23 | 2019-10-25 | 百度在线网络技术(北京)有限公司 | Exchange method, device and the computer readable storage medium of dialogue |
JP7150770B2 (en) | 2019-07-23 | 2022-10-11 | バイドゥ オンライン ネットワーク テクノロジー(ペキン) カンパニー リミテッド | Interactive method, device, computer-readable storage medium, and program |
CN110377716B (en) * | 2019-07-23 | 2022-07-12 | 百度在线网络技术(北京)有限公司 | Interaction method and device for conversation and computer readable storage medium |
US11322153B2 (en) | 2019-07-23 | 2022-05-03 | Baidu Online Network Technology (Beijing) Co., Ltd. | Conversation interaction method, apparatus and computer readable storage medium |
JP2021018797A (en) * | 2019-07-23 | 2021-02-15 | バイドゥ オンライン ネットワーク テクノロジー (ベイジン) カンパニー リミテッド | Conversation interaction method, apparatus, computer readable storage medium, and program |
CN110704595A (en) * | 2019-09-27 | 2020-01-17 | 百度在线网络技术(北京)有限公司 | Dialogue processing method and device, electronic equipment and readable storage medium |
CN111046674A (en) * | 2019-12-20 | 2020-04-21 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111046674B (en) * | 2019-12-20 | 2024-05-31 | 科大讯飞股份有限公司 | Semantic understanding method and device, electronic equipment and storage medium |
CN111104502A (en) * | 2019-12-24 | 2020-05-05 | 携程计算机技术(上海)有限公司 | Dialogue management method, system, electronic device and storage medium for outbound system |
CN111241236A (en) * | 2019-12-30 | 2020-06-05 | 新大陆数字技术股份有限公司 | Task-oriented question-answering method, system, electronic equipment and readable storage medium |
CN111241236B (en) * | 2019-12-30 | 2023-08-22 | 新大陆数字技术股份有限公司 | Task-oriented question-answering method, system, electronic device and readable storage medium |
CN111666387A (en) * | 2020-04-13 | 2020-09-15 | 北京百度网讯科技有限公司 | Dialog generation method and device, electronic equipment and storage medium |
WO2021218061A1 (en) * | 2020-04-28 | 2021-11-04 | 平安科技(深圳)有限公司 | Smart robot deployment method, apparatus, device, and storage medium |
CN112437002A (en) * | 2020-11-23 | 2021-03-02 | 彩讯科技股份有限公司 | Food ordering method, system, equipment and storage medium based on RCS message |
Also Published As
Publication number | Publication date |
---|---|
CN108804536B (en) | 2022-10-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108804536A (en) | Human-computer dialogue and strategy-generating method, equipment, system and storage medium | |
US11295221B2 (en) | Learning user preferences in a conversational system | |
CN112136124B (en) | Dependency graph conversation modeling for human-machine conversation sessions with computer-implemented automated assistants | |
CN107294837A (en) | Engaged in the dialogue interactive method and system using virtual robot | |
US10088972B2 (en) | Virtual assistant conversations | |
CN116737900A (en) | Man-machine interaction processing system and method, storage medium and electronic equipment | |
CN104081382B (en) | Establish the method and system for the user interface that can dynamically specify | |
CN109463004A (en) | The far field of digital assistants service extends | |
KR102389331B1 (en) | Synchronize access control between computing devices | |
US9424838B2 (en) | Pattern processing system specific to a user group | |
CN109564530A (en) | The personal supplementary module for having the selectable state machine traversed is provided | |
CN109656512A (en) | Exchange method, device, storage medium and terminal based on voice assistant | |
CN108733438A (en) | Application program is integrated with digital assistants | |
CN109196464A (en) | User agent based on context | |
US11972095B2 (en) | Voice assistant-enabled client application with user view context and multi-modal input support | |
CN107329990A (en) | A kind of mood output intent and dialogue interactive system for virtual robot | |
CN109814832A (en) | Intelligent digital assistant in multitask environment | |
CN110459222A (en) | Sound control method, phonetic controller and terminal device | |
CN109102802A (en) | System for handling user spoken utterances | |
CN107632706A (en) | The application data processing method and system of multi-modal visual human | |
CN107430852B (en) | The selectivity of the online processing of voice input is stopped in the electronic equipment for supporting voice | |
CN105917405B (en) | Merging of the exogenous large vocabulary model to rule-based speech recognition | |
CN109313668B (en) | System and method for constructing session understanding system | |
CN107480766A (en) | The method and system of the content generation of multi-modal virtual robot | |
CN110019723A (en) | Answering method, terminal device and storage medium based on question answering system |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |