CN107562863A - Chat robots reply automatic generation method and system - Google Patents
Chat robots reply automatic generation method and system Download PDFInfo
- Publication number
- CN107562863A CN107562863A CN201710760328.7A CN201710760328A CN107562863A CN 107562863 A CN107562863 A CN 107562863A CN 201710760328 A CN201710760328 A CN 201710760328A CN 107562863 A CN107562863 A CN 107562863A
- Authority
- CN
- China
- Prior art keywords
- answer
- question
- sentence
- input
- generation module
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Replying automatic generation method and system, methods described the present invention relates to a kind of chat robots includes：The keyword in the sentence of user's input is extracted, corresponding question and answer pair are found as index search in priori knowledge base using the keyword；If finding corresponding question and answer pair, exported the reply sentence of the question and answer centering as revert statement；If not finding corresponding question and answer pair, the input by sentence that the user is inputted answers generation module, obtains revert statement output.The chat robots that invention provides reply automatic generation method and system so that the dialogue that robot automatically generates is more nearly real interpersonal dialogue, and can apply to Opening field.
Description
Technical field
The present invention relates to intelligent robot technology field, and in particular to a kind of chat robots reply automatic generation method and
System.
Background technology
The interactive of Opening field according to user's input and output, it is necessary to meet current session scene and context
Reply.At present, robot, which automatically replies generation method, many kinds, can substantially be divided into following two major class：
(1) the reply generation method based on retrieval：Stored in database<S is inputted, exports R>Question and answer pair, for new input
User's request, inquire about " input S " and new input the sentence similarity, by similarity highest of question and answer centering in database
" " the response that output R " inputs as robot to user corresponding to input S ".The shortcomings that this method is to need to put into substantial amounts of essence
Power builds database, and the question and answer in database are to limited amount, it is difficult to covers all application scenarios.
(2) method based on generation：Using the method for machine learning, generation is inputted according to user and replied, prior art master
If using the machine translation system based on statistics or end to end nerve network system come realize dialogue generation (natural language give birth to
Into), specific way is that sentence s is inputted in dialog procedure to predict by optimizing maximal possibility estimation (MLE) object function
Reply sentence r.The shortcomings that technology is, used maximum estimation objective function of feeling relieved is excessively simple during training pattern
It is single, it is impossible to capture all key takeaways of defining ideal dialogue completely, so as to cause the dialogue of generation is excessively stiff, repeat and
It is general, and lack to deep understanding above.In other words, the reply generated by this method is difficult to ensure that syntactically correct, upper and lower
The uniformity of text etc., the dialogue personification of generation is poor, unlike interpersonal real dialogue.
To sum up, existing interactive can not generate automatically coherent reply sentence, and can not be applied to open neck
Domain, at present, needing badly a kind of can be applied to Opening field and can carry out the chat robots of coherent dialogue.
The content of the invention
For in the prior art the defects of, a kind of chat robots provided by the invention reply automatic generation method and are
System so that the dialogue that robot automatically generates is more nearly real interpersonal dialogue, and can apply to open neck
Domain.
In a first aspect, the invention provides a kind of chat robots to reply automatic generation method, including：
The keyword in the sentence of user's input is extracted, is searched using the keyword as index search in priori knowledge base
To corresponding question and answer pair；
If finding corresponding question and answer pair, exported the reply sentence of the question and answer centering as revert statement；
If not finding corresponding question and answer pair, the input by sentence that the user is inputted answers generation module, is returned
Multiple sentence output.
Preferably, in addition to：If not finding corresponding question and answer pair, the input of user's next step is obtained, according to user
The input of next step judges whether the revert statement that epicycle exports in talking with is correct, is corrected according to judged result and answers generation mould
Block.
Preferably, the training method for answering generation module includes：
The real dialog language material of acquisition is made pauses in reading unpunctuated ancient writings, and according to question and answer pair form sample to obtain true question and answer to s,
R }, wherein, s is input sentence, and r is the true reply of the input sentence；
True question and answer are divided into training set and test set two parts to { s, r }；
Initial answer generation module is established by the training set；
The answer generation module is updated by the test set.
Preferably, it is described to establish initial answer generation module by the training set, including：
To the true question and answer in training set to { si, riSegmented and identify entity therein, and the entity of identification is entered
Row part-of-speech tagging, obtain each { s in training samplei, riCorresponding to question and answer entity to<si1,si2,...,sin>,<ri1,
ri2,...,rim>, wherein, sijFor siIn entity, ritFor riIn entity, j=1,2 ..., n, i=1,2 ..., m；
Respectively to sijAnd ritCarry out reference resolution and disambiguation；
Statistical condition probability P (rit|sij) and joint probability；
Calculate sijAnd ritTerm vector；
Obtain initial answer generation module.
Preferably, it is described that the answer generation module is updated by the test set, including：
The true question and answer in test set are obtained to { s ', r ' }；
Answer input sentence s ' input to generation module, obtain machine and reply R ', composition machine is answered to { s ', R ' }；
Machine is answered judge module is replied to { s ', R ' } input, calculate the confidence of { s ', R ' }；
Generation module is answered according to the confidence level of { s ', R ' } renewal.
Preferably, also include during the generation module is trained：
The true question and answer be labeled as front training sample to { s ', r ' }, the machine answer is labeled as to { s ', R ' }
Negative training sample；
{ s ', the r ' } of tape label and { s ', R ' } is used as training sample, sentenced using clustering algorithm training reply.
Preferably, generation module is answered in the confidence level renewal of the basis { s, r }, including：
According to the confidence level of { s ', R ' }, corresponding conditional probability P (r are adjustedit|sij) and joint probability, answered with updating
Generation module.
Second aspect, the invention provides a kind of chat robots to reply automatic creation system, including：
Priori searching unit, the keyword in sentence for extracting user's input, using the keyword as index
Lookup finds corresponding question and answer pair in priori knowledge base；
First reply unit, if for finding corresponding question and answer pair, using the reply sentence of the question and answer centering as
Revert statement exports；
Second replys unit, if for not finding corresponding question and answer pair, the input by sentence that the user is inputted returns
Generation module is answered, obtains revert statement output.
Preferably, in addition to negative-feedback unit, if for not finding corresponding question and answer pair, user's next step is obtained
Input, judge whether the revert statement that epicycle exports in talking with is correct, is rectified according to judged result according to the input of user's next step
It is positive to answer generation module.
The third aspect, the invention provides a kind of computer-readable recording medium, computer program is stored thereon with, the journey
Any described method in first aspect is realized when sequence is executed by processor.
The chat robots that the present embodiment provides reply automatic generation method and system, and generation is returned with reference to priori
Multiple sentence carries out posteriority judgement, by the way that revert statement is compared with the priori of correlation, to incongruent revert statement
It is adjusted, to improve the output quality of revert statement.End-to-end nerve network system or phrase-based system are based on existing
Meter machine translation system is compared, and the dialogue generated by the method for the present embodiment is more nearly real interpersonal right
Words, that is, talk with logically more coherent consistent and significant.
Brief description of the drawings
The chat robots that Fig. 1 is provided by the embodiment of the present invention reply the flow chart of automatic generation method；
The training schematic flow sheet for the answer generation module that Fig. 2 is provided by the embodiment of the present invention；
The chat robots that Fig. 3 is provided by the embodiment of the present invention reply the structured flowchart of automatic creation system.
Embodiment
The embodiment of technical solution of the present invention is described in detail below in conjunction with accompanying drawing.Following examples are only used for
Clearly illustrate technical scheme, therefore be intended only as example, and the protection of the present invention can not be limited with this
Scope.
It should be noted that unless otherwise indicated, technical term or scientific terminology used in this application should be this hair
The ordinary meaning that bright one of ordinary skill in the art are understood.
As shown in figure 1, present embodiments providing a kind of chat robots replys automatic generation method, including：
Step S1, the keyword in the sentence of extraction user's input, is index search in priori using the keyword
Question and answer pair corresponding to being found in storehouse.
Wherein, the priori storehouse is made up of the true question and answer of the Health For All gathered to { s, r }, and wherein s is question sentence,
R answers to reply sentence corresponding to i.e. question sentence s.
Step S2, if finding corresponding question and answer pair, the reply sentence using the question and answer centering is defeated as revert statement
Go out.
Wherein, the semantic similarity that carries out of the question sentence s of question and answer centering in the sentence of user's input and priori storehouse is sentenced
It is disconnected, obtain the revert statement that the reply sentence of the high question and answer centering of similarity exports as robot.
Step S3, if not finding corresponding question and answer pair, the input by sentence that the user is inputted answers generation module,
Obtain revert statement output.
Wherein, it is that the real corpus based on a large amount of Health For Alls trains what is obtained to answer generation module, specific training method
The present embodiment subsequent content is referred to, here is omitted.
The chat robots that the present embodiment provides reply automatic generation method, first pass through with reference to priori library lookup to conjunction
Suitable revert statement, because the question and answer in priori storehouse are to both from the true sentence during Health For All, therefore, machine
The reply that device people is exported by this method complies fully with Human Natural Language；Revert statement can not obtained by priori storehouse
When, the answer generation module generation obtained using training meets the reply of Human Natural Language rule, to improve revert statement
Export quality.With it is existing based on end-to-end nerve network system or phrase-based statictic machine translation system compared with, pass through this
The dialogue of the method generation of embodiment is more nearly real interpersonal dialogue, that is, talks with logically more coherent
It is consistent and significant.
The method of the present embodiment also includes step S4, if not finding corresponding question and answer pair, obtains user's next step
Input, judge whether the revert statement that epicycle exports in talking with is correct, is rectified according to judged result according to the input of user's next step
It is positive to answer generation module.
Step S4 constitutes a negative feedback model, during using generation module is answered, is directed to according to user
Whether the answer that the reply of robot is made is correct to judge the revert statement for answering generation module generation, for example, working as robot
When reply by answering generation module output is incorrect, user can input the sentences such as " you are wrong ", " what you say ", right
Epicycle is replied and made an appraisal, and now it may determine that the revert statement for answering generation module generation is incorrect, is then adjusted back in real time
Generation module is answered, improves constantly the quality for answering generation module generation revert statement.
In robot and user's question answering process, due to the randomness that the mankind speak, equivalent can use different statements
Mode, such as when inquiring the age of robot, may ask " you how old ", or " how old are you ", because robot is
Answer is searched in priori knowledge base using sentence similarity, accordingly, it is possible to two different replies can be found, such as " 17
Year ", " high 50 centimetres, wide 30 centimetres ".At this time, it may be necessary to ensure the uniformity of return information, especially for some fixation general knowledge
And the fixed attribute of robot these information.Therefore, the method for the present embodiment also includes doing following processing to priori storehouse：In advance
Some fixation problems are first set, enumerate a variety of ways to put questions of these fixation problems, and only set unique answer to make to a variety of ways to put questions
For the revert statement of output.
Due to robot using sentence similarity in the lookup answer of priori knowledge base, and the judgement of similarity is to pass through
The probability that word in parsing sentence occurs is typically chosen the sentence output of maximum probability come what is judged.However, in priori storehouse
Some words largely repeat, and this can greatly increase distracter, such as the pronoun such as " you ", " I ", " he ", " uh ", " "
Deng auxiliary words of mood etc. noise word.In order to solve the above problems, the method for the present embodiment, which is additionally included in priori storehouse, searches
During corresponding question and answer pair, ignore noise word, i.e., ignore noise word in the similarity between calculating sentence.
On the basis of any of the above-described embodiment of the method, in order to improve the precision for answering generation module generation revert statement,
The present embodiment largely really talks with language material by obtaining, and obtains true question and answer to { s, r }, wherein, s (asks for input sentence
Sentence), r is the true reply (being answered corresponding to question sentence) of the input sentence, above-mentioned real corpus is divided into two parts, a part
True question and answer form training set to { s, r }, for answering initially setting up into for generation module, the true question and answer of another part to s,
R } composition test set, for answering generation module optimization renewal.It is trained by these dialogue language materials to answering generation module
And renewal.
The training schematic flow sheet for the answer generation module that Fig. 2 is provided by the embodiment of the present invention.Based on above-mentioned pretreatment
Real corpus afterwards, the present embodiment, which is trained and updated using following methods, answers generation module：
Method one：Generation module is answered in method training and renewal based on probability statistics.
First, initial answer generation module is established, { s, the r } in training set is segmented to obtain sijAnd rit, then
Reference resolution and disambiguation, statistical condition probability P (rit|sij) and combination condition probability；Calculate sijAnd ritTerm vector；Obtain just
The answer generation module of beginning.Specifically include following steps：
Step S501, the real dialog language material in training set is made pauses in reading unpunctuated ancient writings, and sample to obtain according to the form of question and answer pair
True question and answer are to { si, ri, wherein, i is expressed as the true question and answer pair of i-th pair.
Wherein, true question and answer to for n dialogue sentences pair from interpersonal real dialog, being designated as { s respectively1,
r1, { s2,r2, { s3,r3..., { sn,rn, wherein sentence riIt is to sentence s in dialogueiReply.
Step S502, identify true question and answer to { si, riIn entity, and part-of-speech tagging is carried out to the entity of identification, obtained
Each { s in training samplei, riCorresponding to question and answer entity to<si1,si2,...,sin>,<ri1,ri2,...,rim>, wherein,
sijFor siIn entity, ritFor riIn entity, j=1,2 ..., n, i=1,2 ..., m.
Step S503, respectively to sijAnd ritCarry out reference resolution and disambiguation.
Step S504, statistical condition probability P (rit|sij) and joint probability.
Step S505, calculate sijAnd ritTerm vector.
Initial answer generation module is obtained by step S501-S505.
Priori storehouse is formed to { s, r } according to the true question and answer that real dialog language material obtains.
Then, initial answer generation module is constantly trained and optimized, specifically include following steps：
Step S601, the true question and answer in test set are obtained to { s ', r ' }.
Step S602, sentence s ' inputs will be inputted and answer generation module, obtained machine and reply R ', composition machine answer pair
S ', R ' }.
Wherein, step S602 preferred embodiment includes：
Step S201, identify the entity in the input sentence s '.
Step S202, to all entities of identification and the question and answer entity that builds in advance to<si1,si2,...,sin>,<ri1,
ri2,...,rim>In entity sijCarry out entity link.Called entity link refers to find in question and answer entity pair and input sentence
The larger entity s of physical correlation in sub- s 'ij。
Step S203, according to entity link result, obtain the conditional probability of statistics.Wherein, the conditional probability of acquisition is step
The entity s being linked in rapid S202ijCorresponding conditional probability P (rit|sij), i.e., entity s is included in sentence s ' is inputtedijBefore
Put, occur r in revert statementitProbability.
Step S204, according to the conditional probability, from the r of the question and answer entity centeringitMiddle sampling obtains composition machine and returned
Multiple R entity.Conditional probability is higher, occurs r in replyitProbability it is higher.
Step S205, the entity generation machine that R ' is replied according to the composition machine of acquisition reply R '.
Wherein, step S205 is according to Human Natural Language, adds suitable conjunction etc., the reality that will be replied R ' and include
Body composition smoothly replys sentence.
Step S603, machine is answered judge module is replied to { s ', R ' } input, calculate the confidence level of { s ', R ' }.
Wherein, reply judge module and be used for the dialogue sentence of judgement input to being caused by person to person's natural interaction or returning
Answer caused by generation module, it is a binary classifier substantially to reply judge module, and the input of this grader is a dialogue
Sentence is to { s, r }, and output label shows that this dialogue sentence is to come from real interpersonal dialogue to { s, r }, still
From answer generation module.Replying the thinking that judge module judges is：First the s and r that are stitched together are entered with scalable coder
Row coding, it is the Probability p from real dialog+{ s, r } that coding then is changed into { s, r } by binary softmax functions, and
It is Probability p-{ s, the r } from answer generation module, the answer that judge module is replied input according to probable value is true/false to marking
Label, the confidence level of { s, r } is generated according to the probability being calculated.Confidence level is the equal of a fraction, and confidence level is this
The normalized result of probability, fraction height are that the possibility of true question and answer pair is just high.
Step S604, generation module is answered according to the confidence level of { s ', R ' } renewal.
Wherein, the specific method of generation module is answered according to the confidence level of { s ', R ' } renewal to be included：According to putting for { s, r }
Reliability, adjust corresponding conditional probability P (rit|sij) and combination condition probability, to update answer generation module.
Method two：Trained and updated based on machine learning and answer generation module.
First, initial answer generation module is established, the true question and answer in training set are inputted into RNN or LSTM to { s, r },
Every words and the sentence vector representation of each pair question and answer pair and question and answer are obtained respectively to vector representation, use RNN or LSTM scheduling algorithms
Practise and obtain initial answer generation module.
Then, true question and answer in test set are obtained to { s ', r ' }, will be defeated after input sentence s ' carries out sentence vector representation
Enter to answer generation module, obtain machine and reply R ', R ' is inputted into judge module, using R ' judged result and confidence level as feedback
Information, updated with adjustment and answer generation module.
Preferably, the present embodiment reply judge module think sentence be to { s, r } Probability p from real dialog+s,
R } as bonus points answer generation module is fed back to, it is trained by this nitrification enhancement to answering generation module,
The target of training is to maximize the bonus points desired value of generation sentence pair.
Obtain largely truly answering to as training sample, continuous repeat step according to the real dialog language material of acquisition
S601-S604, constantly updated using these training samples and answer generation module, its machine exported is replied R and become closer to
Human Natural Language.In the training process, answer generation module is continued to optimize by replying judge module so that answer and produce mould
Reply of the block to a given input sentence is fitted person to person's natural dialogue as far as possible, promotes to answer the reply of generation module output
Sentence can not be responded judge module and judge it is from real dialog or from answer generation module.
It during generation module is trained, while can also be trained, specifically include to replying judge module：By described in
True question and answer are labeled as front training sample to { s, r }, and the machine is answered and is labeled as negative training sample to { s, R }；By band
{ s, the r } and { s, R } of mark is used as training sample, is trained using clustering algorithm and replys judge module.Using training training sample
Judge module is replied in training, is favorably improved the accuracy of judgement degree for replying judge module, more harsh is really sentenced to obtain
Disconnected standard, the output of generation module is answered with supervision.
The training method of above-mentioned answer generation module, by introducing intensified learning mechanism in resisting network in generation, and change
Enter the object function that intensified learning is used, so that the dialogue of generation network generation can be out-tricked with bigger probability and differentiate net
Network (that is, the dialogue of generation is more nearly real dialogue), so as to lift the conversational quality of generation and personification, solves traditional end
The problem of to terminal nerve network system or conversational quality that the machine translation system based on statistics is generated is undesirable.
Based on answer generation module is obtained using the training of method one, step S1 specific implementation includes：
Step S101, identify the entity in the input sentence s.
Step S102, to all entities of identification and the question and answer entity that builds in advance to<si1,si2,...,sin>,<ri1,
ri2,...,rim>In entity sijCarry out entity link.
Step S103, according to entity link result, obtain the conditional probability of statistics.
Step S104, according to the conditional probability, from the r of the question and answer entity centeringitMiddle sampling obtains composition machine and returned
Multiple R entity.
Step S105, the entity generation machine that R is replied according to the composition machine of acquisition reply R.
For training what is obtained to obtain answer generation module using method two, step S1 specific implementation includes：Will
After inputting sentence s progress sentence vector representations, generation module is answered in input, is obtained machine and is replied R.
As shown in figure 3, it is based on replying automatic generation method identical inventive concept, the present embodiment with above-mentioned chat robots
Provide a kind of chat robots and reply automatic creation system, including：
Priori searching unit, the keyword in sentence for extracting user's input, using the keyword as index
Lookup finds corresponding question and answer pair in priori knowledge base；
First reply unit, if for finding corresponding question and answer pair, using the reply sentence of the question and answer centering as
Revert statement exports；
Second replys unit, if for not finding corresponding question and answer pair, the input by sentence that the user is inputted returns
Generation module is answered, obtains revert statement output.
The system of the present embodiment also includes negative-feedback unit, if for not finding corresponding question and answer pair, obtains user
The input of next step, judge whether the revert statement that exports is correct in epicycle dialogue according to the input of user's next step, according to sentencing
Generation module is answered in disconnected result correction.
Wherein, question and answer to, priori storehouse, answer generation module construction method and above method embodiment in structure side
Method is identical, will not be repeated here.
System and the above method that the present embodiment provides have identical beneficial effect for identical inventive concept, this
Place repeats no more.
Based on inventive concept same as mentioned above, a kind of computer-readable recording medium is present embodiments provided, its
On be stored with computer program, the program realizes any described method in the above method embodiment when being executed by processor.
Finally it should be noted that：Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations；To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that：Its according to
The technical scheme described in foregoing embodiments can so be modified, either which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or replacement, the essence of appropriate technical solution is departed from various embodiments of the present invention technology
The scope of scheme, it all should cover among the claim of the present invention and the scope of specification.
Claims (10)
1. a kind of chat robots reply automatic generation method, it is characterised in that including：
The keyword in the sentence of user's input is extracted, is found using the keyword as index search in priori knowledge base pair
The question and answer pair answered；
If finding corresponding question and answer pair, exported the reply sentence of the question and answer centering as revert statement；
If not finding corresponding question and answer pair, the input by sentence that the user is inputted answers generation module, obtains replying language
Sentence output.
2. according to the method for claim 1, it is characterised in that also include：If not finding corresponding question and answer pair, obtain
The input of user's next step, judge whether the revert statement that epicycle exports in talking with is correct according to the input of user's next step, root
It is judged that generation module is answered in result correction.
3. method according to claim 1 or 2, it is characterised in that the training method for answering generation module includes：
The real dialog language material of acquisition is made pauses in reading unpunctuated ancient writings, and samples to obtain true question and answer to { s, r } according to the form of question and answer pair, its
In, s is input sentence, and r is the true reply of the input sentence；
True question and answer are divided into training set and test set two parts to { s, r }；
Initial answer generation module is established by the training set；
The answer generation module is updated by the test set.
4. according to the method for claim 3, it is characterised in that described that initial answer generation is established by the training set
Module, including：
To the true question and answer in training set to { si, riSegmented and identify entity therein, and word is carried out to the entity of identification
Property mark, obtain each { s in training samplei, riCorresponding to question and answer entity to<si1,si2,...,sin>,<ri1,ri2,...,
rim>, wherein, sijFor siIn entity, ritFor riIn entity, j=1,2 ..., n, i=1,2 ..., m；
Respectively to sijAnd ritCarry out reference resolution and disambiguation；
Statistical condition probability P (rit|sij) and joint probability；
Calculate sijAnd ritTerm vector；
Obtain initial answer generation module.
5. according to the method for claim 3, it is characterised in that described that the answer generation mould is updated by the test set
Block, including：
The true question and answer in test set are obtained to { s ', r ' }；
Answer input sentence s ' input to generation module, obtain machine and reply R ', composition machine is answered to { s ', R ' }；
Machine is answered judge module is replied to { s ', R ' } input, calculate the confidence of { s ', R ' }；
Generation module is answered according to the confidence level of { s ', R ' } renewal.
6. according to the method for claim 3, it is characterised in that also include during the generation module is trained：
The true question and answer be labeled as front training sample to { s ', r ' }, the machine is answered to { s ', R ' } labeled as negatively
Training sample；
{ s ', the r ' } of tape label and { s ', R ' } is used as training sample, sentenced using clustering algorithm training reply.
7. according to the method for claim 6, it is characterised in that the confidence level renewal of the basis { s, r }, which is answered, produces mould
Block, including：
According to the confidence level of { s ', R ' }, corresponding conditional probability P (r are adjustedit|sij) and joint probability, answered with renewal and produce mould
Block.
8. a kind of chat robots reply automatic creation system, it is characterised in that including：
Priori searching unit, the keyword in sentence for extracting user's input, using the keyword as index search
Question and answer pair corresponding to being found in priori knowledge base；
First replys unit, if for finding corresponding question and answer pair, using the reply sentence of the question and answer centering as reply
Sentence exports；
Second replys unit, if for not finding corresponding question and answer pair, the input by sentence that the user is inputted answers production
Raw module, obtain revert statement output.
9. system according to claim 8, it is characterised in that also including negative-feedback unit, if corresponding for not finding
Question and answer pair, then obtain the input of user's next step, according to the input of user's next step judge epicycle talk with the reply that exports
Whether sentence is correct, is corrected according to judged result and answers generation module.
10. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the program is by processor
The method described in one of claim 1-8 is realized during execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710760328.7A CN107562863A (en) | 2017-08-30 | 2017-08-30 | Chat robots reply automatic generation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710760328.7A CN107562863A (en) | 2017-08-30 | 2017-08-30 | Chat robots reply automatic generation method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107562863A true CN107562863A (en) | 2018-01-09 |
Family
ID=60977714
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710760328.7A Pending CN107562863A (en) | 2017-08-30 | 2017-08-30 | Chat robots reply automatic generation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107562863A (en) |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710647A (en) * | 2018-04-28 | 2018-10-26 | 苏宁易购集团股份有限公司 | A kind of data processing method and device for chat robots |
| CN108846124A (en) * | 2018-06-29 | 2018-11-20 | 北京百度网讯科技有限公司 | Training method, device, computer equipment and readable storage medium storing program for executing |
| CN108920603A (en) * | 2018-06-28 | 2018-11-30 | 厦门快商通信息技术有限公司 | A kind of customer service bootstrap technique based on customer service machine mould |
| CN109033336A (en) * | 2018-07-20 | 2018-12-18 | 吴怡 | Legal advice robot and business model based on artificial intelligence |
| CN109086372A (en) * | 2018-07-20 | 2018-12-25 | 吴怡 | It is a kind of based on artificial intelligence application in the more wheel session methods and system of legal advice |
| CN109145106A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | A kind of self-learning type legal advice robot |
| CN109145086A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | It is a kind of for legal advice and intelligent robot easy to spread and its business model |
| CN109146725A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | It is a kind of for support robot provide legal advice service cloud platform |
| CN109192300A (en) * | 2018-08-17 | 2019-01-11 | 百度在线网络技术（北京）有限公司 | Intelligent way of inquisition, system, computer equipment and storage medium |
| CN109241254A (en) * | 2018-08-06 | 2019-01-18 | 深圳市玖胜云智联科技有限公司 | A kind of corpus acquisition method and acquisition device applied to robot |
| CN109284383A (en) * | 2018-10-09 | 2019-01-29 | 北京来也网络科技有限公司 | Text handling method and device |
| CN109299320A (en) * | 2018-10-30 | 2019-02-01 | 上海智臻智能网络科技股份有限公司 | A kind of information interacting method, device, computer equipment and storage medium |
| CN109547320A (en) * | 2018-09-29 | 2019-03-29 | 阿里巴巴集团控股有限公司 | Social contact method, device and equipment |
| CN109710740A (en) * | 2018-12-27 | 2019-05-03 | 杭州美平米科技有限公司 | A kind of robot automatic chatting method based on merchandise news |
| CN110135551A (en) * | 2019-05-15 | 2019-08-16 | 西南交通大学 | A kind of robot chat method of word-based vector sum Recognition with Recurrent Neural Network |
| CN110222164A (en) * | 2019-06-13 | 2019-09-10 | 腾讯科技（深圳）有限公司 | A kind of Question-Answering Model training method, problem sentence processing method, device and storage medium |
| CN110399489A (en) * | 2019-07-08 | 2019-11-01 | 厦门市美亚柏科信息股份有限公司 | A kind of chat data segmentation method, device and storage medium |
| CN110543552A (en) * | 2019-09-06 | 2019-12-06 | 网易（杭州）网络有限公司 | Conversation interaction method and device and electronic equipment |
| CN110597948A (en) * | 2019-07-11 | 2019-12-20 | 东华大学 | Entity relation extraction method based on deep learning |
| CN110705206A (en) * | 2019-09-23 | 2020-01-17 | 腾讯科技（深圳）有限公司 | Text information processing method and related device |
| WO2020020041A1 (en) * | 2018-07-27 | 2020-01-30 | 北京京东尚科信息技术有限公司 | Sentence processing method and system and electronic device |
| CN110866101A (en) * | 2019-11-07 | 2020-03-06 | 昆明理工大学 | Dialogue generation method based on near-end strategy optimization and counterstudy |
| CN110929085A (en) * | 2019-11-14 | 2020-03-27 | 国家电网有限公司 | Power customer service message generation model sample processing system and method based on meta-semantic decomposition |
| CN111090736A (en) * | 2018-10-24 | 2020-05-01 | 马上消费金融股份有限公司 | Question-answering model training method, question-answering method, device and computer storage medium |
| CN111125330A (en) * | 2019-12-19 | 2020-05-08 | 厦门快商通科技股份有限公司 | Intelligent question-answering method, device and equipment based on man-machine conversation |
| CN111143527A (en) * | 2019-12-19 | 2020-05-12 | 厦门快商通科技股份有限公司 | Automatic question answering method, device and equipment based on man-machine conversation |
| CN111401036A (en) * | 2020-05-28 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Method and device for generating reference resolution text and electronic equipment |
| CN111553171A (en) * | 2020-04-09 | 2020-08-18 | 北京小米松果电子有限公司 | Corpus processing method, apparatus and storage medium |
| CN111768784A (en) * | 2019-03-15 | 2020-10-13 | 上海火商智能科技有限公司 | Voice generation method based on generation type countermeasure network |
| CN111813911A (en) * | 2020-06-30 | 2020-10-23 | 神思电子技术股份有限公司 | Knowledge automatic acquisition and updating system based on user supervision feedback and working method thereof |
| CN112597287A (en) * | 2020-12-15 | 2021-04-02 | 深圳市优必选科技股份有限公司 | Statement processing method, statement processing device and intelligent equipment |
| CN113569019A (en) * | 2021-05-24 | 2021-10-29 | 恩亿科(北京)数据科技有限公司 | Method, system, equipment and storage medium for extracting knowledge based on chat conversation |
| CN114510557A (en) * | 2022-01-21 | 2022-05-17 | 民商数字科技(深圳)有限公司 | Automatic return robot and control method thereof |
| CN114970472A (en) * | 2022-04-29 | 2022-08-30 | 宁波火箭数据科技有限公司 | Work order operating system, computer device and readable storage medium |
| CN113569019B (en) * | 2021-05-24 | 2024-04-30 | 恩亿科(北京)数据科技有限公司 | Method, system, equipment and storage medium for knowledge extraction based on chat conversation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101566998A (en) * | 2009-05-26 | 2009-10-28 | 华中师范大学 | Chinese question-answering system based on neural network |
| CN106663128A (en) * | 2016-06-29 | 2017-05-10 | 深圳狗尾草智能科技有限公司 | Extended learning method of chat dialogue system and chat conversation system |
| CN106843461A (en) * | 2016-12-13 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of interactive output intent and robot for robot |
-
2017
- 2017-08-30 CN CN201710760328.7A patent/CN107562863A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101566998A (en) * | 2009-05-26 | 2009-10-28 | 华中师范大学 | Chinese question-answering system based on neural network |
| CN106663128A (en) * | 2016-06-29 | 2017-05-10 | 深圳狗尾草智能科技有限公司 | Extended learning method of chat dialogue system and chat conversation system |
| CN106843461A (en) * | 2016-12-13 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of interactive output intent and robot for robot |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710647A (en) * | 2018-04-28 | 2018-10-26 | 苏宁易购集团股份有限公司 | A kind of data processing method and device for chat robots |
| CN108710647B (en) * | 2018-04-28 | 2023-12-01 | 苏宁易购集团股份有限公司 | Data processing method and device for chat robot |
| CN108920603A (en) * | 2018-06-28 | 2018-11-30 | 厦门快商通信息技术有限公司 | A kind of customer service bootstrap technique based on customer service machine mould |
| CN108920603B (en) * | 2018-06-28 | 2021-12-21 | 厦门快商通信息技术有限公司 | Customer service guiding method based on customer service machine model |
| CN108846124A (en) * | 2018-06-29 | 2018-11-20 | 北京百度网讯科技有限公司 | Training method, device, computer equipment and readable storage medium storing program for executing |
| CN108846124B (en) * | 2018-06-29 | 2022-05-17 | 北京百度网讯科技有限公司 | Training method, training device, computer equipment and readable storage medium |
| CN109145106A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | A kind of self-learning type legal advice robot |
| CN109145106B (en) * | 2018-07-20 | 2022-06-28 | 重庆百事得大牛机器人有限公司 | Self-learning type law consultation robot |
| CN109033336B (en) * | 2018-07-20 | 2021-12-10 | 重庆百事得大牛机器人有限公司 | Legal consultation robot based on artificial intelligence and business model |
| CN109086372B (en) * | 2018-07-20 | 2021-12-10 | 重庆百事得大牛机器人有限公司 | Artificial intelligence-based multi-round conversation method and system applied to legal consultation |
| CN109033336A (en) * | 2018-07-20 | 2018-12-18 | 吴怡 | Legal advice robot and business model based on artificial intelligence |
| CN109145086A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | It is a kind of for legal advice and intelligent robot easy to spread and its business model |
| CN109086372A (en) * | 2018-07-20 | 2018-12-25 | 吴怡 | It is a kind of based on artificial intelligence application in the more wheel session methods and system of legal advice |
| CN109146725A (en) * | 2018-07-20 | 2019-01-04 | 吴怡 | It is a kind of for support robot provide legal advice service cloud platform |
| WO2020020041A1 (en) * | 2018-07-27 | 2020-01-30 | 北京京东尚科信息技术有限公司 | Sentence processing method and system and electronic device |
| CN109241254A (en) * | 2018-08-06 | 2019-01-18 | 深圳市玖胜云智联科技有限公司 | A kind of corpus acquisition method and acquisition device applied to robot |
| CN109192300A (en) * | 2018-08-17 | 2019-01-11 | 百度在线网络技术（北京）有限公司 | Intelligent way of inquisition, system, computer equipment and storage medium |
| CN109547320A (en) * | 2018-09-29 | 2019-03-29 | 阿里巴巴集团控股有限公司 | Social contact method, device and equipment |
| CN109284383A (en) * | 2018-10-09 | 2019-01-29 | 北京来也网络科技有限公司 | Text handling method and device |
| CN111090736A (en) * | 2018-10-24 | 2020-05-01 | 马上消费金融股份有限公司 | Question-answering model training method, question-answering method, device and computer storage medium |
| CN109299320A (en) * | 2018-10-30 | 2019-02-01 | 上海智臻智能网络科技股份有限公司 | A kind of information interacting method, device, computer equipment and storage medium |
| CN109299320B (en) * | 2018-10-30 | 2020-09-25 | 上海智臻智能网络科技股份有限公司 | Information interaction method and device, computer equipment and storage medium |
| CN109710740B (en) * | 2018-12-27 | 2020-08-07 | 杭州美平米科技有限公司 | Robot automatic chat method based on commodity information |
| CN109710740A (en) * | 2018-12-27 | 2019-05-03 | 杭州美平米科技有限公司 | A kind of robot automatic chatting method based on merchandise news |
| CN111768784A (en) * | 2019-03-15 | 2020-10-13 | 上海火商智能科技有限公司 | Voice generation method based on generation type countermeasure network |
| CN110135551A (en) * | 2019-05-15 | 2019-08-16 | 西南交通大学 | A kind of robot chat method of word-based vector sum Recognition with Recurrent Neural Network |
| CN110222164A (en) * | 2019-06-13 | 2019-09-10 | 腾讯科技（深圳）有限公司 | A kind of Question-Answering Model training method, problem sentence processing method, device and storage medium |
| CN110222164B (en) * | 2019-06-13 | 2022-11-29 | 腾讯科技（深圳）有限公司 | Question-answer model training method, question and sentence processing device and storage medium |
| CN110399489B (en) * | 2019-07-08 | 2022-06-17 | 厦门市美亚柏科信息股份有限公司 | Chat data segmentation method, device and storage medium |
| CN110399489A (en) * | 2019-07-08 | 2019-11-01 | 厦门市美亚柏科信息股份有限公司 | A kind of chat data segmentation method, device and storage medium |
| CN110597948A (en) * | 2019-07-11 | 2019-12-20 | 东华大学 | Entity relation extraction method based on deep learning |
| CN110543552A (en) * | 2019-09-06 | 2019-12-06 | 网易（杭州）网络有限公司 | Conversation interaction method and device and electronic equipment |
| CN110705206A (en) * | 2019-09-23 | 2020-01-17 | 腾讯科技（深圳）有限公司 | Text information processing method and related device |
| CN110866101B (en) * | 2019-11-07 | 2022-11-01 | 昆明理工大学 | Dialogue generation method based on near-end strategy optimization and counterstudy |
| CN110866101A (en) * | 2019-11-07 | 2020-03-06 | 昆明理工大学 | Dialogue generation method based on near-end strategy optimization and counterstudy |
| CN110929085B (en) * | 2019-11-14 | 2023-12-19 | 国家电网有限公司 | System and method for processing electric customer service message generation model sample based on meta-semantic decomposition |
| CN110929085A (en) * | 2019-11-14 | 2020-03-27 | 国家电网有限公司 | Power customer service message generation model sample processing system and method based on meta-semantic decomposition |
| CN111143527A (en) * | 2019-12-19 | 2020-05-12 | 厦门快商通科技股份有限公司 | Automatic question answering method, device and equipment based on man-machine conversation |
| CN111125330A (en) * | 2019-12-19 | 2020-05-08 | 厦门快商通科技股份有限公司 | Intelligent question-answering method, device and equipment based on man-machine conversation |
| CN111553171A (en) * | 2020-04-09 | 2020-08-18 | 北京小米松果电子有限公司 | Corpus processing method, apparatus and storage medium |
| CN111553171B (en) * | 2020-04-09 | 2024-02-06 | 北京小米松果电子有限公司 | Corpus processing method, corpus processing device and storage medium |
| CN111401036B (en) * | 2020-05-28 | 2020-09-25 | 支付宝(杭州)信息技术有限公司 | Method and device for generating reference resolution text and electronic equipment |
| CN111401036A (en) * | 2020-05-28 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Method and device for generating reference resolution text and electronic equipment |
| CN111813911A (en) * | 2020-06-30 | 2020-10-23 | 神思电子技术股份有限公司 | Knowledge automatic acquisition and updating system based on user supervision feedback and working method thereof |
| CN112597287A (en) * | 2020-12-15 | 2021-04-02 | 深圳市优必选科技股份有限公司 | Statement processing method, statement processing device and intelligent equipment |
| CN112597287B (en) * | 2020-12-15 | 2024-03-15 | 深圳市优必选科技股份有限公司 | Statement processing method, statement processing device and intelligent equipment |
| CN113569019A (en) * | 2021-05-24 | 2021-10-29 | 恩亿科(北京)数据科技有限公司 | Method, system, equipment and storage medium for extracting knowledge based on chat conversation |
| CN113569019B (en) * | 2021-05-24 | 2024-04-30 | 恩亿科(北京)数据科技有限公司 | Method, system, equipment and storage medium for knowledge extraction based on chat conversation |
| CN114510557A (en) * | 2022-01-21 | 2022-05-17 | 民商数字科技(深圳)有限公司 | Automatic return robot and control method thereof |
| CN114970472A (en) * | 2022-04-29 | 2022-08-30 | 宁波火箭数据科技有限公司 | Work order operating system, computer device and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107562863A (en) | Chat robots reply automatic generation method and system | |
| CN104050160B (en) | Interpreter's method and apparatus that a kind of machine is blended with human translation | |
| CN106295792B (en) | Dialogue data interaction processing method and device based on multi-model output | |
| CN110717018A (en) | Industrial equipment fault maintenance question-answering system based on knowledge graph | |
| CN107301168A (en) | Intelligent robot and its mood exchange method, system | |
| CN110287482B (en) | Semi-automatic participle corpus labeling training device | |
| CN109325780A (en) | A kind of exchange method of the intelligent customer service system in E-Governance Oriented field | |
| CN112270182A (en) | Analysis idea derivation method, system and storage medium based on multi-turn dialogue question-answer mode | |
| CN109460459A (en) | A kind of conversational system automatic optimization method based on log study | |
| CN106326307A (en) | Language interaction method | |
| CN108388944B (en) | A kind of automatic chatting method and robot based on deep neural network | |
| CN108595609A (en) | Generation method, system, medium and equipment are replied by robot based on personage IP | |
| CN116010581A (en) | Knowledge graph question-answering method and system based on power grid hidden trouble shooting scene | |
| CN108628908A (en) | The method, apparatus and electronic equipment of sorted users challenge-response boundary | |
| CN113326367B (en) | Task type dialogue method and system based on end-to-end text generation | |
| CN113705207A (en) | Grammar error recognition method and device | |
| Chowanda et al. | Generative Indonesian conversation model using recurrent neural network with attention mechanism | |
| CN110826341A (en) | Semantic similarity calculation method based on seq2seq model | |
| CN114661864A (en) | Psychological consultation method and device based on controlled text generation and terminal equipment | |
| CN114692615A (en) | Small sample semantic graph recognition method for small languages | |
| CN110991637B (en) | Social network data extraction method and system of company image enhancement system | |
| CN111460106A (en) | Information interaction method, device and equipment | |
| CN117494727B (en) | De-biasing method for large language model | |
| CN112818108B (en) | Text semantic misinterpretation chat robot based on shape and near words and data processing method thereof | |
| CN110489529B (en) | Dialogue generating method based on syntactic structure and reordering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180109 |