CN109885668A - A kind of expansible field interactive system status tracking method and apparatus - Google Patents
A kind of expansible field interactive system status tracking method and apparatus Download PDFInfo
- Publication number
- CN109885668A CN109885668A CN201910071447.0A CN201910071447A CN109885668A CN 109885668 A CN109885668 A CN 109885668A CN 201910071447 A CN201910071447 A CN 201910071447A CN 109885668 A CN109885668 A CN 109885668A
- Authority
- CN
- China
- Prior art keywords
- value
- lstm
- slot
- feature
- dialogue
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
Abstract
The present invention provides the expansible field interactive system status tracking method and apparatus of one kind and determines current dialogue states using slot-value model the method includes establishing status tracking model；Training data feature set is split, acquisition each round dialogue, and is acquired to characteristic item therein, is split to each characteristic item；Prepare contextual information；Model training based on machine learning method.Since the status tracking model in this method carries out independent prediction to each feature, without being defined to feature value range, it can be changed with the dynamic of supported feature value range, to not need to re-start the training of language understanding model during upgrading service, so that field interactive system scalability is stronger.
Description
Technical field
The present invention relates to artificial intelligence fields, in particular to interactive system field.
Background technique
With the development of machine learning especially speech recognition and understanding technology, many artificial conversational systems are gradually man-machine
Conversational system replaces.Existing interactive system is broadly divided into two major classes, and one kind is small degree intelligent sound box etc, based on chat
The system of mode, this kind of system achieve the purpose that carry out normal dialog with people by the study to mankind's every-day language；Other one
Class is the similar system of voice customer service, by human-computer dialogue, is reached to the certain type of business service of user.Two class human-computer dialogues
The purpose of design of system is different, and used model and the limitation faced are also different.The interactive system of Chat mode,
Can be using a large amount of dialogue materials in social networks as training data, and purpose only safeguards the smoothly context of dialogue, is not related to
Join specific business；The interactive system of specific area is trained speech recognition modeling in some specific area, is suitable for
The training dataset quantity in some field is simultaneously few, and this kind of system background associated services, needs to make the wish of user
It is specific to determine.
Language understanding model used in current field interactive system generally uses speech recognition-language understanding-
Status tracking-Choice of Countermeasures mode carries out, and status tracking step therein is according to corresponding field business by each state
Value all fix several selectable value, the customization of status tracking process has thus been carried out according to business scenario.In order to improve
Recognition accuracy in the interactive system of field, people improve the accurate of language understanding generally in language understanding model, such as
Patent CN108334496A " human-computer dialogue understanding method and system and relevant device for specific area " in model training,
Additional part-of-speech information is introduced, the part of speech of next input word is predicted using part of speech prediction interval, by semantic tagger, intention
Identification, part of speech predict that three tasks carry out Combined Treatment, and using the semantic information shared between three tasks, it is accurate to have reached identification
The promotion of rate.However realization of the existing research to status tracking, it is still according to the fixed value of the artificial specific characteristic of field scene
Mode.In real system application, due to the addition of upgrading service or new business function, it is often necessary to cause field man-machine
Feature value selectable value is changed in status tracking in conversational system.Due in existing field interactive system to state with
The use of characteristic value is to be used in combination in track, interactional, existing for the change of feature value in response state tracking
Field interactive system needs to carry out re -training to the language understanding model before status tracking, this obviously will cause system
Complexity and scalability in upgrading are deteriorated.
Summary of the invention
To achieve the goals above, the invention provides the following technical scheme: a kind of expansible field interactive system
Status tracking method, it is characterised in that:
Status tracking model is established, current dialogue states are determined using slot-value model；Specifically, value is set
Value range is Ci∈ 1......C, wherein i indicates the possible quantity of the value of value in a specific slot, slot value model
It encloses for Tk∈ 1......N, wherein k indicates slot number, context of dialogue D, then dialogue state tracking is represented by given one
A<slot, value>combined value, so that intersection entropy indicated by following formula is minimum:
Wherein P (y) indicates the distribution function that training data is concentrated,Indicate the distribution letter of prediction result
Number, y indicate the slot-value combined value of prediction.
Training data feature set is split, acquisition each round dialogue, and is acquired to characteristic item therein, to each feature
Item is split；It is adopted using the characteristic item that the SLU feature set creation method that DSTC2 data set carries talks with each round
Collection；For the characteristic item after splitting, respectively to each slot tissue training data set in characteristic item, carries out characteristic item and individually instruct
Practice.Before individually train to characteristic item, to<slot, value>combined value carries out logic discrimination, if this is to<slot,
Value>combined value is very, to be then trained, if not then abandoning this to<slot, value>combined value as true.
Prepare contextual information；Using LSTM and level LSTM model, using different LSTM handle respectively word, sentence and
The input of paragraph rank, and ability is extracted and rebuild using the file characteristics of autocoder detection LSTM.
Each round is talked with, acquires contextual information with the following method:
Feature coding: Es=LSTM is carried out according to current sentence using LSTMsentence(Sentj), j indicates current sentence
Number；
Feature coding: Ed=LSTM is carried out according to the context of current sentence using level LSTMdialogue
[LSTMsentence 1...j-1(Sentj)]；
Feature coding: Ea=is carried out according to the context of current sentence and the movement of corresponding business using LSTM
LSTMDialogueAct 1...K(DAK), wherein K indicates business amount of action, DAKFor k-th Dialogue Action；
Then according to the above method, obtain contextual information is described as D=[Es, Ed, Ea].Based on described < slot,
The preparation of value > training dataset and the contextual information, using machine learning model engage in the dialogue state tracking instruct
Practice, which can be abstracted are as follows:
To realize that the scalability of interactive system is supported.
Model training based on machine learning method, according to the training dataset after being split, and to each feature into
Row is described based on the contextual information of LSTM and level LSTM, forms training dataset to each feature, and then to each feature
It is individually predicted, then stateful combination, is denoted as the status tracking result at current time.
Meanwhile the present invention also provides a kind of electricity using above-mentioned expansible field interactive system status tracking method
Sub- equipment.
Compared with prior art, the beneficial effect of the embodiment of the present invention is: status tracking model provided by the invention is to every
A feature carries out independent prediction, does not limit feature value range, therefore can be changed with the dynamic of the range of supported feature value
Become.This status tracking model for not limiting feature value, does not need to re-start language understanding mould when upgrading service
The training of type, so that field interactive system scalability is stronger.
Detailed description of the invention
It, below will be to use required in embodiment in order to illustrate more clearly of the technical solution of embodiment of the present invention
Attached drawing be briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not to be seen as
It is the restriction to range, it for those of ordinary skill in the art, without creative efforts, can be with root
Other relevant attached drawings are obtained according to these attached drawings.
Fig. 1 is slot-value status tracking model schematic of the present invention.
Fig. 2 is the training process schematic diagram provided in an embodiment of the present invention with two features of A, B.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
The term used in embodiments of the present invention is only to be not intended to be limiting merely for for the purpose of describing particular embodiments
The present invention.In the embodiment of the present invention and the "an" of singular used in the attached claims, " described " and "the"
It is also intended to including most forms, unless the context clearly indicates other meaning, " a variety of " generally comprise at least two, but not
It excludes to include at least one situation.
It should be appreciated that term "and/or" used herein is only a kind of incidence relation for describing affiliated partner, indicate
There may be three kinds of relationships, for example, A and/or B, can indicate: individualism A, exist simultaneously A and B, individualism B these three
Situation.In addition, character "/" herein, typicallys represent the relationship that forward-backward correlation object is a kind of "or".
Depending on context, word as used in this " if ", " if " can be construed to " ... when " or
" when ... " or " in response to determination " or " in response to detection ".Similarly, context is depended on, phrase " if it is determined that " or " such as
Fruit detection (condition or event of statement) " can be construed to " when determining " or " in response to determination " or " when detection (statement
Condition or event) when " or " in response to detection (condition or event of statement) ".
In addition, the step timing in following each method embodiments is only a kind of citing, rather than considered critical.
The preferred embodiment that the invention will now be described in detail with reference to the accompanying drawings.
When the present invention due to operation system according to upgrading in the prior art, the model of feature value in status tracking is changed
It encloses, to need the problem of being trained again to language understanding model, provides a kind of expansible field man-machine state tracking
Model and method carry out independent prediction to each feature, without being defined to feature value range, therefore can be with supported feature
The dynamic of value range changes.
The essence of status tracking is the state slot-value combination safeguarded in conversation procedure, each slot-value group
Close the input parameter that all can serve as later period Tactic selection.As shown in Figure 1, a kind of value combination of all slot can be unique
Determine a kind of session status.
For specific area, slot-value model can determine current dialogue states, it is assumed that value value range is Ci
∈ 1......C, wherein i indicates that the possible quantity of the value of value in a specific slot, slot value range are Tk∈
1......N, wherein k indicates slot number, context of dialogue D, then dialogue state tracking is represented by given one < slot,
Value > combined value, so that intersection entropy indicated by following formula is minimum:
Wherein P (y) indicates the distribution function that training data is concentrated,Indicate the distribution letter of prediction result
Number, y indicate the slot-value combined value of prediction.
The dialogue of collecting training data user's each round, and characteristic item therein is acquired.It here can be using conventional
Session characteristics gathering method, such as the SLU feature set creation method (Word Understanding or the spoken language that are carried using DSTC2 data set
Understanding method).
After characteristic item collects completion, each characteristic item is split, each slot in characteristic item is organized respectively
Training dataset.The mode that this each characteristic item is individually trained is different from original session tracking system, by all characteristic items
Tissue is a feature vector, carries out the mode of associated prediction.In original mode, due to carrying out joint training, needing will be each
Item value is limited within several specific value ranges, and existing way individually trains each feature, it is only necessary to be differentiated
Some<slot, whether value>combination is very.The mode that this training dataset is split, so that training dataset scale
It increases exponentially, but no longer the value range of output result is defined.
This patent contextual information uses LSTM model, and the full name of LSTM is Long Short-Term Memory, it is
One kind of RNN (Recurrent Neural Network).The characteristics of LSTM is due to its design, is highly suitable for clock synchronization ordinal number
According to modeling, to preferably capture the dependence of relatively long distance.The LSTM of level is handled respectively using different LSTM
Word, sentence and the input of paragraph rank, and the file characteristics for detecting using autocoder (autoencoder) LSTM extract with
Reconstruction ability.To every wheel user conversation, all acquisition following context information:
Feature coding: Es=LSTM is carried out according to current sentence using LSTMsentence(Sentj), j indicates current sentence
Number；
Feature coding: Ed=LSTM is carried out according to the context of current sentence using level LSTMdialogue
[LSTMsentence 1...j-1(Sentj)]；
Feature coding: Ea=is carried out according to the context of current sentence and the movement of corresponding business using LSTM
LSTMDialogueAct 1...K(DAK), wherein K indicates business amount of action, DAKFor k-th Dialogue Action；
According to the preparation of above-mentioned contextual information, available contextual information is described as D=[Es, Ed, Ea], then base
In above-mentioned<slot, the preparation of the preparation of value>training dataset and contextual information can use existing machine learning
Model, the track training for the state that conversates, the procedural abstraction are following formula:
Due to not limiting the value range of session status feature, session status feature value model in above-mentioned training process
The variation enclosed, the training data for being only embodied in subsequent machine-learning process are concentrated, do not need to carry out subsequent machine learning model
Update, the scalability so as to field interactive system where providing session status tracking system supports.
According to above-mentioned steps, training dataset is split according to feature, and each feature is carried out based on LSTM and
The contextual information of level LSTM describes, to form training dataset to each feature, and then individually carries out to each feature
Prediction.
Each feature is carried out after individually predicting, stateful combination, as the status tracking result at current time.
In Fig. 2, the training process of two features of A, B is illustrated: training dataset is split by each feature；To each spy
Sign is described with tri- LSTM models of Es, Ed, Ea, thus the training dataset of each feature of component, later with full connection nerve net
Each feature is respectively trained in network, obtains the predicted value of each feature, and the predicted value of all features combines, as dialogue shape
The result of state tracking.
The method that the present invention is individually learnt using session characteristics avoids session tracking process in the interactive system of field
Need to define session characteristics value this problem in advance, to bring bigger scalability to field interactive system.In order to test
Algorithm performance is demonstrate,proved, we are in DSTC2 data set, respectively using the test data for combining Word Understanding and speech understanding to generate
Collection, using full Connection Neural Network learning model, in conjunction with Adam model training optimization method, obtain the session status style of cooking, place,
Price, integrated forecasting accuracy rate are as follows:
| Test set generates model | The style of cooking | Place | Price | It is comprehensive |
| Word Understanding | 84.0 | 88.8 | 91.7 | 70.7 |
| Speech understanding | 78.7 | 90.3 | 91.6 | 67.5 |
As can be seen from the comparison result, the expansible field interactive system session status that this patent proposes tracks mould
Type and method can provide very high session tracking accuracy rate on the basis of feature value is expansible.
In order to be more clear explanation this patent, citing is illustrated.
In the field interactive system involved by " dining room recommendation business ", there is<a style of cooking, place, price>tri- slot,
The corresponding value of these three slot is if it is<Beijing cuisine, five road junctions, 200>, the Choice of Countermeasures scheme of operation system will recommend
User's " office's gas in six layers of Hua Lian shopping center ".
Assuming that existing training dataset are as follows:
" I wants to go to five stutters, 200 yuan of Beijing cuisine per capita ", status tracking result are<Beijing cuisine, five road junctions, 200>
" I wants to go to five stutters, 50 yuan of fast food per capita ", status tracking result are<fast food, five road junctions, 50>
" I want to go to five stutter grilled fish ", status tracking result are<Sichuan cuisine, five road junctions, none>
Training dataset is split, to each slot-value combined prediction, such as to first training data:
" I wants to go to five stutters, 200 yuan of Beijing cuisine per capita ",<style of cooking, Beijing cuisine>result is 1；<the style of cooking, Sichuan cuisine>knot
Fruit is 0；<the style of cooking, Shaanxi wheaten food>result are 0
Same processing is also done to other training datas and other features.
Context of dialogue information is modeled using LSTM model and level LSTM model again, in conjunction with the instruction of above-mentioned generation
Practice data acquisition system, final training dataset is generated to each characteristic dimension.Use (such as the full connection of existing machine learning model
Neural network+Adam model training optimization method), so that it may session status is predicted.
When new user inputs " I wants to go to Fengtai Technology Park and eat the face biangbiang ", for the predicted value of " style of cooking " slot
For " Shaanxi wheaten food ", the predicted value for " place " slot is " Fengtai Technology Park ", and the predicted value for " price " slot is
" none ", thus predict come session status be<Shaanxi wheaten food, Fengtai Technology Park, none>.
Meanwhile the present invention goes back while providing a kind of using above-mentioned expansible field interactive system status tracking model
With the electronic equipment of method.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of expansible field interactive system status tracking method, it is characterised in that:
Status tracking model is established, current dialogue states are determined using slot-value model；
Training data feature set is split, and acquisition each round dialogue and is acquired characteristic item therein, to each characteristic item into
Row is split；
Prepare contextual information；
Model training based on machine learning method.
2. according to the method described in claim 1, it is characterized by:
It is described to establish in status tracking model step, value value range is set as Ci∈ 1......C, wherein i indicates a spy
Determine the possible quantity of value of value in slot, slot value range is Tk∈ 1......N, wherein k indicates slot number, dialogue
Context is D, then dialogue state tracking is represented by given one<slot, value>combined value, so that indicated by following formula
Intersection entropy it is minimum:
Wherein P (y) indicates the distribution function that training data is concentrated,Indicate the distribution function of prediction result, y table
Show the slot-value combined value of prediction.
3. the method according to claim 1, wherein the SLU feature set generation side carried using DSTC2 data set
The characteristic item that method talks with each round is acquired.
4. according to claim 1 or method described in any one of 3, which is characterized in that right respectively for the characteristic item after splitting
Each slot tissue training data set in characteristic item carries out characteristic item and individually trains.
5. according to the method described in claim 4, it is characterized in that, before individually train to characteristic item, to < slot,
Value>combined value carries out logic judgment, if this is to<slot, value>combined value is very, to be then trained, if not such as
Very, then this is abandoned to<slot, value>combined value.
6. making the method according to claim 1, wherein contextual information uses LSTM and level LSTM model
Word, sentence and the input of paragraph rank are handled respectively with different LSTM, and use the file characteristics of autocoder detection LSTM
Extract and rebuild ability.
7. according to the method described in claim 6, it is characterized in that, acquiring context with the following method to each round dialogue
Information:
Feature coding: Es=LSTM is carried out according to current sentence using LSTMsentence(Sentj), j indicates current sentence number；
Feature coding: Ed=LSTM is carried out according to the context of current sentence using level LSTMdialogue[LSTMsentence 1...j-1
(Sentj)]；
Feature coding: Ea=is carried out according to the context of current sentence and the movement of corresponding business using LSTM
LSTMDialogueAct 1...K(DAK), wherein K indicates business amount of action, DAKFor k-th Dialogue Action；
Then according to the above method, obtain contextual information is described as D=[Es, Ed, Ea].
8. method according to any one of claims 1-7, which is characterized in that based on described<slot, value>training number
According to the preparation of collection and the contextual information, engaged in the dialogue the track training of state using machine learning model, which can table
It is shown as:
To realize that the scalability of interactive system is supported.
9. according to claim 1 with method described in any one of 7, according to the training dataset after being split, and to each
Feature described based on the contextual information of LSTM and level LSTM, forms training dataset to each feature, and then to every
A feature is individually predicted that then stateful combination is denoted as the status tracking result at current time.
10. a kind of electronic equipment for applying method as claimed in any one of claims 1-9 wherein.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910071447.0A CN109885668A (en) | 2019-01-25 | 2019-01-25 | A kind of expansible field interactive system status tracking method and apparatus |
| PCT/CN2019/073477 WO2020151017A1 (en) | 2019-01-25 | 2019-01-28 | Scalable field human-machine dialogue system state tracking method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910071447.0A CN109885668A (en) | 2019-01-25 | 2019-01-25 | A kind of expansible field interactive system status tracking method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109885668A true CN109885668A (en) | 2019-06-14 |
Family
ID=66926795
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910071447.0A Pending CN109885668A (en) | 2019-01-25 | 2019-01-25 | A kind of expansible field interactive system status tracking method and apparatus |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109885668A (en) |
| WO (1) | WO2020151017A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111026857A (en) * | 2019-12-10 | 2020-04-17 | 苏州思必驰信息科技有限公司 | Conversation state tracking method, man-machine conversation method and system |
| CN112818097A (en) * | 2021-01-26 | 2021-05-18 | 山西三友和智慧信息技术股份有限公司 | Off-task training system based on dialog box state tracking model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170228366A1 (en) * | 2016-02-05 | 2017-08-10 | Adobe Systems Incorporated | Rule-based dialog state tracking |
| US9977778B1 (en) * | 2016-11-03 | 2018-05-22 | Conduent Business Services, Llc | Probabilistic matching for dialog state tracking with limited training data |
| CN108415923A (en) * | 2017-10-18 | 2018-08-17 | 北京邮电大学 | The intelligent interactive system of closed domain |
| CN108710704A (en) * | 2018-05-28 | 2018-10-26 | 出门问问信息科技有限公司 | Determination method, apparatus, electronic equipment and the storage medium of dialogue state |
| CN109063035A (en) * | 2018-07-16 | 2018-12-21 | 哈尔滨工业大学 | A kind of man-machine more wheel dialogue methods towards trip field |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108304489B (en) * | 2018-01-05 | 2021-12-28 | 广东工业大学 | Target-guided personalized dialogue method and system based on reinforcement learning network |
-
2019
- 2019-01-25 CN CN201910071447.0A patent/CN109885668A/en active Pending
- 2019-01-28 WO PCT/CN2019/073477 patent/WO2020151017A1/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170228366A1 (en) * | 2016-02-05 | 2017-08-10 | Adobe Systems Incorporated | Rule-based dialog state tracking |
| US9977778B1 (en) * | 2016-11-03 | 2018-05-22 | Conduent Business Services, Llc | Probabilistic matching for dialog state tracking with limited training data |
| CN108415923A (en) * | 2017-10-18 | 2018-08-17 | 北京邮电大学 | The intelligent interactive system of closed domain |
| CN108710704A (en) * | 2018-05-28 | 2018-10-26 | 出门问问信息科技有限公司 | Determination method, apparatus, electronic equipment and the storage medium of dialogue state |
| CN109063035A (en) * | 2018-07-16 | 2018-12-21 | 哈尔滨工业大学 | A kind of man-machine more wheel dialogue methods towards trip field |
Non-Patent Citations (1)
| Title |
|---|
| 杜晓宇: ""基于LSTM的对话状态追踪模型研究与实现"", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111026857A (en) * | 2019-12-10 | 2020-04-17 | 苏州思必驰信息科技有限公司 | Conversation state tracking method, man-machine conversation method and system |
| CN111026857B (en) * | 2019-12-10 | 2022-07-08 | 思必驰科技股份有限公司 | Conversation state tracking method, man-machine conversation method and system |
| CN112818097A (en) * | 2021-01-26 | 2021-05-18 | 山西三友和智慧信息技术股份有限公司 | Off-task training system based on dialog box state tracking model |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020151017A1 (en) | 2020-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106407178B (en) | A kind of session abstraction generating method, device, server apparatus and terminal device | |
| CN108255934B (en) | Voice control method and device | |
| US10891322B2 (en) | Automatic conversation creator for news | |
| EP3711031A1 (en) | Analyzing spatially-sparse data based on submanifold sparse convolutional neural networks | |
| CN114556354A (en) | Automatically determining and presenting personalized action items from an event | |
| CN110287461A (en) | Text conversion method, device and storage medium | |
| CN110222827A (en) | The training method of text based depression judgement network model | |
| CN110032630A (en) | Talk about art recommendation apparatus, method and model training equipment | |
| CN110245221A (en) | The method and computer equipment of training dialogue state tracking classifier | |
| CN110364146A (en) | Audio recognition method, device, speech recognition apparatus and storage medium | |
| CN110807566A (en) | Artificial intelligence model evaluation method, device, equipment and storage medium | |
| CN111027838B (en) | Crowd-sourced task pushing method, device, equipment and storage medium thereof | |
| Shen et al. | Kwickchat: A multi-turn dialogue system for aac using context-aware sentence generation by bag-of-keywords | |
| US20230088445A1 (en) | Conversational recommendation method, method of training model, device and medium | |
| KR101677859B1 (en) | Method for generating system response using knowledgy base and apparatus for performing the method | |
| US10770072B2 (en) | Cognitive triggering of human interaction strategies to facilitate collaboration, productivity, and learning | |
| CN110019691A (en) | Conversation message treating method and apparatus | |
| CN109741735A (en) | The acquisition methods and device of a kind of modeling method, acoustic model | |
| Elshaer et al. | Transfer learning from sound representations for anger detection in speech | |
| Windiatmoko et al. | Developing FB chatbot based on deep learning using RASA framework for university enquiries | |
| CN113314119A (en) | Voice recognition intelligent household control method and device | |
| Dang et al. | Dynamic multi-rater gaussian mixture regression incorporating temporal dependencies of emotion uncertainty using kalman filters | |
| CN115953521A (en) | Remote digital human rendering method, device and system | |
| CN117251057A (en) | AIGC-based method and system for constructing AI number wisdom | |
| CN109885668A (en) | A kind of expansible field interactive system status tracking method and apparatus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |
|
| RJ01 | Rejection of invention patent application after publication |