CN110019730A - Automatic interaction system and intelligent terminal - Google Patents
Automatic interaction system and intelligent terminal Download PDFInfo
- Publication number
- CN110019730A CN110019730A CN201711420778.8A CN201711420778A CN110019730A CN 110019730 A CN110019730 A CN 110019730A CN 201711420778 A CN201711420778 A CN 201711420778A CN 110019730 A CN110019730 A CN 110019730A
- Authority
- CN
- China
- Prior art keywords
- answer
- customer problem
- prediction probability
- interaction system
- knowledge
- 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
- 230000003993 interaction Effects 0.000 title claims abstract description 22
- 238000012216 screening Methods 0.000 claims abstract description 34
- 238000013507 mapping Methods 0.000 claims abstract description 33
- 230000004044 response Effects 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 4
- 238000000034 method Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 abstract description 6
- 230000002452 interceptive effect Effects 0.000 abstract description 5
- 230000000875 corresponding effect Effects 0.000 description 11
- 238000012549 training Methods 0.000 description 7
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 235000013351 cheese Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
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
- G06F16/3329—Natural language query formulation or dialogue systems
-
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Animal Behavior & Ethology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of automatic interaction system and intelligent terminal.The automatic interaction system includes: that customer problem obtains module, to obtain customer problem；Answer obtains module, and to obtain multiple answers for the customer problem in the way of at least two, the mode includes knowledge base, knowledge mapping and learning model；Answer screening module, to be screened according to the source parameter and prediction probability of each answer to the multiple answer, the source parameter and prediction probability of the answer are determined according to the acquisition modes of the answer；Answer output module, the optimal answer obtained to export screening.The accuracy and continuity responded in question and answer interactive process can be improved in technical solution through the invention.
Description
Technical field
The present invention relates to natural language processing technique field more particularly to a kind of automatic interaction systems and intelligent terminal.
Background technique
In artificial intelligence technology application field, there are more and more intelligent answer products.Typically, for user
The reply accuracy and reply speed of problem are an important factor for influencing intelligent answer product quality.
A variety of question and answer processing modes exist in the prior art, there are commonly rule-based mode, based on template matching
Mode, the mode based on retrieval, based on production mode etc..Wherein, the mode based on retrieval be by retrieval knowledge library
Some knowledge points generate answer, and knowledge base generally includes multiple knowledge points, and each knowledge point includes that standard asks about its corresponding expansion
Exhibition is asked and answer；Answer feedback mechanism based on production is the Automatic generation of information according to active user's input by word sequence
Arrange the answer of composition.
But in rule-based, template matching, the mode of retrieval, template, example or database have limitation, and
And lack effective language understanding, cause to exist in the accuracy and flexibility of answer certain insufficient；Based on production mode
It needs to establish and training pattern, model complexity height, the stability of answer acquisition process is low.
Summary of the invention
Present invention solves the technical problem that being how to improve the accuracy and continuity responded in question and answer interactive process.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of automatic interaction system, comprising:
Customer problem obtains module, to obtain customer problem；
Answer obtains module, described to obtain multiple answers for the customer problem in the way of at least two
Mode includes knowledge base, knowledge mapping and learning model；
Answer screening module, to be carried out according to the source parameter and prediction probability of each answer to the multiple answer
Screening, the source parameter and prediction probability of the answer are determined according to the acquisition modes of the answer；The source of the answer
Parameter includes that the generation of answer is time-consuming, and the answer screening module specifically includes: the answer screening module includes: second important
Property weight determination unit, determines the corresponding importance weight of the answer to the source parameter according to each answer；Product calculates
Unit, to calculate the importance weight of each answer and the product of prediction probability；Second effective score calculating unit, to count
It calculates the product of each answer and generates time-consuming quotient, using effective score as each answer；Third filtering unit, to basis
Effective score of each answer screens the multiple answer；
Answer output module, the optimal answer obtained to export screening.
Optionally, the prediction probability of the one or more determining answers of the answer screening module in the following ways:
If the answer comes from the knowledge base, calculate the customer problem and the knowledge base Plays ask and/
Or the semantic similarity that extension is asked, using the prediction probability as the answer；
If the answer comes from the knowledge mapping, according to the confidence level for the answer that the knowledge mapping determines
Determine the prediction probability of the answer；
If the answer carrys out self learning model, determined according to the sum of the conditional probability between the adjacent word of the answer
The prediction probability of the answer.
Optionally, the source parameter includes priority, and the priority of the answer from knowledge base, which is higher than, comes from knowledge graph
The priority of the answer of spectrum, the priority of the answer from knowledge mapping are higher than the priority for carrying out the answer of self learning model.
Optionally, the answer screening module includes:
First screening unit successively judges the prediction probability of each answer to the priority sequence according to answer
Whether it is greater than given threshold, and the answer that the prediction probability judged for the first time is greater than the given threshold is answered as optimal
Case.
Optionally, the answer acquisition module includes:
The language asked is asked and/or extended to first answer acquiring unit to calculate the customer problem and knowledge base Plays
Adopted similarity, and determine the first answer from the knowledge base；
Second answer acquiring unit to match the customer problem with the knowledge in knowledge mapping, and determines
The second answer from the knowledge mapping；
Third answer acquiring unit the customer problem is inputted learning model, and determines the learning model
Output is third answer.
Optionally, the customer problem is voice；It includes: the first voice converting unit that the customer problem, which obtains module,
The customer problem is converted to text, the answer output module includes: the second voice converting unit, will obtain
The optimal answer be converted to voice after be sent to user.
Optionally, the customer problem obtains module in response to executing after the indication message that receives.
Optionally, the intention assessment is the result is that using the intent classifier model that training is completed in advance to the customer problem
Carry out what intention assessment obtained.
The embodiment of the invention also discloses a kind of intelligent terminals comprising above-mentioned automatic interaction system.
Optionally, the intelligent terminal is service robot, mobile phone or tablet computer.
Compared with prior art, the technical solution of the embodiment of the present invention has the advantages that
Technical solution of the present invention obtains customer problem；It is obtained in the way of at least two for the multiple of the customer problem
Answer；The multiple answer is screened according to the source parameter of each answer and prediction probability, the source of the answer
Parameter and prediction probability are determined according to the acquisition modes of the answer；The optimal answer that output screening obtains.The technology of the present invention
Scheme obtains multiple answers in the way of at least two, and optimal answer output is then screened in multiple answers；It is answered due to obtaining
The mode of case is different, therefore can promote the rich of answer from multiple angles, can to avoid using single mode obtain less than
The case where answer, guarantees the sustainability interacted with user's question and answer, promotes user experience.In addition, according to each from multiple answers
The source parameter and prediction probability of a answer choose optimal answer, it is ensured that it is accurate that answer is replied for customer problem
Property.Technical solution of the present invention obtains answer using knowledge base, knowledge mapping or learning model, be based on different technical principles come
It obtains, therefore can further promote the rich of answer；In addition, being asked by what above-mentioned three kinds of approach were got for user
The accuracy of the answer of topic is higher, and the accuracy of the optimal answer obtained after further screening is higher, further increases and asks
Answer the accuracy responded in interactive process.
Detailed description of the invention
Fig. 1 is a kind of structural schematic diagram of automatic interaction device of the embodiment of the present invention；
Fig. 2 is a kind of structural schematic diagram for specific embodiment that answer shown in Fig. 1 obtains module 502；
Fig. 3 is a kind of structural schematic diagram of specific embodiment of answer screening module shown in Fig. 1 503；
Fig. 4 is the structural schematic diagram of another specific embodiment of answer screening module shown in Fig. 1 503.
Specific embodiment
As described in the background art, the prior art is in rule-based, template matching, the mode of retrieval, template, example or
Database has limitation, and lacks effective language understanding, causes to exist in the accuracy and flexibility of answer certain
It is insufficient；It needs to establish based on production mode and training pattern, model complexity height, answer acquisition process stability is low.
Technical solution of the present invention obtains multiple answers in the way of at least two, and optimal answer then is screened in multiple answers
Case output；Since the mode for obtaining answer is different, the rich of answer can be promoted from multiple angles, it can be to avoid use
Single mode obtains the case where less than answer, guarantees the sustainability interacted with user's question and answer, promotes user experience.In addition, from
Optimal answer is chosen according to the source parameter and prediction probability of each answer in multiple answers, it is ensured that answer is directed to user
The accuracy that problem is replied.
To make the above purposes, features and advantages of the invention more obvious and understandable, with reference to the accompanying drawing to the present invention
Specific embodiment be described in detail.
As shown in Figure 1, automatic interaction system 50 may include, customer problem obtains module 501, answer obtains module 502,
Answer screening module 503 and answer output module 504.
Wherein, customer problem obtains module 501 to obtain customer problem；Answer obtains module 502 to using at least
Two ways obtains multiple answers for the customer problem；Answer screening module 503 is to the source according to each answer
Parameter and prediction probability screen the multiple answer, and the source parameter and prediction probability of the answer are answered according to this
What the acquisition modes of case determined；The optimal answer that answer output module 504 is obtained to export screening.
In the present embodiment, customer problem, which obtains module 501, can obtain customer problem, example using any enforceable mode
Such as customer problem directly can be collected from outside, customer problem can also be obtained in such a way that interface calls.User asks
Topic can have semanteme.Specifically, customer problem can be the forms such as voice, text.
The answer acquisition module 502 can use at least two modes and obtain at least two answers.Specifically, obtaining
There are many modes of the answer of customer problem, for example, being matched using knowledge base to customer problem, and will be with customer problem phase
Matched standard asks corresponding answer as the answer for being directed to customer problem；Also can use searching algorithm take notice of in map space into
Row search, and using the answer searched as the answer for being directed to customer problem.Wherein, it is intended that space can be it is preset, can also
To be constantly replenished by on-line study.Alternatively, code identification can also be carried out customer problem using deep learning model
And the corresponding answer of decoded output.
Due to obtaining answer using various ways, every kind of acquisition modes obtain answer using different principles, therefore can be with
Promote the rich of answer.
Furthermore, when obtaining the answer of customer problem using single mode, it may appear that the case where answer can not be obtained,
Such as the answer not matched with customer problem in knowledge base, there is no the knots to match with customer problem in knowledge mapping
Point.So, answer is obtained by using various ways, can guarantees the stability that answer obtains to avoid above situation.
In order to determine the more accurately answer for being directed to customer problem, answer screening module 503 screens multiple answers.
The foundation of screening is the source parameter and prediction probability of each answer.
Wherein, the source parameter of the answer and prediction probability are determined according to the acquisition modes of the answer.Due to adopting
The accuracy and importance of the answer obtained with different acquisition modes have differences, therefore characterize difference using source parameter
The accuracy and importance of answer, and then when screening answer, it can use source choice of parameters and go out accuracy or importance more
High answer, to improve the accuracy of optimal answer.It is come from for example, the accuracy of the answer obtained using database mode is higher than
The source parameter of the answer of knowledge mapping, the answer obtained as a result, using database mode is different from answering from knowledge mapping
The source parameter of case.
The prediction probability of answer can characterize the accuracy that answer replies customer problem.Prediction probability is bigger,
The accuracy of corresponding answer is higher.The higher answer of accuracy can be filtered out using the prediction probability of answer as a result, to improve
The accuracy of optimal answer.
The answer output module 504 exports answer screening module 503 and screens obtained optimal answer.Specifically, can
It is presented to the user with directly exporting optimal answer, such as is presented to the user in a manner of text, voice etc..It can also be based on optimal
Answer carries out the interaction of other operation realizations and user.For example, executing sequence of operations based on optimal answer.
The embodiment of the present invention obtains multiple answers in the way of at least two, and optimal answer is then screened in multiple answers
Output；Since the mode for obtaining answer is different, the rich of answer can be promoted from multiple angles, it can be to avoid using single
One mode obtains the case where less than answer, guarantees the sustainability interacted with user's question and answer, promotes user experience.In addition, from more
Optimal answer is chosen according to the source parameter and prediction probability of each answer in a answer, it is ensured that answer is asked for user
Inscribe the accuracy replied.
Automatic interaction system 50 in the present embodiment can independently execute, the instruction independent of other computer instructions.
Preferably, at least two mode is selected from knowledge base, knowledge mapping and learning model.
In the present embodiment, it can use knowledge base and obtain answer.Knowledge base includes problem and answer, by by customer problem
It is matched with problem in knowledge base to obtain answer.
The present embodiment also can use knowledge mapping and obtain answer.Knowledge mapping is a kind of semantic network, including node and
The side of link node.Node represents entity or concept, the various semantic relations between Bian Daibiao entity/concept.Specifically, know
The data known in map are stored with triple data mode, it may be assumed that<entity A, relationship, entity B>, such as:<Liu Dehua, out
Radix Rehmanniae, Hong Kong >.If customer problem are as follows: " where the birthplace of Liu Dehua is? ", it is " fragrant for obtaining answer using knowledge mapping
Port ", prediction probability 0.9986.
It should be noted that the network structure of knowledge mapping is also possible to any enforceable mode in the prior art, this
Inventive embodiments are without limitation.
The present embodiment can also obtain answer using learning model.Learning model can be deep learning model or engineering
Model is practised, such as can be shot and long term memory models (long-short term memory, LSTM).The problem of for inputting,
Learning model can automatically generate answer according to neural network.
It is possible to further the prediction probability of one or more determining answers in the following ways: if the answer
From the knowledge base, then calculates the customer problem and the semantic similarity asked is asked and/or extended to the knowledge base Plays,
Using the prediction probability as the answer；If the answer comes from the knowledge mapping, determined according to the knowledge mapping
The confidence level of the answer determine the prediction probability of the answer；If the answer carrys out self learning model, according to
The sum of conditional probability between the adjacent word of answer determines the prediction probability of the answer.
In specific implementation, when obtaining answer using knowledge storehouse matching, customer problem and the knowledge base can be got the bid
The prediction probability that the semantic similarity asked can indicate gained answer is asked and/or extended to standard；Answer is being obtained using knowledge mapping
When, knowledge mapping has the marking of confidence level for its answer determined, as a result, can be with according to the marking of the confidence level to answer
Determine its prediction probability；When the production mode based on deep learning model obtains answer, the prediction probability of answer be can be
The sum of conditional probability in answer between the word of front and back.
Further, the source parameter includes priority, and the priority of the answer from knowledge base, which is higher than, comes from knowledge
The priority of the answer of map, the priority of the answer from knowledge mapping are higher than the priority for carrying out the answer of self learning model.
In the present embodiment, be in view of the problems in knowledge base and answer it is pre-configured, it is obtained using knowledge base
The accuracy of answer is higher.Learning model needs are trained in advance, and the training effect of learning model is trained used
The influence of corpus, the accuracy using the answer of learning model generation are lower.The answer got using knowledge mapping it is accurate
Property falls between.The priority of the corresponding source parameter of above-mentioned three kinds of modes is followed successively by knowledge base from big to small, knows as a result,
Know map and learning model.
It should be noted that the demand to answer may also be different according to the difference of practical application scene, therefore come
The priority of source parameter can also carry out the configuration of adaptability according to concrete application scene, and the embodiment of the present invention does not limit this
System.
In a kind of specific embodiment of the present invention, as shown in Fig. 2, it may include first that the answer, which obtains module 502,
The semantic similarity asked is asked and/or extended to answer acquiring unit 5021 to calculate the customer problem and knowledge base Plays,
And determine the first answer from the knowledge base；
Second answer acquiring unit 5022, the customer problem to be matched with the knowledge in knowledge mapping, and
Determine the second answer from the knowledge mapping；
Third answer acquiring unit 5023 customer problem is inputted learning model, and determines the study mould
The output of type is third answer.
It is that customer problem is asked and/or expanded with the standard in knowledge base when obtaining answer using knowledge base in the present embodiment
Exhibition, which is asked, to be matched, and is asked or is extended if there is the standard that the semantic similarity with customer problem reaches given threshold and asks, then will
The standard, which is asked or extended, asks corresponding answer as the first answer.
It is to match customer problem with the triple data in knowledge mapping when obtaining answer using knowledge mapping,
If there is the triple data to match with customer problem, then using the triple data interior joint as the second answer.
It is that customer problem is input to the learning model when obtaining answer using learning model, which can be certainly
It is dynamic to generate the answer for being directed to customer problem, and using this as third answer.
In a specific embodiment of the invention, answer screening module 503 includes the first screening unit (not shown), to
According to the priority sequence of answer, successively judge whether the prediction probability of each answer is greater than given threshold, and will for the first time
Judge that obtained prediction probability is greater than the answer of the given threshold as optimal answer.
It is preferential to choose the high answer of priority in the present embodiment.That is, first judging the pre- of the higher answer of priority
Survey whether probability is greater than given threshold, if it is, using the answer as optimal answer.Otherwise, continue to judge next priority
The prediction probability of answer whether be greater than given threshold, until filtering out optimal answer.The quantity of answer have it is multiple, to guarantee
Optimal answer can be obtained.
In another specific embodiment of the invention, as shown in figure 3, answer screening module 503 may include:
First importance weight determination unit 5031, to determine the important of the answer according to the source parameter of each answer
Property weight；
Accuracy weight determination unit 5032, to determine that the accuracy of the answer is weighed according to the prediction probability of each answer
Value；
First effective score calculating unit 5033, to the importance weight and accuracy weight computing using each answer
Effective score of each answer；
Second screening unit 5034, to be screened according to effective score of each answer to the multiple answer.
Relative to the source parameter for first considering answer in previous embodiment, the prediction probability of answer is considered further that；The present embodiment
Middle source parameter and prediction probability by answer is considered simultaneously.
Specifically, effective score can be importance weight and accuracy weight and obtain by any enforceable mathematical operation
It arrives, can be importance weight and accuracy weights sum, be also possible to the product of importance weight and accuracy weight, this hair
Bright embodiment is without limitation.
The first importance weight determination unit 5031 and accuracy weight determination unit 5032 of the present embodiment are according to answer
Source parameter can determine corresponding importance weight, that is to say, that the acquisition modes of answer can have different important
Property weight.Furthermore, the accuracy for the answer that different acquisition modes are got and the importance weight of the answer are positively correlated.
Corresponding accuracy weight can be determined according to the prediction probability of answer.The size of accuracy weight and the size of prediction probability are just
It is related.
The effective score calculating unit 5033 of the first of the present embodiment and the second screening unit 5034, utilize the weight of each answer
The property wanted weight and accuracy weight computing obtain effective score of answer.Effective score of answer can be with the comprehensive characterization answer
Accuracy.Effective score is higher, and the accuracy of answer is higher.Optimal answer is the effectively highest answer of score in each answer.
Further, the source parameter of the answer may include that the generation of answer is time-consuming.
As shown in figure 4, answer screening module 503 may include: in another specific embodiment of the invention
Second importance weight determination unit 5035 determines that the answer is corresponding to the source parameter according to each answer
Importance weight；
Product computing unit 5036, to calculate the importance weight of each answer and the product of prediction probability；
Second effective score calculating unit 5037, to calculate the product of each answer and generate time-consuming quotient, using as
Effective score of each answer；
Third filtering unit 5038, to be screened according to effective score of each answer to the multiple answer.
The source parameter and prediction probability of answer are considered simultaneously relative in embodiment illustrated in fig. 3.The present invention is real
Example is applied also for the time-consuming considerations as optimal answer of the generation of answer.
In specific implementation, generate time-consuming negatively correlated with effective score.Generation is time-consuming longer, and the validity of answer is lower.?
When second effective score calculating unit 5037 calculates effective score of answer, importance weight and prediction probability can obtained
After product, the product and generation time-consuming are done into quotient, to obtain effective score.
The course of work of the remaining element of Fig. 4 can refer to the corresponding unit of Fig. 3, and details are not described herein.
In a concrete application scene of the invention, the customer problem is voice；The customer problem obtains module 501
It include: the first voice converting unit (not shown), the customer problem is converted to text, the answer output module
504 include: the second voice converting unit (not shown), to be sent to use after the optimal answer obtained is converted to voice
Family.
That is, after needing to convert voice data into text, then execute subsequent step.Why digitize the speech into
It is to calculate customer problem and standard in the next steps and ask and/or extend the semantic similarity asked for text.
In order to guarantee the consistency interacted with user, when user is interacted using this mode of voice, use is fed back to
The optimal answer at family also uses voice.As a result, when optimal answer is textual form, optimal answer is converted to defeated again after voice
Out to user.
In another concrete application scene of the invention, the customer problem obtains module 501 and cuts in response to what is received
It is executed after changing instruction information.
Further, the indication message be using professional knowledge library to the customer problem carry out it fails to match
When issue.It is professional due to question and answer in the question and answer interactive process for carrying out professional domain using professional knowledge library, it may
The case where appearance can not obtain answer, therefore response is executed by indication message instruction switching, it can guarantee to get to answer
Case promotes user experience to realize the continuity of interaction.For example, being matched using professional knowledge library to the customer problem
When failure, by executing response, the optimal answer got is " hello ", into chat mode, realizes interactive continuity.
Further, the indication message is the intention assessment result in the customer problem and default intent classifier
It is issued when successful match.In the present embodiment, relative to using professional knowledge library to the customer problem carry out it fails to match when
Indication message is issued, the embodiment of the present invention is before matching the customer problem using professional knowledge library to user
Problem is classified, if the intention assessment result of customer problem and default intent classifier successful match, issues switching instruction
Information executes response with instruction switching.For example, default intent classifier is to chat classification, the intention assessment result of customer problem is
When chatting classification, then indication message is issued, response is executed with instruction switching.
Further, the intention assessment is the result is that ask the user using the intent classifier model that training is completed in advance
Topic carries out what intention assessment obtained.
In specific implementation, the question and answer corpus training that can advance with accumulation obtains intent classifier model.What training was completed
Intent classifier model can carry out intention assessment to the customer problem.For example, user inputs " I am somewhat unhappy ", it is intended that point
Class model can be classified as chatting classification.Then indication message is issued, response is executed with instruction switching, into machine
Chat mode, and export optimal answer " small i fools you, and say cheese, the people of smile is most charming！".
It should be noted that intent classifier model can be implemented using existing any enforceable sorting algorithm, the present invention
Example is without limitation.
The embodiment of the invention also discloses a kind of intelligent terminal, the intelligent terminal may include above-mentioned automatic interaction system
System, so that intelligent terminal is just provided with the function of intelligent answer.
Specifically, the intelligent terminal can be service robot, mobile phone or tablet computer.
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (10)
1. a kind of automatic interaction system characterized by comprising
Customer problem obtains module, to obtain customer problem；
Answer obtains module, to obtain multiple answers for the customer problem, the mode in the way of at least two
Including knowledge base, knowledge mapping and learning model；
Answer screening module, to be sieved according to the source parameter and prediction probability of each answer to the multiple answer
Choosing, the source parameter and prediction probability of the answer are determined according to the acquisition modes of the answer；Join in the source of the answer
Number includes that the generation of answer is time-consuming, and the answer screening module specifically includes: the answer screening module includes: the second importance
Weight determination unit determines the corresponding importance weight of the answer to the source parameter according to each answer；Product calculates single
Member, to calculate the importance weight of each answer and the product of prediction probability；Second effective score calculating unit, to calculate
The product of each answer and the quotient for generating time-consuming, using effective score as each answer；Third filtering unit, to according to each
Effective score of a answer screens the multiple answer；
Answer output module, the optimal answer obtained to export screening.
2. automatic interaction system according to claim 1, which is characterized in that the answer screening module is in the following ways
One or more determining answers prediction probability:
If the answer comes from the knowledge base, calculates the customer problem and ask and/or expand with the knowledge base Plays
The semantic similarity asked is opened up, using the prediction probability as the answer；
If the answer comes from the knowledge mapping, the confidence level of the answer determined according to the knowledge mapping is determined
The prediction probability of the answer；
If the answer carrys out self learning model, determined according to the sum of conditional probability between the adjacent word of the answer described in
The prediction probability of answer.
3. automatic interaction system according to claim 1, which is characterized in that the source parameter includes priority, is come from
The priority of the answer of knowledge base is higher than the priority of the answer from knowledge mapping, the priority of the answer from knowledge mapping
It is higher than the priority for carrying out the answer of self learning model.
4. automatic interaction system according to claim 1, which is characterized in that the answer screening module includes:
First screening unit, to the priority sequence according to answer, successively judge each answer prediction probability whether
It is greater than the answer of the given threshold as optimal answer greater than given threshold, and using the prediction probability judged for the first time.
5. automatic interaction system according to claim 1, which is characterized in that the answer obtains module and includes:
The semantic phase asked is asked with knowledge base Plays and/or extended to first answer acquiring unit to calculate the customer problem
Like degree, and determine the first answer from the knowledge base；
Second answer acquiring unit, to match the customer problem with the knowledge in knowledge mapping, and determination comes from
Second answer of the knowledge mapping；
Third answer acquiring unit the customer problem is inputted learning model, and determines the output of the learning model
For third answer.
6. automatic interaction system according to claim 1, which is characterized in that the customer problem is voice；The user
It includes: the first voice converting unit that problem, which obtains module, and the customer problem is converted to text, the answer exports mould
Block includes: the second voice converting unit, to be sent to user after the optimal answer obtained is converted to voice.
7. automatic interaction system according to claim 1, which is characterized in that the customer problem obtains module in response to connecing
It is executed after the indication message received.
8. automatic interaction system according to claim 2, which is characterized in that the intention assessment is the result is that utilize instruction in advance
Practice the intent classifier model completed and what intention assessment obtained is carried out to the customer problem.
9. a kind of intelligent terminal, which is characterized in that including automatic interaction system such as described in any item of the claim 1 to 8.
10. intelligent terminal as claimed in claim 9, which is characterized in that the intelligent terminal is service robot, mobile phone or flat
Plate computer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711420778.8A CN110019730B (en) | 2017-12-25 | 2017-12-25 | Automatic interaction system and intelligent terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711420778.8A CN110019730B (en) | 2017-12-25 | 2017-12-25 | Automatic interaction system and intelligent terminal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110019730A true CN110019730A (en) | 2019-07-16 |
| CN110019730B CN110019730B (en) | 2024-07-26 |
Family
ID=67187115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711420778.8A Active CN110019730B (en) | 2017-12-25 | 2017-12-25 | Automatic interaction system and intelligent terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110019730B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021169640A1 (en) * | 2020-02-25 | 2021-09-02 | 京东方科技集团股份有限公司 | Query device and method, apparatus, and storage medium |
| CN114153961A (en) * | 2022-02-07 | 2022-03-08 | 杭州远传新业科技有限公司 | Knowledge graph-based question and answer method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729402A (en) * | 2013-11-22 | 2014-04-16 | 浙江大学 | Method for establishing mapping knowledge domain based on book catalogue |
| CN105893535A (en) * | 2016-03-31 | 2016-08-24 | 上海智臻智能网络科技股份有限公司 | Intelligent question and answer method, knowledge base optimizing method and device and intelligent knowledge base |
| US20160379120A1 (en) * | 2015-06-25 | 2016-12-29 | International Business Machines Corporation | Knowledge Canvassing Using a Knowledge Graph and a Question and Answer System |
| WO2017041372A1 (en) * | 2015-09-07 | 2017-03-16 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN106649825A (en) * | 2016-12-29 | 2017-05-10 | 上海智臻智能网络科技股份有限公司 | Voice interaction system, establishment method and device thereof |
| WO2017092380A1 (en) * | 2015-12-03 | 2017-06-08 | 华为技术有限公司 | Method for human-computer dialogue, neural network system and user equipment |
| CN107369098A (en) * | 2016-05-11 | 2017-11-21 | 华为技术有限公司 | The treating method and apparatus of data in social networks |
-
2017
- 2017-12-25 CN CN201711420778.8A patent/CN110019730B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729402A (en) * | 2013-11-22 | 2014-04-16 | 浙江大学 | Method for establishing mapping knowledge domain based on book catalogue |
| US20160379120A1 (en) * | 2015-06-25 | 2016-12-29 | International Business Machines Corporation | Knowledge Canvassing Using a Knowledge Graph and a Question and Answer System |
| WO2017041372A1 (en) * | 2015-09-07 | 2017-03-16 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| WO2017092380A1 (en) * | 2015-12-03 | 2017-06-08 | 华为技术有限公司 | Method for human-computer dialogue, neural network system and user equipment |
| CN106844368A (en) * | 2015-12-03 | 2017-06-13 | 华为技术有限公司 | For interactive method, nerve network system and user equipment |
| CN105893535A (en) * | 2016-03-31 | 2016-08-24 | 上海智臻智能网络科技股份有限公司 | Intelligent question and answer method, knowledge base optimizing method and device and intelligent knowledge base |
| CN107369098A (en) * | 2016-05-11 | 2017-11-21 | 华为技术有限公司 | The treating method and apparatus of data in social networks |
| CN106649825A (en) * | 2016-12-29 | 2017-05-10 | 上海智臻智能网络科技股份有限公司 | Voice interaction system, establishment method and device thereof |
Non-Patent Citations (2)
| Title |
|---|
| 刘康;张元哲;纪国良;来斯惟;赵军;: "基于表示学习的知识库问答研究进展与展望", 自动化学报, no. 06 * |
| 王元卓;贾岩涛;刘大伟;靳小龙;程学旗;: "基于开放网络知识的信息检索与数据挖掘", 计算机研究与发展, no. 02 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021169640A1 (en) * | 2020-02-25 | 2021-09-02 | 京东方科技集团股份有限公司 | Query device and method, apparatus, and storage medium |
| CN114153961A (en) * | 2022-02-07 | 2022-03-08 | 杭州远传新业科技有限公司 | Knowledge graph-based question and answer method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110019730B (en) | 2024-07-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107908803A (en) | The response method and device, storage medium, terminal of question and answer interaction | |
| CN110019838A (en) | Intelligent Answer System and intelligent terminal | |
| CN110019729A (en) | Intelligent answer method and storage medium, terminal | |
| CN110019728A (en) | Automatic interaction method and storage medium, terminal | |
| CN105391730B (en) | A kind of information feedback method, apparatus and system | |
| CN109961780A (en) | Man-machine interaction method, device, server and storage medium | |
| CN111435482A (en) | Outbound model construction method, outbound method, device and storage medium | |
| CN109960722B (en) | Information processing method and device | |
| CN109101216A (en) | Audio method of adjustment, device, electronic equipment and storage medium | |
| CN108519998B (en) | Problem guiding method and device based on knowledge graph | |
| CN114090755A (en) | Reply sentence determination method and device based on knowledge graph and electronic equipment | |
| CN110223358A (en) | Visible pattern design method, training method, device, system and storage medium | |
| CN113778871A (en) | Mock testing method, device, equipment and storage medium | |
| CN111124898B (en) | Question-answering system testing method and device, computer equipment and storage medium | |
| CN110019730A (en) | Automatic interaction system and intelligent terminal | |
| CN113840040A (en) | Man-machine cooperation outbound method, device, equipment and storage medium | |
| CN106356056B (en) | Audio recognition method and device | |
| CN113901194A (en) | Customer service matching method and related equipment | |
| CN117750050A (en) | Information processing method and device based on large language model and electronic equipment | |
| CN109002498A (en) | Interactive method, device, equipment and storage medium | |
| CN109255016A (en) | Answer method, device and computer readable storage medium based on deep learning | |
| CN113691403B (en) | Topology node configuration method, related device and computer program product | |
| CN115375965A (en) | Preprocessing method for target scene recognition and target scene recognition method | |
| CN115543090A (en) | Topic transfer method, topic transfer device, electronic equipment and storage medium | |
| CN111985901B (en) | Marketing product configuration method, device, equipment and storage medium in telecom industry |
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 |