CN109272999A - Information processing unit, its interactive method and storage medium - Google Patents
Information processing unit, its interactive method and storage medium Download PDFInfo
- Publication number
- CN109272999A CN109272999A CN201811092347.8A CN201811092347A CN109272999A CN 109272999 A CN109272999 A CN 109272999A CN 201811092347 A CN201811092347 A CN 201811092347A CN 109272999 A CN109272999 A CN 109272999A
- Authority
- CN
- China
- Prior art keywords
- information
- field
- feedback
- processing unit
- candidate
- 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
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/1822—Parsing for meaning understanding
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L2015/223—Execution procedure of a spoken command
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L2015/225—Feedback of the input speech
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L2015/226—Procedures used during a speech recognition process, e.g. man-machine dialogue using non-speech characteristics
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L2015/226—Procedures used during a speech recognition process, e.g. man-machine dialogue using non-speech characteristics
- G10L2015/227—Procedures used during a speech recognition process, e.g. man-machine dialogue using non-speech characteristics of the speaker; Human-factor methodology
Abstract
The present invention provides information processing unit, its interactive method and the storage medium for being able to carry out human-computer dialogue processing.The information processing unit includes: input information receiving unit, receives the language message of user's input, as input information；Feedback information and characteristic information obtaining portion, it receives input information in response to input information receiving unit, industry service in recalls information processing unit, and for each candidate field relevant to the industry service of calling, the characteristic information of feedback information and each candidate field of the input information in each candidate field is generated；And output field determining section, characteristic information based on each candidate field that feedback information and characteristic information obtaining portion extract, each candidate field is ranked up, and the optimal candidate field that will sort is determined as output field, wherein, the feedback information includes feedback language message and/or feedback action information.
Description
Technical field
The present invention relates to the information processing units for being able to carry out human-computer dialogue processing, and the information processing unit, its is man-machine right
Words method and storage medium.
Background technique
Interactive is a research direction of artificial intelligence field, and generally, human-computer dialogue is exactly to allow user
It can be interacted by the language (i.e. natural language) of the mankind with computer.In general, human-computer dialogue can be with according to interactive mode
It is divided into following four classes: dialogue (more wheels), question and answer (single-wheel) and the recommendation that open field chat, task based access control drive.
Wherein more wheels dialogue of task based access control driving refers to that user inputs information and contains specific purpose, is met with expectation
The information of specific restrictive condition or service etc., such as: making a reservation, book tickets, finding music, film or certain commodity etc..Because with
The demand at family may be more complicated, it may be necessary to a point more wheels are stated, user may also in dialog procedure constantly modification or
Improve the demand of oneself.In addition, when the demand of the statement of user is not specific enough or clear, interactive system can also be with
User is helped to find satisfied result by inquiry, clarification or confirmation.
The man-machine dialogue system of traditional task based access control driving generally includes the processing of following three phases:
User is inputted voice or text parses by natural language understanding (SLU), is converted to dialogue movement (Dialog
Acts internal representation)；
Dialogue state tracks (DST), according to the user session behavior of input and the context status information of user, determines
Ownership goal generates system acting corresponding with user action；And
Spatial term (NLG), the sentence of natural language is converted to according to the semantic expressiveness of system acting, feeds back to use
Family.
Above-mentioned traditional man-machine interactive system is able to solve the task interaction problems in a certain field, but in interactive system
When needing to undertake multiple-task (or technical ability) of different field, existing technical solution (such as publication No. is CN105068661A
Chinese patent application) be first to user input carry out intent classifier, choose be intended to the highest field technical ability of confidence score.But
Be, the input semanteme handled for semantic ambiguity or field itself relatively in the case where, traditional man-machine interactive system
Due to can not accurately select field, lead to not obtain proper reply.In addition, needing needle when newly-increased field or technical ability
Model training is re-started to intent classifier with increase the intent classifier of frontier/with increase correspond to frontier expert
System leads to the poor expandability of system.
Summary of the invention
In view of the above problems, the present invention is intended to provide a kind of be able to carry out with the field choice accuracy improved and can expand
The information processing unit of the human-computer dialogue processing of malleability, the interactive method and Jie's storage media of the information processing unit.
The first aspect of the present invention provides a kind of information processing unit for being able to carry out human-computer dialogue processing, at the information
Reason device includes: input information receiving unit, receives the language message of user's input, as input information；Feedback information and spy
Information obtaining section is levied, receives input information in response to input information receiving unit, the field clothes in recalls information processing unit
Business, and for each candidate field relevant to the industry service of calling, generate feedback letter of the input information in each candidate field
The characteristic information of breath and each candidate field；And output field determining section, it is extracted based on feedback information and characteristic information obtaining portion
Each candidate field characteristic information, each candidate field is ranked up, and the optimal candidate field that will sort is determined as exporting
Field, wherein the feedback information includes feedback language message and/or feedback action information.
By the industry service in recalls information processing unit, and based on each candidate neck relevant to the industry service of calling
The characteristic information in domain is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field, thus
It realizes to the Optimal Decision-making of cross-domain dialogue, the output for inputting high matching degree with user can be obtained, not only have and consider for a long time
The movement selective advantage of return, and the more robust plan for allowing clarification and determination to remove ASR and SLU mistake can be generated
Slightly.
Further, since the supported field clothes in feedback language message and characteristic information obtaining portion recalls information processing unit
Business, such as can call all industry services, therefore even if in the case where newly-increased industry service, can also call in time from
And task-driven conversational system is improved to the scalability of newly-increased technical ability.
As a preferred option, industry service call unit, the industry service supported in recalls information processing unit, with
Determine the relevant each candidate field of the industry service institute called；Human-computer dialogue processing unit, for each candidate field, to input
Information carries out the human-computer dialogue processing of task based access control driving to generate the feedback information in each candidate field；And domain features extract
Unit extracts the characteristic information in each candidate field for each candidate field.
As a preferred option, the domain features extraction unit extracts the feature letter in each field in the way of the grouping of field
Breath.As a preferred option, the domain features extraction unit is grouped field by characteristic similarity, in each grouping
Field limit field respectively and need the characteristic dimension that returns, and the field sharing feature power in each grouping in the training process
Weight.
As a preferred option, the human-computer dialogue processing unit further comprises: natural language understanding unit, to work
Input information for the language message of user's input parses, and the movement for being converted to user session indicates；Dialogue state with
The movement of track unit, the user session obtained according to natural language understanding unit indicates and the context state of input information is believed
Breath determines ownership goal, and generates system acting information corresponding with user session, as feedback action information；Natural language
Generation unit generates natural language information according to the semantic expressiveness of system acting, as feedback language message.
As a preferred option, the characteristic information that domain features extraction unit extracts includes: user's calling in user's portrait
Each field preference；And the human-computer dialogue that carries out of industry service call unit handle the processing result of each step with it is defeated
Enter the correlation of information.
By extracting the characteristic information of reaction user preference program, information is inputted for ambiguous user, can will be used
The technical ability of family more preference is supplied to user, and by extracting correlative character, information is inputted for ambiguous user, can be incited somebody to action
Higher-quality technical ability is supplied to user.
In addition, information processing unit further include: output section, by input information output field feedback language message,
Clarification/confirmation message for prompting user to clarify or confirm, and/or information will be inputted in the feedback action in output field
Information output.
According to a second aspect of the present invention, information processing unit according to a first aspect of the present invention can also include that user is intended to
The candidate user of identification part, identification input information is intended to, and obtains the confidence level that each candidate user is intended to, and industry service
Call unit, which is called, reaches the relevant industry service of the candidate user intention of predetermined threshold to confidence level.
By reaching the limitation of predetermined threshold to confidence level, the precision of field selection can be improved, it can using confidence level
The calling of unrelated technical ability is reduced, to reduce the performance cost of system.
Scheme as an alternative, the feedback information and domain features obtaining portion also call the industry service of specific area,
The specific area includes bad intention assessment efficiency, accuracy rate and the low field of recall rate and/or does not carry out intention knowledge
Other field.
As a preferred option, the characteristic information that the domain features extraction unit extracts further include in following two extremely
One item missing: each candidate user is intended to the confidence level for each candidate field；Preference of the user to each candidate field in user's portrait
Degree.
According to a third aspect of the present invention, a kind of interactive method for information processing unit is provided, this method includes
Following steps: input information receiving step receives the language message of user's input, as input information；Feedback information and spy
Levy information acquisition step, in response to receiving input information in input information receiving step, call industry service, and for
The relevant each candidate field of the industry service of calling generates feedback information and each candidate neck of the input information in each candidate field
The characteristic information in domain；And output field determines step, obtains each time extracted in step based on feedback information and characteristic information
The characteristic information for selecting field is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field,
In, the feedback information includes feedback language message and/or feedback action information.
As a preferred option, in interactive method according to a third aspect of the present invention, feedback information and feature letter
It includes: industry service invocation step that breath, which obtains step, and the industry service supported in recalls information processing unit is called with determining
The relevant each candidate field of industry service institute；Human-computer dialogue processing step carries out input information for each candidate field
The human-computer dialogue of task based access control driving is handled to generate the feedback information in each candidate field；And domain features extraction step,
For each candidate field, the characteristic information in each candidate field is extracted.
As a preferred option, in interactive method according to a third aspect of the present invention, the domain features are extracted
Step extracts the characteristic information in each field in such a way that field is grouped.
According to a fourth aspect of the present invention, a kind of computer readable storage medium is provided, processor can be made by being stored thereon with
Execute the computer program of interactive method as described in the third aspect of the present invention.
In summary, according to the technique and scheme of the present invention, it at least can be realized following effect: 1) improving task-driven pair
Scalability of the telephone system to newly-increased technical ability；2) information is inputted for ambiguous user, it is intended to by higher-quality technical ability
It is supplied to user；3) information is inputted for ambiguous user, it is intended to which the technical ability of user's more preference is supplied to user.
Detailed description of the invention
Fig. 1 is the schematic diagram for showing the hardware construction of information processing unit 1000 according to the present invention；
Fig. 2 shows the use environment schematic diagrames of information processing unit according to the present invention；
Fig. 3 shows the flow chart of the interactive method of first embodiment according to the present invention；
Fig. 4 shows feedback language message and characteristic information in the interactive method of first embodiment according to the present invention
Obtain the flow chart of processing；
Fig. 5 shows the flow chart of human-computer dialogue processing according to the present invention；
Fig. 6 shows the modular structure of the information processing unit of first embodiment according to the present invention；
Fig. 7 shows the flow chart of the interactive method of second embodiment according to the present invention；
Fig. 8 shows the modular structure of the information processing unit 1100 of second embodiment according to the present invention.
Specific embodiment
Now, the exemplary embodiment that present invention will be described in detail with reference to the accompanying.It should be pointed out that unless specifically stated otherwise,
The relative configuration of the component, digital representation and the numerical value that describe in these embodiments does not limit the scope of the invention.
[term definition]
It is of the invention for ease of understanding, term used herein is carried out as described below.
Industry service (also referred to as technical ability) refers to service provided by module relevant to dialogue field, can be web application
The service of offer is also possible to the service that the application software of such as APP provides.
[hardware configuration of information processing unit]
Fig. 1 is the figure for showing the hardware construction of the information processing unit 1000 in the present embodiment.In the present embodiment, with intelligence
Energy phone provides description as the example of information processing unit.Although it is noted that illustrating smart phone in the present embodiment
As information processing unit 1000, but it is clear that without being limited thereto, information processing unit of the invention can be mobile terminal (intelligence
Mobile phone, smartwatch, Intelligent bracelet), intelligent sound box, personal computer, laptop, tablet computer, PDA (individual digital
Assistant), robot or server etc. have the function of the information processing function comprising natural language processing and good in interactive function
The various devices such as internet device (such as digital camera, refrigerator, television set etc.).
Firstly, the hardware configuration of block diagram description information processing unit 1000 referring to Fig.1.In addition, making in the present embodiment
Following construction is described for example, but information processing unit of the invention is not limited to construction shown in FIG. 1.
Information processing unit 1000 includes the input interface 102 being connected to each other via system bus, CPU 103, ROM
104, RAM 105, storage device 106, output interface 107, display 108, communication unit 109 and short-range wireless communication unit
110.Input interface 102 is the interface for receiving voice messaging and/or text information that user is inputted, and is to use
Connecing for input information is inputted using the operating unit (not shown) of microphone, keyboard or touch screen etc. in receiving user
Mouthful.Note that the display 108 being described later on and operating unit can be at least partly integrated, also, for example, it may be
Picture output is carried out in same picture and receives the construction of user's operation.
CPU 103 is system control unit, and generally comprehensively controls information processing unit 1000.In addition, for example,
CPU 103 carries out the display control of the display 108 of information processing unit 1000.ROM 104 stores CPU 103 and executes such as
The fixed data of tables of data, control program and operating system (OS) program etc..In the present embodiment, it is stored in ROM 104
Each control program, for example, such as being dispatched under the management of the OS stored in ROM 104, task switches and interrupt processing
Deng software execute control.
RAM 105 is constructed such as SRAM (static random access memory), DRAM as needing backup power source.This
In the case of, RAM 105 can store the significant data of control variable of program etc. in a non-volatile manner.In addition, for depositing
The storage region for storing up setting information, management data of information processing unit 1000 of information processing unit 1000 etc. is also disposed in
In RAM 105.In addition, RAM 105 is used as the working storage and main memory of CPU 103.
Storage device 106 stores such as predefined dictionary, corpus, model library and application program, for executing according to this
The application program etc. of the man-machine dialogue system method of invention.In addition, the storage of storage device 106 is such as via communication unit
109 send/received information transmission/receiving control program etc. with other devices (such as server 2000 in Fig. 2)
The various information that various programs and these programs use.
Output interface 107 is for being controlled display 108 to show or broadcast information or/and using journey
The display picture of sequence and/or the interface of sound.Output equipment includes such as LCD (liquid crystal display) or OLED (organic electroluminescence hair
Electric display) display 108 and loudspeaking sound device (not shown) etc..There is such as numerical value by arranging on a display device 108
The soft keyboard of the key of enter key, mode setting button, decision key, cancel key and power key etc. also can receive via display 108
It is from the user such as text information input, operational order.
Information processing unit 1000 is via communication unit 109 for example, by channel radios such as Wi-Fi (Wireless Fidelity) or bluetooth
Letter method executes data communication with external device (ED) (not shown).
In addition, information processing unit 1000 can also via short-range wireless communication unit 110, in short-range with
External device (ED) etc. is wirelessly connected and executes data communication.And short-range wireless communication unit 110 by with communication unit
109 different communication means are communicated.It is, for example, possible to use its communication range is shorter than the communication means of communication unit 109
Communication means of the Bluetooth Low Energy (BLE) as short-range wireless communication unit 110.In addition, as short-distance wireless communication list
The communication means of member 110, for example, it is also possible to use the communication parties such as NFC (near-field communication) or Wi-Fi perception (Wi-Fi Aware)
Method.
Fig. 2 shows apparatus for management of information use environment schematic diagrames of the invention.As shown in Fig. 2, such as server or individual
Multiple information processings such as multiple apparatus for management of information (1000,1100) of computer and mobile terminal as user terminal
Device (2000,2100) can communicate each other via network 3000.It in this example, will be at apparatus for management of information and information
Reason device is shown as 2, it is clear that the above quantity is merely illustrative, and it can be one or more that quantity, which is also distinguished,.
Information processing unit 1000/1100 is user using come the subscriber terminal equipment that is operated, subscriber terminal equipment
The various application programs of upper storage can provide the service in each field for user, fill in addition, user can also be handled with operation information
It sets and receives the service provided from apparatus for management of information via network.
Network 3000 can be the inside of public network, entity, mechanism and/or tissue of internet etc etc.
Network, dedicated network and/or public network etc..Network 3000 can be any kind of cable network, wireless network and two
The combination of person, including but not limited to telecommunication network (such as local area network (LAN), wide area network (WAN), satellite network, cable network,
Mobile communications network (such as 2G, 3G, 4G, 5G), short range network (such as Wi-Fi network, WiMax network, bluetooth etc.) are at least
The combination of one.Network 3000 can utilize communication protocol, such as internet including agreement packet-based and/or based on datagram
Agreement (IP), transmission control protocol (TCP), User Datagram Protocol (UDP) or other kinds of agreement etc..
[first embodiment]
[interactive method of first embodiment according to the present invention]
It is said referring to interactive method of the Fig. 3 to the task based access control driving of first embodiment according to the present invention
It is bright.
As shown in figure 3, such as microphone of the information processing unit 1000 of user's operation such as smart phone or integrated
The language message of operating unit the input such as voice or text of touch screen over the display etc. receives step in input information
In rapid S110, information processing unit 1000 via the input interface 102 being connect with operating unit receive voice that user inputs or
The language message of person's text, as input information.Herein if user's input is voice messaging, voice messaging can be converted
For text information.
Next, being obtained in step S120 into feedback information and characteristic information, in response in input information receiving step
S110 receives input information, calls industry service, and for each candidate field relevant to the industry service of calling, generates
The feedback language message of information and the characteristic information in each candidate field are inputted in each candidate field.Wherein, the feedback information packet
It includes feedback language message and/or such as indicates the feedback action information of the system acting of information processing unit 1000.
Step S120 is obtained to feedback language message and characteristic information referring to Fig. 4 to be described in detail.Firstly, leading
Domain service call step S1210, the industry service supported in recalls information processing unit, to determine the industry service institute phase called
Each candidate field closed.Specifically, by the routine interface of the application program in recalls information processing unit 1000, to determine
The service that the application program called is capable of providing is related to which field, and these fields are determined as candidate field.
As a kind of basic scheme for improving scalability, in industry service invocation step S1210, recalls information processing
The all spectra service supported in device 1000.Even if new application is increased in information processing unit 1000 in the case
Program, and include the service for being related to frontier, the neck for still calling this newly-increased in the service of new application offer
Domain service.As a kind of alternative mode, it is clear that can also be serviced with the certain fields supported in recalls information processing unit 1000.
Then, into human-computer dialogue processing step S1220, for each candidate field, input information is based on
The human-computer dialogue of task-driven is handled to generate the feedback information in each candidate field；And in domain features extraction step S1230
In, for each candidate field, extract the characteristic information in each candidate field.
Wherein, in order to further realize the scalability in field, the spy in each field can be extracted in such a way that field is grouped
Reference breath.Specifically, being grouped by characteristic similarity to field, field is limited to the field in each grouping respectively and needs to return
The characteristic dimension returned, and the field sharing feature weight in each grouping in the training process.
Extracted characteristic information may include: the preference in each field that user calls in user's portrait；And neck
The human-computer dialogue that domain service call unit carries out handles the correlation of the processing result and input information of each step.For example, calculating
Score (such as the weighting confidence of the SLOT obtained during SLU of each stage output of SLU, DST, NLG and user's input correlation
Spend score).
Referring to Fig. 5, human-computer dialogue processing step S1220 is described in detail.As shown in figure 5, in natural language
Understand (SLU) step S12101, the input information of the language message inputted as user is parsed, and be converted to user couple
The movement of words indicates；In dialogue state tracking step (DST) S12103, according to what is obtained in natural language understanding step S12101
The movement of user session indicates and the context status information of input information, determines ownership goal, and generation and user session
Corresponding system acting information, as feedback action information；Then, dynamic according to system in spatial term step S12105
The semantic expressiveness of work generates natural language information, as feedback language message.
Next, being determined in step S130 in output field, extracted based on feedback information and characteristic information obtaining portion each
The characteristic information in candidate field is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field.
Specifically, to the field of all candidates, carry out model sequence, the model that can be used for example be GBDT, learn2rank or
Other regression models etc..
In addition, it can include output step S140, at least one of in the following manner leads input information in output
The feedback information in domain is exported via output interface 107 on the output equipment of display 108 and/or loudspeaking sound device: by feedback letter
Breath display or voice play to user；Prompt user clarifies or confirms to ambiguous input information further progress；It will be defeated
Enter information output field feedback action information exported, such as output to information processing unit central controller, with
Such as execute feedback action (such as broadcasting music, video or starting other application etc.).
In the case that output is for prompting clarification/confirmation message of user's clarification or confirmation in exporting step S140,
Return step S110 receives the input information of user's next round dialogue.
Otherwise, in the case where such as in exporting step S140 exporting feedback action information, feedback action is performed and locates
Reason terminates.
[information processing unit of first embodiment according to the present invention]
It is illustrated referring to modular structure of the Fig. 6 to the information processing unit of first embodiment according to the present invention.
As shown in fig. 6, the information processing unit includes: input information receiving unit 110, the language message of user's input is received, as
Input information；Feedback information and characteristic information obtaining portion 120 receive input information in response to input information receiving unit, adjust
With industry service, and for each candidate field relevant to the industry service of calling, input information is generated in each candidate field
Feedback information and each candidate field characteristic information；And output field determining section 130, it is based on feedback information and characteristic information
The characteristic information in each candidate field that obtaining portion is extracted is ranked up each candidate field, and by optimal candidate field of sorting
It is determined as output field, wherein the feedback information includes feedback language message and/or feedback action information.Further, it is also possible to
Including output section 140, information will be inputted in the feedback language message in output field, and/or prompt user's clarification/confirmation
Clarification/confirmation message, and/or by input information output field feedback action information export.
Wherein, feedback information and characteristic information obtaining portion include: industry service call unit 1210, recalls information processing
The industry service supported in device, to determine the relevant each candidate field of the industry service called institute；Human-computer dialogue processing unit
1220, for each candidate field, the human-computer dialogue processing of task based access control driving is carried out to generate each candidate neck to input information
The feedback information in domain；And domain features extraction unit 1230 extracts the feature letter in each candidate field for each candidate field
Breath.
Wherein, human-computer dialogue processing unit 1220 further comprises: natural language understanding unit 12101, to as with
The input information of the language message of family input is parsed, and the movement for being converted to user session indicates；Dialogue state tracking is single
The movement of member 12103, the user session obtained according to natural language understanding unit indicates and the context state of input information
Information determines ownership goal, and generates system acting information corresponding with user session, as feedback action information；Natural language
It says generation unit 12105, natural language information is generated according to the semantic expressiveness of system acting, as feedback language message to feed back
To user.
In summary, the technical solution of first embodiment according to the present invention, can be realized following effect: 1) improving and appoint
Scalability of the business driving conversational system to newly-increased technical ability；2) information is inputted for ambiguous user, it is intended to will be more high-quality
The technical ability of amount is supplied to user；3) information is inputted for ambiguous user, it is intended to be supplied to the technical ability of user's more preference
User.
[second embodiment]
[interactive method of first embodiment according to the present invention]
Next, being illustrated referring to Fig. 7 to interactive method according to the present invention.Fig. 7 shows according to the present invention
The flow chart of the interactive method of two embodiments.
For simplicity, be only illustrated to the step of being different from the first embodiment in second embodiment, omit
To the explanation of same steps.
As shown in fig. 7, the difference of second embodiment and first embodiment is, increase user's intention assessment step S115,
The candidate user of identification input information is intended to, and obtains the confidence level that each candidate user is intended to.Then, into feedback information and neck
Characteristic of field obtains step S121, wherein in industry service invocation step S1211, calls and reaches predetermined threshold with confidence level
Candidate user is intended to relevant industry service.Wherein, in industry service invocation step S1211, it may call upon specific area
Industry service, specific area include bad intention assessment efficiency, accuracy rate and the low field of recall rate and/or not into
The field of row intention assessment.By reaching the limitation of predetermined threshold to confidence level, the precision of field selection can be improved, using setting
Reliability can reduce the calling of unrelated technical ability, to reduce the performance cost of system.
Specifically, in step sl 15, using machine learning (such as SVM) or deep learning method (such as CNN) or mode
Matching process (such as regular expression), the intention for inputting information itself to user identify, obtain multiple candidate users and are intended to,
And obtain the confidence level for being intended to corresponding each field with multiple candidate users.
In order to make it easy to understand, being described in detail below in conjunction with specific example.For example, in step S110, information processing
Device 1000 receives the information " playing Les Miserables " of user's input, and processing enters user intention assessment step S115, knows
Not Shu Ru the candidate user of information be intended to, and obtain the confidence level that each candidate user is intended to.The confidence level that candidate user is intended to can
To be used to determine whether to call related fields service.According to for example, (Intent:{ music:0.8, audio:0.7,
Knowledge:0.5, weather:0.1 }.Wherein, for the field video, although being not intended to classifier, as specific
Still it calls in field.
Step S121 is obtained subsequently into feedback information and characteristic information, specifically, in industry service invocation step
In S1211, if threshold value is set as 0.5, it is determined that candidate field music (music), audio (audio) and knowledge
(knowledge) field, and video (video) is also used as specific area, it is determined as candidate field.Then it is handled in human-computer dialogue
In step S1220, for each candidate field, according to respective business characteristic, complete independently SLU- > DST- > NLG process.
Such as audio is serviced, the query (inquiry) of upper 2 wheel is " Q1: play a nursery rhymes Q2: playing opera ", this
Wheel receives " Q3: playing Les Miserables ", then SLU:(plays the Les Miserables [pattern] [title]) ([title=is sad by -> DST
Miserable world opera version, 1.0], [category=nursery rhymes, 0.5], [category=opera, 0.8]) and ([title=is sad by OFFER
The miserable world, 1.0], [category=opera, 0.8]) and -> NLG (playing opera Les Miserables for you).
In addition, in domain features extraction step S1230 extract in human-computer dialogue processing step S1220 SLU- > DST- >
The characteristic information that the follow-up decision collected during NLG needs.Moreover it is preferred that the institute in domain features extraction step S1230
The characteristic information of extraction can also include at least one in following two: each candidate user is intended to setting for each candidate field
Reliability；Preference of the user to each candidate field in user's portrait.
Specifically, the popular journey of the slot position confidence level of update, the matching degree of resource and slot position information, resource can be extracted
The information such as degree are as characteristic information.Such as above-mentioned example extraction text feature (word feature, term vector etc.), SLU stage are known
The weighting point of other SLOT, DST phase resource and the matching degree of user query, the temperature of resource, dialogue state (dialogue
State) and dialogue acts (dialog act) etc., as characteristic information.
Equally, corresponding human-computer dialogue processing result and spy can also be generated for the candidate field such as music/knowledge
Reference breath.
Then, it is determined in step S130 in output field, uses these characteristic informations as decision-making foundation, to each candidate neck
Domain is ranked up.Specifically, for each candidate field, the characteristic information based on return generates corresponding candidate field respectively
Feature vector completes model prediction and sequence.Disaggregated model (BOOST, LR, SVM etc.) can be used, mould is calculated separately to each field
Type score selects the candidate field of highest scoring, as optimal field.Sort method can also be learnt using Lambdarank etc.
Directly the feature of different field is ranked up, and most preceding candidate field of sorting, is determined as output field.
By above step, the precision of final field selection (output field determines) can be further increased, confidence is utilized
Degree can reduce the calling of unrelated technical ability (industry service).
[information processing unit of second embodiment according to the present invention]
It is said referring to modular structure of the Fig. 8 to the information processing unit 1100 of first embodiment according to the present invention
It is bright.For simplicity, be only illustrated to the step of being different from the first embodiment in second embodiment, omit to identical
The explanation of module.
Referring to Fig. 8, the difference of second embodiment and first embodiment is, the information processing of second embodiment of the invention
Device 1100 increases user's intention assessment portion 115, and the candidate user of identification input information is intended to, and obtains each candidate user meaning
The confidence level of figure.And in feedback information and domain features obtaining unit 121, industry service call unit 1211 is called and is set
The candidate user that reliability reaches predetermined threshold is intended to relevant industry service.Wherein, industry service call unit 1211 can be with
Call the industry service of specific area, specific area include bad intention assessment efficiency, accuracy rate and the low field of recall rate, with
And/or person does not carry out the field of intention assessment.
As a result, in second embodiment of the invention on the basis of the technical effect of first embodiment, also by confidence
Degree reaches the limitation of predetermined threshold, can be improved the precision of field selection, the calling of unrelated technical ability can be reduced using confidence level,
To reduce the performance cost of system.
To sum up, the technical solution of first embodiment of the invention and second embodiment can not only be realized to cross-domain dialogue
Optimal Decision-making, the output that high matching degree is inputted with user can be obtained, not only have consider return for a long time movement selection
Advantage, and the more robust strategy for allowing clarification and determination to remove ASR and SLU mistake can be generated.
Wherein, present invention further provide that
A, a kind of information processing unit for the human-computer dialogue processing for being able to carry out task based access control driving, the information processing unit
Include:
Information receiving unit is inputted, the language message of user's input is received, as input information；
Feedback information and characteristic information obtaining portion receive input information in response to input information receiving unit, call neck
Domain service, and for each candidate field relevant to the industry service of calling, it is anti-in each candidate field to generate input information
The characteristic information of feedforward information and each candidate field；And
Output field determining section, the feature letter based on each candidate field that feedback information and characteristic information obtaining portion extract
Breath is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field,
Wherein, the feedback information includes feedback language message and/or feedback action information.
A2. according to the information processing unit of A1, wherein the feedback information and characteristic information obtaining portion include:
Industry service call unit, the industry service supported in recalls information processing unit, to determine the field called
The relevant each candidate field of service institute；
Human-computer dialogue processing unit carries out the man-machine right of task based access control driving to input information for each candidate field
Words are handled to generate the feedback information in each candidate field；And
Domain features extraction unit extracts the characteristic information in each candidate field for each candidate field.
A3. according to the information processing unit of A2, wherein the domain features extraction unit extracts in such a way that field is grouped
The characteristic information in each field.
A4. according to the information processing unit of A3, wherein the domain features extraction unit, by characteristic similarity to field
It is grouped, the characteristic dimension that field needs to return, and each point in the training process is limited respectively to the field in each grouping
Field sharing feature weight in group.
A5. according to the information processing unit of A2, wherein the human-computer dialogue processing unit further comprises:
Natural language understanding unit parses the input information of the language message inputted as user, and converts
It is indicated for the movement of user session；
The movement of dialogue state tracking cell, the user session obtained according to natural language understanding unit indicates and input
The context status information of information determines ownership goal, and generates system acting information corresponding with user session, as feedback
Action message；
Spatial term unit generates natural language information according to the semantic expressiveness of system acting, as feedback language
Information.
A6. the information processing unit according to A2 or A3, wherein the industry service call unit recalls information processing
The all spectra service supported in device.
A7. the information processing unit according to A2 or A3, wherein the feature letter that the domain features extraction unit extracts
Breath includes:
The preference in each field that user calls in user's portrait；And
The human-computer dialogue that industry service call unit carries out handles the correlation of the processing result and input information of each step.
A8. the information processing unit according to any one of A2 or A3, wherein the information processing unit further include:
Output section will input information in the feedback language message in output field, and output is used to user and/or prompt
Family clarification or confirmation, and/or feedback action information of the information in output field will be inputted and exported.
A9. the information processing unit according to A2 or 3, wherein the information processing unit further includes that user is intended to know
The candidate user in other portion, identification input information is intended to, and obtains the confidence level that each candidate user is intended to, and
Industry service call unit, which is called, reaches the relevant industry service of the candidate user intention of predetermined threshold to confidence level.
A10. the information processing unit according to A9, wherein the feedback information and domain features obtaining portion are also called
The industry service of specific area, the specific area include bad intention assessment efficiency, accuracy rate and the low field of recall rate, with
And/or person does not carry out the field of intention assessment.
A11. the information processing unit according to A10, wherein the characteristic information that the domain features extraction unit extracts
Further include at least one in following two:
Each candidate user is intended to the confidence level for each candidate field；
Preference of the user to each candidate field in user's portrait.
B12. a kind of interactive method that the task based access control for information processing unit drives, this method includes following step
It is rapid:
Information receiving step is inputted, the language message of user's input is received, as input information；
Feedback information and characteristic information obtain step, in response to receiving input information in input information receiving step,
Industry service is called, and for each candidate field relevant to the industry service of calling, generates input information in each candidate field
In feedback information and each candidate field characteristic information；And
Output field determines step, and the spy in each candidate field extracted in step is obtained based on feedback information and characteristic information
Reference breath is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field,
Wherein, the feedback information includes feedback language message and/or feedback action information.
C13. a kind of computer readable storage medium, the storage medium include the program of storage, wherein in described program
Equipment where controlling the storage medium when operation is executed and is driven described in above-mentioned B12 for the task based access control of information processing unit
Interactive method.
D14, a kind of electronic equipment, the electronic equipment include: one or more processors；
Memory, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes the interactive method driven described in above-mentioned B12 for the task based access control of information processing unit.
Claims (10)
1. a kind of information processing unit for being able to carry out human-computer dialogue processing, the information processing unit include:
Information receiving unit is inputted, the language message of user's input is received, as input information；
Feedback information and characteristic information obtaining portion receive input information in response to input information receiving unit, call field clothes
Business, and for each candidate field relevant to the industry service of calling, generate feedback letter of the input information in each candidate field
The characteristic information of breath and each candidate field；And
Output field determining section, the characteristic information based on each candidate field that feedback information and characteristic information obtaining portion extract are right
Each candidate field is ranked up, and the optimal candidate field that will sort is determined as output field,
Wherein, the feedback information includes feedback language message and/or feedback action information.
2. information processing unit according to claim 1, wherein the feedback information and characteristic information obtaining portion include:
Industry service call unit, the industry service supported in recalls information processing unit, to determine the industry service called
The relevant each candidate field of institute；
Human-computer dialogue processing unit carries out at the human-computer dialogue of task based access control driving input information for each candidate field
It manages to generate the feedback information in each candidate field；And
Domain features extraction unit extracts the characteristic information in each candidate field for each candidate field.
3. information processing unit according to claim 2, wherein the domain features extraction unit mentions in such a way that field is grouped
Take the characteristic information in each field.
4. information processing unit according to claim 3, wherein the domain features extraction unit, by characteristic similarity to neck
Domain is grouped, and limits the characteristic dimension that field needs to return respectively to the field in each grouping, and each in the training process
Field sharing feature weight in grouping.
5. information processing unit according to claim 2, wherein the human-computer dialogue processing unit further comprises:
Natural language understanding unit parses the input information of the language message inputted as user, and is converted to use
The movement of family dialogue indicates；
The movement of dialogue state tracking cell, the user session obtained according to natural language understanding unit indicates and input information
Context status information, ownership goal is determined, and generate system acting information corresponding with user session, as feedback action
Information；
Spatial term unit, according to the semantic expressiveness of system acting generate natural language information, as feedback language message,
Wherein, the feedback information further includes clarification/confirmation message.
6. information processing unit according to claim 2 or 3, wherein at the industry service call unit recalls information
The all spectra service supported in reason device.
7. information processing unit according to claim 2 or 3, wherein the feature that the domain features extraction unit extracts
Information includes:
The preference in each field that user calls in user's portrait；And
The human-computer dialogue that industry service call unit carries out handles the correlation of the processing result and input information of each step.
8. a kind of interactive method for information processing unit, method includes the following steps:
Information receiving step is inputted, the language message of user's input is received, as input information；
Feedback information and characteristic information obtain step, in response to receiving input information in input information receiving step, call
Industry service, and for each candidate field relevant to the industry service of calling, input information is generated in each candidate field
The characteristic information of feedback information and each candidate field；And
Output field determines step, and the feature letter in each candidate field extracted in step is obtained based on feedback information and characteristic information
Breath is ranked up each candidate field, and the optimal candidate field that will sort is determined as output field,
Wherein, the feedback information includes feedback language message and/or feedback action information.
9. a kind of computer readable storage medium, the storage medium includes the program of storage, wherein in described program operation
Equipment where controlling the storage medium is executed and is driven described in the claims 8 for the task based access control of information processing unit
Interactive method.
10. a kind of electronic equipment, which is characterized in that the electronic equipment includes: one or more processors；
Memory, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The interactive method now driven as claimed in claim 8 for the task based access control of information processing unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811092347.8A CN109272999B (en) | 2018-09-19 | 2018-09-19 | Information processing unit, its interactive method and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811092347.8A CN109272999B (en) | 2018-09-19 | 2018-09-19 | Information processing unit, its interactive method and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109272999A true CN109272999A (en) | 2019-01-25 |
| CN109272999B CN109272999B (en) | 2019-08-16 |
Family
ID=65197044
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811092347.8A Active CN109272999B (en) | 2018-09-19 | 2018-09-19 | Information processing unit, its interactive method and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109272999B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111737423A (en) * | 2020-08-26 | 2020-10-02 | 北京声智科技有限公司 | Domain identification method and device, electronic equipment and storage medium |
| WO2022089546A1 (en) * | 2020-10-28 | 2022-05-05 | 华为云计算技术有限公司 | Label generation method and apparatus, and related device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160110347A1 (en) * | 2014-10-15 | 2016-04-21 | Voicebox Technologies Corporation | System and method for providing follow-up responses to prior natural language inputs of a user |
| CN106960005A (en) * | 2017-02-20 | 2017-07-18 | 北京光年无限科技有限公司 | A kind of output intent and robot for artificial intelligence robot |
| US20170359707A1 (en) * | 2016-06-08 | 2017-12-14 | Google Inc. | Providing a personal assistant module with a selectively-traversable state machine |
| CN107589828A (en) * | 2016-07-07 | 2018-01-16 | 深圳狗尾草智能科技有限公司 | The man-machine interaction method and system of knowledge based collection of illustrative plates |
| CN108399169A (en) * | 2017-02-06 | 2018-08-14 | 阿里巴巴集团控股有限公司 | Dialog process methods, devices and systems based on question answering system and mobile device |
-
2018
- 2018-09-19 CN CN201811092347.8A patent/CN109272999B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160110347A1 (en) * | 2014-10-15 | 2016-04-21 | Voicebox Technologies Corporation | System and method for providing follow-up responses to prior natural language inputs of a user |
| US20170359707A1 (en) * | 2016-06-08 | 2017-12-14 | Google Inc. | Providing a personal assistant module with a selectively-traversable state machine |
| CN107589828A (en) * | 2016-07-07 | 2018-01-16 | 深圳狗尾草智能科技有限公司 | The man-machine interaction method and system of knowledge based collection of illustrative plates |
| CN108399169A (en) * | 2017-02-06 | 2018-08-14 | 阿里巴巴集团控股有限公司 | Dialog process methods, devices and systems based on question answering system and mobile device |
| CN106960005A (en) * | 2017-02-20 | 2017-07-18 | 北京光年无限科技有限公司 | A kind of output intent and robot for artificial intelligence robot |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111737423A (en) * | 2020-08-26 | 2020-10-02 | 北京声智科技有限公司 | Domain identification method and device, electronic equipment and storage medium |
| WO2022089546A1 (en) * | 2020-10-28 | 2022-05-05 | 华为云计算技术有限公司 | Label generation method and apparatus, and related device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109272999B (en) | 2019-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6873333B2 (en) | Method using voice recognition system and voice recognition system | |
| US11030412B2 (en) | System and method for chatbot conversation construction and management | |
| CN103761242A (en) | Indexing method, indexing system and natural language understanding system | |
| CN103279508A (en) | Method for voice response correction and natural language conversational system | |
| CN103280218A (en) | Voice recognition-based selection method and mobile terminal device and information system thereof | |
| CN107291690A (en) | Punctuate adding method and device, the device added for punctuate | |
| CN103268315A (en) | Natural language conservation method and system | |
| CN101233559A (en) | Context-sensitive communication and translation methods for enhanced interactions and understanding among speakers of different languages | |
| CN102292766A (en) | Method, apparatus and computer program product for providing compound models for speech recognition adaptation | |
| CN112328849A (en) | User portrait construction method, user portrait-based dialogue method and device | |
| CN109272999B (en) | Information processing unit, its interactive method and storage medium | |
| CN108345612A (en) | A kind of question processing method and device, a kind of device for issue handling | |
| CN113139036A (en) | Information interaction method and equipment | |
| CN115668206A (en) | Facilitating parallel hypothesis reasoning for multi-language, multi-turn, multi-domain virtual assistants | |
| CN111128175B (en) | Spoken language dialogue management method and system | |
| KR20180060903A (en) | Electronic device and method thereof for performing translation by sharing context of utterance | |
| CN109783733A (en) | User's portrait generating means and method, information processing unit and storage medium | |
| WO2021139486A1 (en) | Text incrementation method and apparatus, and terminal device | |
| CN113140138A (en) | Interactive teaching method, device, storage medium and electronic equipment | |
| CN110309339A (en) | Picture tag generation method and device, terminal and storage medium | |
| US20220270617A1 (en) | Electronic device for supporting artificial intelligence agent services to talk to users | |
| CN113302628A (en) | Micrometric neural conversational model using domain knowledge | |
| CN113590798A (en) | Dialog intention recognition, training method for model for recognizing dialog intention | |
| CN116052646B (en) | Speech recognition method, device, storage medium and computer equipment | |
| CN116306685A (en) | Multi-intention recognition method and system for power business scene |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200804 Address after: 518000 Nanshan District science and technology zone, Guangdong, Zhejiang Province, science and technology in the Tencent Building on the 1st floor of the 35 layer Patentee after: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd. Address before: 100029, Beijing, Chaoyang District new East Street, building No. 2, -3 to 25, 101, 8, 804 rooms Patentee before: Tricorn (Beijing) Technology Co.,Ltd. |