CN110232108A - Interactive method and conversational system - Google Patents
Interactive method and conversational system Download PDFInfo
- Publication number
- CN110232108A CN110232108A CN201910395865.5A CN201910395865A CN110232108A CN 110232108 A CN110232108 A CN 110232108A CN 201910395865 A CN201910395865 A CN 201910395865A CN 110232108 A CN110232108 A CN 110232108A
- Authority
- CN
- China
- Prior art keywords
- user
- sentence
- behavior
- user behavior
- conversational system
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- 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
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
Abstract
This application involves artificial intelligence technologys, provide a kind of interactive method, comprising: conversational system obtains first user's sentence of user's input, wherein first user's sentence indicates the first user behavior;Determine that there are first user behavior be abnormal behaviour according to first user's sentence;It is that conversational system can identify in abnormal behaviour, but in the case where the behavior that can not be executed, the reason of abnormal behaviour can not be performed is identified, and generate the first revert statement replied abnormal behaviour, wherein, the reason of the first revert statement is for prompting the first user behavior that cannot be performed;Or, in the case where abnormal behaviour is the unrecognized behavior of conversational system, identify that user is intended to according to first user's sentence, and generate the second revert statement replied abnormal behaviour, wherein, the second revert statement is for prompting the conversational system not support user to be intended to.
Description
Technical field
This application involves artificial intelligence field more particularly to a kind of interactive method and conversational systems.
Background technique
Conversational system may include two kinds at present: Task conversational system and chat-type conversational system.Wherein, Task
Dialogue refers to the dialogue carried out to complete particular task.In Task dialogue, conversational system is needed according to user's sentence
Understand the intention of user, extracts the information in user's sentence, and the task that user specifies is completed according to the information extracted.
But those skilled in the art is in the lower discovery that studies for a long period of time, robust of the conversational system to the processing of abnormal behaviour
Property is excessively poor.
Summary of the invention
This application provides a kind of interactive method and conversational systems, can clearly prompt user session system can not
The reason of executing abnormal behaviour, to improve the robustness of processing of the conversational system to abnormal behaviour.
In a first aspect, providing a kind of interactive method, comprising:
Conversational system obtains first user's sentence of the first user behavior of instruction of user's input.Conversational system is according to
First user's sentence determines that first user behavior is abnormal behaviour by Activity recognition model.Wherein, the abnormal behaviour
It is the behavior that the conversational system can not be handled normally.In the abnormal behaviour it is that the conversational system can identify, but nothing
In the case where the behavior that method executes, conversational system identifies the reason of abnormal behaviour can not be performed, and generates to described different
The first revert statement that Chang Hangwei is replied.Wherein, first revert statement is for prompting first user behavior not
The reason of capable of being performed.Alternatively, in the case where the abnormal behaviour is the unrecognized behavior of the conversational system, dialogue system
System identifies that user is intended to according to the first user sentence, and generates second replied the abnormal behaviour and reply language
Sentence.Wherein, second revert statement is for prompting the conversational system that the user is not supported to be intended to.
In above scheme, user, which is clear, knows that conversational system is unable to complete from the revert statement that conversational system exports
The reason of user behavior, the direction of adjustment dialogue in time, so that dialogue is gone on smoothly.
In some possible designs, conversational system can be generated by the following method and be replied the abnormal behaviour
The first revert statement:
The reason of conversational system can not be performed according to the abnormal behaviour searches corresponding first template.Wherein, described
First template contains associated first slot position of first user behavior.Conversational system is obtained for filling first slot position
The first slot position information.The first slot position information is inserted first slot position by conversational system, is obtained described first and is replied language
Sentence.
In some possible designs, conversational system searches the second template including the second slot position.Conversational system obtains institute
State the first association user behavior of the first user behavior, wherein the first association user behavior is and the first user row
For the degree of correlation be greater than threshold value user behavior.The first association user behavior is inserted second slot position by conversational system,
Obtain third revert statement, wherein the third revert statement is closed for prompting the conversational system to be able to carry out described first
Join user behavior.
In above scheme, conversational system can clearly indicate and user's sentence before according to user's sentence before
Therefore the user behavior that relevant conversational system can be supported can guide the direction of user session well, so that dialogue energy
It is enough more clearly to carry out.
In some possible designs, conversational system can be generated by the following method and be replied the abnormal behaviour
The second revert statement: conversational system search third template, wherein the third template contains third slot position;By the use
Family is intended to insert the third slot position, obtains second revert statement.
In some possible designs, conversational system searches the second template including the second slot position.Conversational system, which obtains, to be used
The second user sentence before the first user sentence of family input.Wherein, the second user sentence indicates second
User behavior.Conversational system obtains the second association user behavior of the second user behavior.Wherein, second association user
Behavior is the user behavior for being greater than threshold value with the degree of correlation of the second user behavior.Conversational system is by second association user
Second slot position is inserted in behavior, obtains the 4th revert statement.Wherein, the 4th revert statement is for prompting the dialogue system
System is able to carry out the second association user behavior.
In above scheme, conversational system can clearly indicate and active user's sentence phase according to current user's sentence
Therefore the user behavior that the conversational system of pass can be supported can guide the direction of user session well, so that dialogue can
More clearly carry out.
In some possible designs, conversational system identifies user's meaning according to the first user sentence in the following manner
Figure:
First user input by sentence user's intention assessment model is intended to by conversational system to obtain the user,
Wherein, user's intention assessment model is trained by known users sentence and known users intention.Its
In, user's intention assessment model is classifier, the classifier be k nearest neighbor classifier, support vector machines, decision tree and
Any one in random forest.
Second aspect provides a kind of conversational system, including input module, determining module and reply module.
The input module is used to obtain first user's sentence of user's input, wherein the first user sentence instruction
First user behavior.
The determining module is used to determine first user by Activity recognition model according to the first user sentence
Behavior is abnormal behaviour, wherein the abnormal behaviour is the behavior that the conversational system can not be handled normally.
The reply module is used in the abnormal behaviour be the row that the conversational system can be identified, but can not be executed
For in the case where, the reason of abnormal behaviour can not be performed is identified, and generate the replied the abnormal behaviour
One revert statement, wherein first revert statement is for the reason of prompting first user behavior that cannot be performed.Or
Person,
The reply module is also used in the case where the abnormal behaviour is the unrecognized behavior of the conversational system,
It identifies that user is intended to according to the first user sentence, and generates the second revert statement replied the abnormal behaviour,
Wherein, second revert statement is for prompting the conversational system that the user is not supported to be intended to.
In some possible designs, the reply module is also used to the reason of can not being performed according to the abnormal behaviour
Search corresponding first template, wherein first template contains associated first slot position of first user behavior;Pass through
Slot position identification model obtains the first slot position information for filling first slot position;It will be described in the first slot position information filling
First slot position obtains first revert statement.
In some possible designs, the reply module is also used to search the second template, wherein the second template packet
Include the second slot position;Obtain the first association user behavior of first user behavior, wherein the first association user behavior is
It is greater than the user behavior of threshold value with the degree of correlation of first user behavior;By the first association user behavior filling described the
Two slot positions obtain third revert statement, wherein the third revert statement is described for prompting the conversational system to be able to carry out
First association user behavior.
In some possible designs, the reply module is also used to search third template, wherein the third template packet
Third slot position is contained;The user is intended to insert the third slot position, obtains second revert statement.
In some possible designs, the reply module is also used to search the second template, wherein the second template packet
Include the second slot position;Obtain the second user sentence of user's input, wherein the second user sentence is in the first user sentence
Before, the second user sentence indicates second user behavior;Obtain the second association user row of the second user behavior
For, wherein the second association user behavior is the user behavior for being greater than threshold value with the degree of correlation of the second user behavior;It will
Second slot position is inserted in the second association user behavior, to obtain the 4th revert statement, wherein the described 4th replys language
To the second user behavior relevant behavior of the sentence for prompting the conversational system to be able to carry out.
In some possible designs, the reply module is also used to be intended to know by the first user input by sentence user
Other model is intended to obtain the user, wherein user's intention assessment model is by known users sentence and
Know what user's intention was trained.Wherein, user's intention assessment model is classifier, and the classifier is k nearest neighbor
Any one in classifier, support vector machines, decision tree and random forest.
The third aspect provides a kind of computer program product, when the computer program product is read by calculating equipment
And when executing, as the described in any item methods of first aspect will be performed.
Fourth aspect provides a kind of computer non-transitory storage media, including instruction, when described instruction is calculating equipment
When upper operation, so that the calculating equipment executes such as the described in any item methods of first aspect.
5th aspect, provides a kind of intelligent terminal, including processor and memory, and the processor executes the storage
Code in device executes such as the described in any item methods of first aspect.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application or in background technique below will be implemented the application
Attached drawing needed in example or background technique is illustrated.
Fig. 1 be this application involves a kind of Task dialogue application scenarios schematic diagram;
Fig. 2 be this application involves the first situation abnormal behaviour Task dialogue schematic diagram;
Fig. 3 be this application involves second situation abnormal behaviour Task dialogue schematic diagram;
Fig. 4 be this application involves the abnormal behaviour for the first situation using showing of being replied of the first revert statement
It is intended to;
Fig. 5 be this application involves the abnormal behaviour for the first situation language is replied using the first revert statement and third
The schematic diagram that sentence is replied;
Fig. 6 be this application involves the abnormal behaviour for second situation using showing of being replied of the second revert statement
It is intended to;
Fig. 7 be this application involves the abnormal behaviour for second situation using the second revert statement and the 4th reply language
The schematic diagram that sentence is replied;
Fig. 8 is a kind of structural schematic diagram for calculating equipment provided by the present application;
Fig. 9 is a kind of structural schematic diagram of computing cluster provided by the present application;
Figure 10 is a kind of structural schematic diagram of conversational system provided by the present application.
Specific embodiment
The term that embodiments herein part uses is only used for explaining specific embodiments of the present invention, rather than purport
Limiting the present invention.
Refering to fig. 1, Fig. 1 be this application involves a kind of Task dialogue application scenarios schematic diagram.In the applied field
Under scape, user and conversational system carry out Task dialogue.As shown in Figure 1, being user's sentence of user's input, Fig. 1 on the right of Fig. 1
The left side be revert statement that conversational system is exported according to user's sentence.Firstly, user inputs user's sentence, " Pekinese Ding Qu flies
Air ticket.", then, and conversational system output revert statement " it is good, 8 points of tomorrow is found from Shenzhen to Pekinese's air ticket for you, you need
Subscribe? ".Then, user inputs user's sentence again and " me is helped to subscribe.", conversational system output revert statement " it is good,
It makes a reservation for you.May I ask can I help you there are also what? ".User inputs user's sentence " without thanks " again.It should be understood that
Application scenarios shown in FIG. 1 are only the example of one of Task dialogue, and in practical applications, Task dialogue can be with
Be about the dialogue made a phone call, about querying geographical position dialogue, about order take-away dialogue, about inquiry weather dialogue
And it about the dialogue etc. for ordering hotel, is not especially limited herein.
In Task dialogue, user behavior is generally comprised in user's sentence of user's input.Wherein, the user behavior
It is the behavior that user claims to conversational system.By taking user's sentence " Pekinese Ding Qu air ticket " as an example, user's sentence is to right
Telephone system proposes the requirement for ordering air ticket, therefore, user behavior " ordering air ticket " is contained in user's sentence.It should be understood that above-mentioned
The citing of user behavior is merely possible to a kind of example, and in other embodiments, user behavior can also be " making a phone call ", " look into
Ask geographical location ", " ordering take-aways ", " inquire weather " and " ordering hotel " etc., be not especially limited herein.
User behavior can be what conversational system identified user's input by sentence Activity recognition model.It is specific one
In embodiment, Activity recognition model can be classifier, for example, k nearest neighbor classifier, support vector machines, decision tree and random
Forest etc. is not especially limited herein.Activity recognition model can according to conversational system support user behavior to classify into
Row divides.For example, the user behavior that conversational system is supported include " ordering air ticket ", " making a phone call ", " ordering take-aways ", " inquire weather " with
And " ordering hotel ", then, the classification of behavior identification model just includes " ordering air ticket ", " making a phone call ", " ordering take-away ", " inquiry day
Gas " and " ordering hotel ".User's sentence " Pekinese Ding Qu air ticket " that Activity recognition model is inputted according to user, can determine and answer
This is included into " ordering air ticket " classification, to identify that the user behavior that user's sentence includes is " ordering air ticket ".
User's sentence can also include slot position information.Wherein, slot position information is for filling slot relevant to user behavior
Value.By taking user behavior " ordering air ticket " as an example, slot position relevant to " ordering air ticket " includes departure time, departure place and purpose
Place.Corresponding to above-mentioned slot position relevant to " ordering air ticket ", user's sentence " stipulating 10 points of day from Shenzhen to Pekinese's air ticket "
In, slot position information includes departure time=" 10 points of tomorrow ", departure place=" Shenzhen " and destination=" Beijing " etc..
Conversational system needs all to fill slot position relevant to user behavior completely, could execute user behavior.For example illustrate, talk with
System needs that " air ticket will be ordered " that relevant slot position departure time, departure place and destination are all filled out and be filled with, and could execute
" ordering air ticket " this user behavior.
User's input by sentence slot position identification model can be identified slot position information by conversational system.In a specific implementation
In example, slot position identification model can be conventional machines learning model, such as conditional random field models, cyclic convolution network, or
Dictionary search method etc., is not especially limited herein.Slot position identification model can extract the crucial letter provided in user session
Breath.For example, order air ticket type of slots include " departure place " and " destination ", slot position identification model need extraction " departure place " and
The information of " destination ".User's sentence " I will order Beijing to the air ticket in Shanghai " that slot position identification model is inputted according to user is known
It does not obtain the result is that " departure place: Beijing " and " destination: Shanghai ", so that conversational system be given to provide slot position information.
It is appreciated that above-mentioned Activity recognition model and slot position identification model can be two different models respectively, it can also
To be integrated in the same model, it is not especially limited herein.
User behavior can be divided into normal behaviour and abnormal behaviour two major classes.Wherein, conversational system is based on user behavior
Whether belong to conversational system support behavior and user behavior whether can be performed distinguish user behavior be normal behaviour also
It is abnormal behaviour.If user behavior is the behavior for belonging to conversational system support, also, user behavior can be performed, then should
User behavior is normal behaviour.Conversely, the user behavior is abnormal behaviour.Wherein, abnormal behaviour may include two kinds: first
Kind, abnormal behaviour belongs to conversational system support, still, it is not possible to the user behavior being performed.Second, abnormal behaviour is
It is not belonging to the behavior of conversational system support.
Conversational system can be judged by the following manner whether user behavior is the behavior for belonging to conversational system support: continue
By taking classifier as an example, it is assumed that user behavior is divided into " ordering air ticket ", " making a phone call ", " ordering take-away ", " inquiry weather " by classifier
And " ordering hotel " etc. multiple classification.So, if user's sentence of user's input is " ordering high guaranteed votes back ", classify
Device can not be classified to any of the above classification, at this point, conversational system may determine that user behavior is not to belong to conversational system
The behavior of support.Conversely, conversational system may determine that user behavior is the behavior for belonging to conversational system support.
Conversational system can be judged by the following manner whether user behavior can be performed." day is stipulated with user's sentence
10 points from Shenzhen to Pekinese's air ticket " for, user behavior is " ordering air ticket ", and slot position information includes departure time=" tomorrow ten
Point ", departure place=" Shenzhen " and destination=" Beijing ".But conversational system is found after scanning for, and is only searched
8 points of tomorrow is from Shenzhen to Pekinese's flight, that is, and the slot position information of user's sentence can not be matched with search result, then,
Conversational system judges that user behavior can not be performed, conversely, then conversational system judges that user behavior can be performed.
When user behavior is normal behaviour, conversational system normally exports revert statement, specifically may refer to Fig. 1, herein
Not reinflated description.
When user behavior is abnormal behaviour, it is all so that conversational system can export " not understanding " " the problem of cannot answering you "
The revert statement of class.Corresponding to two kinds of above-mentioned abnormal behaviours, conversational system will be discussed in detail respectively below in both exceptions
The revert statement exported under behavior.
In the first case, abnormal behaviour belongs to conversational system support, still, it is not possible to the behavior being performed.
As shown in Fig. 2, being user's sentence of user's input on the right of Fig. 2, the left side of Fig. 2 is the revert statement of conversational system.Firstly, with
Family input user's sentence " stipulates the Pekinese Tian Qu plane ticket.", then, conversational system determines that user behavior " ordering air ticket " is to belong to
The behavior that conversational system is supported, finds 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket according to the user behavior, and defeated
Revert statement " good, to find 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket for you, you need to subscribe " out.User after
Continuous input user's sentence " me is helped to change the plane ticket of a Zhang Shidian ".Conversational system determines that user behavior " ordering air ticket " is to belong to dialogue
The behavior that system is supported, still, conversational system find that slot position information departure time=" 10 points of tomorrow " can not be with search result
Match, conversational system exports revert statement " sorry, I am not as understanding ".User can not determine that conversational system is unable to complete user
The reason of behavior, it is not known that talk with how this continues, dialogue is caused to fail.
In second situation, abnormal behaviour is to be not belonging to the behavior of conversational system support.As shown in figure 3, the right of Fig. 3
It is user's sentence of user's input, the left side of Fig. 3 is the revert statement of conversational system.It " is stipulated firstly, user inputs user's sentence
The Pekinese Tian Qu plane ticket ", then, conversational system determine that the behavior of " ordering air ticket " is to belong to the behavior of User support, according to this
User behavior finds 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket, and export revert statement " it is good, found for you bright
8 points of its morning from Shenzhen to Pekinese's plane ticket, you need subscribe ".User continues input user's sentence and " first looks at the day after tomorrow
Either with or without high guaranteed votes back ", still, conversational system determination is to be not belonging to the behavior that conversational system is supported then to export back
Multiple sentence " sorry, I am not as understanding ".User can not determine the reason of conversational system is unable to complete user behavior, it is not known that
Talk with how this continues, dialogue is caused to fail.
In the above method, when user behavior is abnormal behaviour, conversational system only can simply export " not understanding " " cannot
The suchlike revert statement of the problem of answering you ", causes user to fall into puzzlement, can not determine that conversational system is unable to complete user
The reason of behavior, causes dialogue that can not continue, and eventually leads to dialogue failure.
For two different abnormal behaviours, conversational system can be handled by two different modes, below will
It is described in detail respectively.
It is directed to the first abnormal behaviour, conversational system can directly prompt the inexecutable reason of user behavior.Such as Fig. 4
It is shown, it is user's sentence of user's input on the right of Fig. 4, the left side of Fig. 4 is the revert statement of conversational system.Firstly, user is defeated
Access customer sentence " stipulates the Pekinese Tian Qu plane ticket.", then, conversational system determines that user behavior " ordering air ticket " is conversational system
The behavior of support finds 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket according to the user behavior, and exports reply language
Sentence " good, to find 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket for you, you need to subscribe ".User continues input and uses
Family sentence " helps me to change the plane ticket of a Zhang Shidian ".Conversational system determines that user behavior " ordering air ticket " is the row that conversational system is supported
For still, conversational system discovery can not match slot position information departure time=" 10 points of tomorrow " with search result, and conversational system is defeated
The revert statement of " sorry, which only has this class from Shenzhen to Pekinese's flight " out.The reply that user exports from conversational system
Sentence clearly knows the reason of can not executing " ordering air ticket " user behavior, and then, user continues input user's sentence, and " well, just
This ".Conversational system determines that the user behavior of " ordering air ticket " is the behavior for belonging to conversational system support, executes " ordering air ticket "
User behavior, and export " well, has been your airline reservation ".
Fig. 2 and Fig. 4 are compared it can be found that in the corresponding method of Fig. 2, conversational system " helps me can not execute
Change the plane ticket of a Zhang Shidian " corresponding user behavior when, the revert statement of " sorry, I seem do not understand " can be exported, led
Family of applying falls into puzzlement, can not determine the reason of conversational system is unable to complete user behavior, causes dialogue that can not continue, but
It is that in the corresponding method of Fig. 4, conversational system can not execute " me is helped to change the plane ticket of a Zhang Shidian " corresponding user behavior
When, the revert statement of " sorry, which only has this class from Shenzhen to Pekinese's flight " can be exported, user is clear from dialogue
The reason of conversational system is unable to complete user behavior is known in the revert statement of system output, the direction of adjustment dialogue in time, dialogue
It goes on smoothly.
In a specific embodiment, " sorry, the day is from Shenzhen to Beijing for the first revert statement of conversational system output
Flight there was only this class " can be and generate in the following manner: conversational system obtain corresponding first template " it is sorry, should
Flight day from { departure place } to { destination } only has this class ", then, conversational system " stipulates day to go from user's sentence
It is " Beijing " that slot position information destination is got in Pekinese's plane ticket ", and slot position letter is got from global positioning system
Ceasing departure place is " Shenzhen " (departure place is defaulted as current located sites).Later, conversational system is by departure place and purpose
{ departure place } and { destination } in the first template is inserted in place, to obtain that " sorry, the day is from Shenzhen to Pekinese
Flight only has this class ".
In a specific embodiment, corresponding first template of conversational system acquisition, which can be, to be accomplished by the following way
: conversational system obtains user behavior " ordering air ticket ", enquiry machine ticket data library, determines that the reason of user behavior can not execute is not
Have and find the Flight Information for meeting user demand, to obtain the system action of " notice no ticket ".Conversational system is according to " notice nothing
The classification of ticket " system action finds the first template in corresponding from multiple modules, wherein multiple system actions and multiple moulds
Corresponding relationship between plate can be pre-set.
Conversational system, can be with prompted dialog other than it can directly prompt the inexecutable reason of active user's behavior
The other users behavior that system is supported, for example, conversational system can prompt the user bigger with active user's behavior correlation
Behavior.As shown in figure 5, being user's sentence of user's input on the right of Fig. 5, the left side of Fig. 5 is the revert statement of conversational system.It is first
First, user inputs user's sentence and " stipulates the Pekinese Tian Qu plane ticket.", then, conversational system determines user behavior " ordering air ticket "
It is the behavior that conversational system is supported, 8 points of tomorrow morning is found from Shenzhen to Pekinese's plane ticket according to the user behavior, and
Export revert statement " good, to find 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket for you, you need to subscribe ".User
Continue to input user's sentence " me is helped to change the plane ticket of a Zhang Shidian ".Conversational system determines that user behavior " ordering air ticket " is dialogue system
The behavior that system is supported, still, conversational system discovery can not match slot position information departure time=" 10 points of tomorrow " with search result,
Conversational system exports except first revert statement of " sorry, which only has this class from Shenzhen to Pekinese's flight ", also exports
The third revert statement of " you can recommend ticket information after renewal air ticket or inquiry ".User exported from conversational system
One revert statement clearly knows the reason of can not executing " ordering air ticket " user behavior, is also known clearly by third revert statement
The other users behavior that conversational system is supported.Then, user continues to input user's sentence " well, with regard to this ".Dialogue system
It unites and determines that the user behavior of " ordering air ticket " is the behavior for belonging to conversational system support, execute the user behavior of " ordering air ticket ", and defeated
" well, has been your airline reservation " out.
Here, therefore the user behavior that conversational system clearly indicates that conversational system is supported can guide use well
The direction of family dialogue, so that dialogue can be carried out more clearly.
In a specific embodiment, conversational system output third revert statement " you can after renewal air ticket or
Ticket information is recommended in inquiry " it can be and generate in the following manner: conversational system obtains corresponding second template, and " you can be after
Continuous { recommended user's behavior 1 } or { recommended user's behavior 2 } ", then, conversational system " helps me to change one ten according to user's sentence
The corresponding user behavior " ordering air ticket " of the plane ticket of point ", finds and user behavior " ordering air ticket " phase in user behavior list
The biggish user behavior of closing property " ticket information is provided " and " ticket information is recommended in inquiry ".Later, conversational system " will provide machine
{ recommended user's behavior 1 } and { recommended user's behavior in ticket information " and " ticket information is recommended in inquiry " second template of filling
2 }, to obtain third revert statement " you can recommend ticket information after renewal air ticket or inquiry ".
In a specific embodiment, user behavior list can be set in advance in conversational system.Wherein, Yong Huhang
It include multiple user behaviors that conversational system can be supported for list.After in user behavior list, setting completed, conversational system
User behavior list can be kept constant, user behavior list can also be deleted during use, add and
Update the operation such as user behavior.In a specific embodiment, user behavior list can be as shown in table 1:
1 user behavior list of table
Serial number | User behavior |
1 | Order air ticket |
2 | Ticket information is provided |
3 | Ticket information is recommended in inquiry |
4 | Order hotel |
… | … |
It should be understood that above-mentioned user behavior list is only a kind of example, in other examples, user behavior list is also
It may include more or less user behavior, also, the user behavior in user behavior list can also be with the user in table 1
Behavior part is identical or entirely different, should not constitute specific restriction herein.
In a specific embodiment, corresponding second template of conversational system acquisition, which can be, to be accomplished by the following way
: conversational system obtains user behavior " ordering air ticket ", enquiry machine ticket data library, does not find the flight letter for meeting user demand
Breath, to obtain the system action of " prompt user behavior ".Conversational system according to the classification of " prompt user behavior " system action,
The second template in corresponding is found from multiple template, wherein the corresponding relationship between multiple system actions and multiple template can
To be pre-set.
In a specific embodiment, conversational system is corresponding according to user's sentence " me is helped to change the plane ticket of a Zhang Shidian "
User behavior " ordering air ticket ", the user behavior higher with " ordering air ticket " correlation is found in user behavior list and " is mentioned
For ticket information " and " inquiry recommend ticket information " can be achieved in that and user behavior is classified, and will be same
The degree of correlation of the user behavior of class is arranged relatively high, and the degree of correlation of inhomogeneous user behavior is arranged relatively low.Then, root
The relatively high user behavior of the degree of correlation is selected from user behavior list according to target user's behavior to recommend.Shown in table 1
User behavior list for, " ordering air ticket ", " provide ticket information " and " ticket information is recommended in inquiry " are to belong to " machine
The user behavior of ticket " classification, and, the user behavior of " ordering hotel " is not belonging to the user behavior of " air ticket " classification, therefore, " orders machine
The degree of correlation of ticket ", " providing ticket information " and " ticket information is recommended in inquiry " can be set relatively high, the phase of " ordering hotel "
Guan Du can be set relatively low.Conversational system can be " ordering air ticket " according to target user's behavior, from user behavior list
The degree of correlation relatively high " ticket information is provided " and " ticket information is recommended in inquiry " is selected to be recommended.
Above-mentioned Fig. 4 and relevant example are only a kind of example, should not constitute specific restriction, in other embodiment party
In formula, it can also be the dialogue under other scenes, be not especially limited herein.
It is directed to second of abnormal behaviour, conversational system can identify that user is intended to, and prompt that user is not supported to be intended to correspond to
Function.As shown in fig. 6, being user's sentence of user's input on the right of Fig. 6, the left side of Fig. 6 is the revert statement of conversational system.
Firstly, user inputs user's sentence " stipulating the Pekinese Tian Qu plane ticket ", then, conversational system determines user behavior " ordering air ticket "
It is the user behavior that conversational system is supported, 8 points of tomorrow morning is found from Shenzhen to Pekinese's plane ticket according to user behavior,
And export revert statement " good, to find 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket for you, you need to subscribe ".With
Continue to input user's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " in family.Conversational system determination is not belonging to conversational system branch
The user behavior held, then, conversational system identify that user is intended to " ordering high guaranteed votes " according to user's sentence, and export " it is sorry, temporarily
It does not support to order high-speed rail ticket function ".The revert statement that user exports from conversational system clearly knows that " ordering high guaranteed votes " is conversational system
The function of not supporting, then, user input user's sentence " that stipulates the plane ticket in 8 points of day ".Conversational system determines user's row
It is the user behavior for belonging to conversational system support for " ordering air ticket ", executes the user behavior of " ordering air ticket ", and export " it is good,
For your airline reservation ".
Fig. 3 and Fig. 6 are compared it can be found that conversational system is encountering " ordering high guaranteed votes " in the corresponding method of Fig. 3
The revert statement that when user behavior that this conversational system can not be supported, can export " sorry, I seems not understand ", causes to use
Family falls into puzzlement, can not determine the reason of conversational system is unable to complete user behavior, causes dialogue that can not continue, still,
In the corresponding method of Fig. 6, conversational system is right when encountering the user behavior that " ordering high guaranteed votes " this conversational system can not be supported
Telephone system can identify that user is intended to according to user's sentence, and export second reply of " sorry, wouldn't to support to order high-speed rail ticket function "
Sentence, user, which is clear, knows that conversational system is not supported to order high-speed rail ticket function from the revert statement that conversational system exports, and
When adjustment dialogue direction, dialogue goes on smoothly.
In a specific embodiment, conversational system output the second revert statement " it is sorry, wouldn't support to order high guaranteed votes
Function " can be to be generated in the following manner: conversational system obtain corresponding third template " it is sorry, wouldn't support { user's meaning
Figure } function ", then, conversational system is intended to according to user's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " identification user
For " ordering high guaranteed votes ".Later, user is intended to " ordering high guaranteed votes " filling third template by conversational system, thus obtain " it is sorry, it wouldn't
High-speed rail ticket function is ordered in support ".
In a specific embodiment, the corresponding third template of conversational system acquisition, which can be, to be accomplished by the following way
: conversational system discovery can not be according to user's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " by corresponding user behavior
It is categorized into any one user behavior of conversational system support, it is thus determined that conversational system does not support the user behavior, to obtain
The system action of " notice does not support function ".Conversational system is according to the classification of " notice do not support function " system action, from multiple
Template finds the third template in corresponding, wherein the corresponding relationship between multiple system actions and multiple template can be pre-
First it is arranged.
In a specific embodiment, conversational system obtain according to user's sentence identify user be intended to can be by with
What under type was realized: user's input by sentence intention assessment model can be identified that user is intended to by conversational system.It is specific one
Embodiment in, it is intended that identification model can be classifier, for example, k nearest neighbor classifier, support vector machines, decision tree and with
Machine forest etc., is not especially limited herein.User can be intended to be divided by Activity recognition model supports scene with conversational system
The related and common classification in life, for example, order high guaranteed votes, photos and sending messages, open application, search cuisines, go by taxi etc..Meaning
User's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " that figure identification model is inputted according to user, can be it is determined that return
Enter " ordering high guaranteed votes " classification, to identify that the user that user's sentence includes is intended that " ordering high guaranteed votes ".
In a specific embodiment, it is intended that identification model can be conversational system by a large amount of known users sentences with
And a large amount of known users are intended to be trained.Training below by taking deep neural network as an example, to intention assessment model
Process is described in detail.
Essence to the training of intention assessment model is the weight square determined in deep neural network by trained process
Battle array.Specifically: each layer of work in deep neural network can use mathematic(al) representationTo describe: from object
Each layer of work in reason level deep neural network can be understood as by five kinds to the input space (collection of input vector
Close) operation, complete the transformation (i.e. Row rank to column space) of the input space to output space, this five kinds operations are wrapped
It includes: 1, rising dimension/dimensionality reduction;2, amplification/diminution;3, it rotates;4, it translates;5, " bending ".Wherein 1,2,3 operation byIt completes, 4
Operation by+b complete, 5 operation is then by a () Lai Shixian.Here why stated with " space " two word is because being classified
Object be not single things, but a kind of things, space refers to all groups of individuals of this kind of things.Wherein, W is weight
Vector, each of vector value indicate the weighted value of a neuron in this layer of neural network.Vector W decides
The input space described in text to output space spatial alternation, i.e., each layer of weight W control how transformation space.Training is deep
The purpose of neural network is spent, that is, finally obtains all layers of weight matrix of trained neural network (by plurality of layers
Vector it is W-shaped at weight matrix).Therefore, the training process of neural network is substantially exactly the mode of study control spatial alternation,
It is exactly more specifically to learn weight matrix.
Process to the training of intention assessment model is exactly constantly to reduce the pre- of deep neural network by updating weight
The process of difference between measured value and target value.Since it is desired that the output of deep neural network is as close as really desired prediction
Value, it is possible to by comparing the predicted value and really desired target value of current network, further according to difference between the two
Situation come update each layer of neural network weight vectors (process of initialization certainly, is usually had before first time updates,
Each layer as in deep neural network is pre-configured with parameter), for example, if the predicted value of network is high, just adjust weight to
It is lower that amount allows it to predict, continuous to adjust, until deep neural network can predict really desired target value.Therefore,
With regard to needing pre-defined " how the difference between comparison prediction value and target value ", this is loss function (loss
Function) or objective function (objective function), they are the difference for measuring predicted value and target value
Important equation.Wherein, it is illustrated with loss function, the higher expression difference of output valve (loss) of loss function is bigger, then depth
The training of neural network has reformed into the process for reducing this loss as far as possible.Specifically, by a known users input by sentence
Deep neural network can be obtained by a predicted value by forward-propagating.Then, by the way that the known users sentence is corresponding
Known users intention is compared, to obtain difference value between the two, carries out backpropagation according to the difference value, so that it may
Update the weight vectors W of each layer of neural network.Constantly using a large amount of known users sentences and its is corresponding a large amount of known
User is intended to adjustment weight vectors W, and through deep neural network can predict really desired target value.
In a specific embodiment, a large amount of known users sentences that training deep neural network uses be can be manually
Input dialogue system, it can be conversational system and voluntarily obtain what the dialog history of user and conversational system obtained, be also possible to
Pass through what is obtained from by way of production method (self-play).It wherein, from production method is able to achieve by engineer one
The conversational system of Task dialog logic and one can cover user's simulator of status information as much as possible.Allow user
Simulator and dialog engine carry out automatic interaction, generate corresponding dialogue data (comprising user's sentence).Training depth nerve net
A large amount of known users that network uses are intended to can be by carrying out what artificial mark obtained to a large amount of known users sentences.
Conversational system, can be with prompted dialog system other than it can directly prompt that user is not supported to be intended to corresponding function
The other users behavior that system is supported, for example, conversational system can prompt the user row bigger with active user's behavior correlation
For.As shown in fig. 7, being user's sentence of user's input on the right of Fig. 7, the left side of Fig. 7 is the revert statement of conversational system.It is first
First, user inputs user's sentence " stipulating the Pekinese Tian Qu plane ticket ", and then, conversational system determines that user behavior " ordering air ticket " is
The user behavior that conversational system is supported, finds 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket according to user behavior, and
Export revert statement " good, to find 8 points of tomorrow morning from Shenzhen to Pekinese's plane ticket for you, you need to subscribe ".User
Continue to input user's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back ".Conversational system, which determines, is not belonging to conversational system support
User behavior, then, conversational system according to user's sentence identify user be intended to " ordering high guaranteed votes ", export " sorry, Zan Buzhi
Hold and order high-speed rail ticket function " revert statement, also, also output " you can after renewal air ticket or inquiry recommend ticket information "
Revert statement.The revert statement that user exports from conversational system clearly knows that " ordering high guaranteed votes " is the function that conversational system is not supported
Can, also know clearly the other users behavior that conversational system is supported.Then, user continue input user's sentence " that stipulates day
8 points of plane ticket ".Conversational system determines that user behavior " ordering air ticket " is the user behavior for belonging to conversational system support, executes
The user behavior of " ordering air ticket ", and export " good, to be your airline reservation ".
In a specific embodiment, conversational system output the 4th revert statement " you can after renewal air ticket or
Ticket information is recommended in inquiry " it can be and generate in the following manner: conversational system obtains corresponding second template, and " you can be after
Continuous { recommended user's behavior 1 } or { recommended user's behavior 2 } ", then, conversational system " is stipulated according to upper one user's sentence
The corresponding user behavior " ordering air ticket " of the Pekinese Tian Qu plane ticket " finds in user behavior list and " orders machine with user behavior
Ticket " the biggish user behavior of correlation " ticket information is provided " and " ticket information is recommended in inquiry ".Later, conversational system will
{ the recommended user's behavior 1 } and { recommend to use that " providing ticket information " and " ticket information is recommended in inquiry " is inserted in second template
Family behavior 2 }, to obtain the 4th revert statement " you can recommend ticket information after renewal air ticket or inquiry ".
Above-mentioned Fig. 6, Fig. 7 and relevant example are only a kind of example, should not constitute specific restriction, other real
It applies in mode, can also be the dialogue under other scenes, be not especially limited herein.
A kind of abnormal behaviour answering method provided by the present application, can clearly prompt user session system that can not execute exception
The reason of behavior, to improve the robustness of processing of the conversational system to abnormal behaviour.The abnormal row of the embodiment of the present application
For answering method, comprising the following steps:
S101: first user's sentence of user's input is obtained.
It include the first user behavior in first user's sentence of user's input in a specific embodiment.Wherein, institute
Stating user behavior is the behavior that user proposes first requirement to conversational system.It is with first user's sentence " Pekinese Ding Qu air ticket "
Example, the first user sentence propose the first requirement of " ordering air ticket " to conversational system, therefore, include in first user's sentence
First user behavior " ordering air ticket ".It should be understood that the citing of above-mentioned first user behavior is merely possible to a kind of example, it is other
In embodiment, the first user behavior can also be " making a phone call ", " querying geographical position ", " ordering take-away ", " inquiry weather " and
" ordering hotel " etc., is not especially limited herein.
In a specific embodiment, the first user behavior can be conversational system for first user's input by sentence behavior
What identification model identified.Wherein, Activity recognition model can be classifier, for example, k nearest neighbor classifier, support vector machines,
Decision tree and random forest etc. are not especially limited herein.The use that Activity recognition model can be supported according to conversational system
Family behavior divides classification.For example, the user behavior that conversational system is supported includes " ordering air ticket ", " making a phone call ", " orders outer
Sell ", " inquiry weather " and " ordering hotel ", then, the classification of behavior identification model just include " ordering air ticket ", " making a phone call ",
" ordering take-away ", " inquiry weather " and " ordering hotel ".User's sentence " Pekinese Ding Qu that Activity recognition model is inputted according to user
Air ticket ", can be it is determined that be included into " ordering air ticket " classification, so that the user behavior for identifying that user's sentence includes is " to order machine
Ticket ".
In a specific embodiment, first user's sentence can also include slot position information.Wherein, slot position information is to use
In the value for filling slot relevant to the first user behavior.It is relevant to " ordering air ticket " by taking the first user behavior " ordering air ticket " as an example
Slot position includes departure time, departure place and destination.Corresponding to above-mentioned slot position relevant to " ordering air ticket ", used first
In family sentence " stipulating 10 points of day from Shenzhen to Pekinese's air ticket ", slot position information includes departure time=" 10 points of tomorrow ", sets out
Place=" Shenzhen " and destination=" Beijing " etc..Conversational system needs all fill slot position relevant to user behavior
It is full, user behavior could be executed.For example illustrate, conversational system needs that " air ticket will be ordered " the relevant slot position departure time, goes out
Hair place and destination are all filled out and are filled with, and " ordering air ticket " this user behavior could be executed.
In a specific embodiment, slot position information can be conversational system for user's input by sentence slot position identification model
It obtains.Wherein, slot position identification model can be conventional machines learning model, such as conditional random field models, cyclic convolution net
Network or dictionary search method etc., are not especially limited herein.Slot position identification model can be extracted to be provided in user session
Key message.For example, ordering air ticket type of slots includes " departure place " and " destination ", slot position identification model, which needs to extract, " sets out
The information on ground " and " destination "." I will order Beijing to the machine in Shanghai to user's sentence that slot position identification model is inputted according to user
Ticket ", identification obtain the result is that " departure place: Beijing " and " destination: Shanghai ", to give conversational system offer slot position information.
S102: determine that there are abnormal behaviours according to the first user sentence, wherein the abnormal behaviour is the dialogue
The behavior that system can not be handled normally.
In a specific embodiment, abnormal behaviour may include two kinds: the first, abnormal behaviour is to belong to dialogue system
What system was supported, still, it is not possible to the user behavior being performed.Second, abnormal behaviour is the row for being not belonging to conversational system support
For.
In a specific embodiment, conversational system can be judged by the following manner whether the first user behavior is to belong to
In the behavior that conversational system is supported: continuing by taking classifier as an example, it is assumed that user behavior is divided into " ordering air ticket ", " beaten by classifier
The multiple classification of phone ", " ordering take-away ", " inquiry weather " and " ordering hotel " etc..So, if user's sentence of user's input
For " ordering high guaranteed votes back ", then classifier can not be classified to any of the above classification, at this point, conversational system may determine that
User behavior is not the behavior for belonging to conversational system support.Conversely, conversational system may determine that the first user behavior is to belong to pair
The behavior that telephone system is supported.
In a specific embodiment, conversational system can be judged by the following manner whether user behavior can be held
Row.By taking user's sentence " stipulating 10 points of day from Shenzhen to Pekinese's air ticket " as an example, user behavior is " ordering air ticket ", slot position packet
Include departure time=" 10 points of tomorrow ", departure place=" Shenzhen " and destination=" Beijing ".But conversational system carries out
It is found after search, only searches 8 points of tomorrow from Shenzhen to Pekinese's flight, that is, the slot position information of user's sentence can not and be retrieved
As a result it is matched, then, conversational system judges that user behavior can not be performed, conversely, then conversational system judges user's row
For that can be performed.
S103: being that the conversational system can identify in the abnormal behaviour, still, the case where the behavior that can not be executed
Under, identify the reason of abnormal behaviour can not be performed, and generate first replied the abnormal behaviour and reply language
Sentence.
In a specific embodiment, first revert statement is for prompting first user behavior that cannot be held
Capable reason.The first revert statement " sorry, which only has this class from Shenzhen to Pekinese's flight " exported with conversational system
For, the first revert statement can be to be generated in the following manner: conversational system obtain corresponding first template " it is sorry, should
Flight day from { departure place } to { destination } only has this class ", then, conversational system " stipulates day to go from user's sentence
It is " Beijing " that slot position information destination is got in Pekinese's plane ticket ", and slot position letter is got from global positioning system
Ceasing departure place is " Shenzhen " (departure place is defaulted as current located sites).Later, conversational system is by departure place and purpose
{ departure place } and { destination } in the first template is inserted in place, to obtain that " sorry, the day is from Shenzhen to Pekinese
Flight only has this class ".
In a specific embodiment, first template contains associated first slot of first user behavior
Position.By taking the first template the flight of the day from { departure place } to { destination } " sorry, there was only this class " as an example, the first slot position
Including { departure place } and { destination }.
In a specific embodiment, corresponding first template of conversational system acquisition, which can be, to be accomplished by the following way
: the reason of conversational system can not be performed according to the abnormal behaviour searches corresponding first template, wherein first mould
Plate contains associated first slot position of first user behavior;It obtains and believes for filling with the first slot position of first slot position
Breath;The first slot position information is inserted into first slot position, obtains first revert statement.For example illustrate, talk with
System obtains user behavior " ordering air ticket ", enquiry machine ticket data library, determines that the reason of user behavior can not execute is not find
Meet the Flight Information of user demand, to obtain the system action of " notice no ticket ".Conversational system is according to " notice no ticket " system
The classification of system behavior finds the first template in corresponding from multiple modules, wherein between multiple system actions and multiple template
Corresponding relationship can be it is pre-set.
In a specific embodiment, conversational system can also export third revert statement, wherein the third is replied
The behavior that sentence is used to that the conversational system to be prompted to be able to carry out.Conversational system searches the second template, wherein second template
Including the second slot position;Obtain the first association user behavior of first user behavior, wherein the first association user behavior
It is the user behavior for being greater than threshold value with the degree of correlation of first user behavior;It will be described in the first association user behavior filling
Second slot position, to obtain third revert statement.It is that " you can be after renewal air tickets with the third revert statement that conversational system exports
Or ticket information is recommended in inquiry " for, third revert statement can be to be generated in the following manner: conversational system obtains pair
The second template " you can continue { recommended user's behavior 1 } or { recommended user's behavior 2 } " answered, then, conversational system according to
The corresponding user behavior " ordering air ticket " of second user sentence " me is helped to change the plane ticket of a Zhang Shidian " before first user's sentence,
Found in user behavior list with the biggish user behavior of user behavior " ordering air ticket " correlation " provide ticket information " with
And " ticket information is recommended in inquiry ".Later, " provide ticket information " and " ticket information is recommended in inquiry " is inserted the by conversational system
{ recommended user's behavior 1 } and { recommended user's behavior 2 } in two templates, to obtain third revert statement, " you can continue
It orders air ticket or ticket information is recommended in inquiry ".
In a specific embodiment, user behavior list can be set in advance in conversational system.Wherein, Yong Huhang
It include multiple user behaviors that conversational system can be supported for list.After in user behavior list, setting completed, conversational system
User behavior list can be kept constant, user behavior list can also be deleted during use, add and
Update the operation such as user behavior.
In a specific embodiment, the second template includes the second slot position.With the second template, " you can continue { to recommend
User behavior 1 } or { recommended user's behavior 2 } " for, the second slot position includes: { recommended user's behavior 1 } and { recommended user
Behavior 2 } ".
In a specific embodiment, corresponding second template of conversational system acquisition, which can be, to be accomplished by the following way
: conversational system obtains user behavior " ordering air ticket ", enquiry machine ticket data library, does not find the flight letter for meeting user demand
Breath, to obtain the system action of " prompt user behavior ".Conversational system according to the classification of " prompt user behavior " system action,
The second template in corresponding is found from multiple template, wherein the corresponding relationship between multiple system actions and multiple template can
To be pre-set.
In a specific embodiment, conversational system is corresponding according to user's sentence " me is helped to change the plane ticket of a Zhang Shidian "
User behavior " ordering air ticket ", the user behavior higher with " ordering air ticket " correlation is found in user behavior list and " is mentioned
For ticket information " and " inquiry recommend ticket information " can be achieved in that and user behavior is classified, and will be same
The degree of correlation of the user behavior of class is arranged relatively high, and the degree of correlation of inhomogeneous user behavior is arranged relatively low.Then, root
The relatively high user behavior of the degree of correlation is selected from user behavior list according to target user's behavior to recommend.Shown in table 1
User behavior list for, " ordering air ticket ", " provide ticket information " and " ticket information is recommended in inquiry " are to belong to " machine
The user behavior of ticket " classification, and, the user behavior of " ordering hotel " is not belonging to the user behavior of " air ticket " classification, therefore, " orders machine
The degree of correlation of ticket ", " providing ticket information " and " ticket information is recommended in inquiry " can be set relatively high, the phase of " ordering hotel "
Guan Du can be set relatively low.Conversational system can be " ordering air ticket " according to target user's behavior, from user behavior list
The degree of correlation relatively high " ticket information is provided " and " ticket information is recommended in inquiry " is selected to be recommended.
S104: in the case where the abnormal behaviour is the unrecognized behavior of the conversational system, according to described first
User's sentence identifies that user is intended to, and generates the second revert statement replied the abnormal behaviour.
In a specific embodiment, second revert statement is for prompting the conversational system not support the use
Family is intended to.It is the second reply for " sorry, wouldn't to support to order high-speed rail ticket function " with the second revert statement that conversational system exports
Sentence can be to be generated in the following manner: third template is searched, the user is intended to insert the third slot position, thus
Obtain the second revert statement.For being " sorry, wouldn't to support { user's intention } function " with third template, conversational system according to
Family sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " identification user is intended to " ordering high guaranteed votes ".Later, conversational system will
User is intended to " ordering high guaranteed votes " and inserts third template, to obtain " sorry, wouldn't to support to order high-speed rail ticket function ".
In a specific embodiment, the corresponding third template of conversational system acquisition, which can be, to be accomplished by the following way
: conversational system discovery can not be according to user's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " by corresponding user behavior
It is categorized into any one user behavior of conversational system support, it is thus determined that conversational system does not support the user behavior, to obtain
The system action of " notice does not support function ".Conversational system is according to the classification of " notice do not support function " system action, from multiple
Template finds the third template in corresponding, wherein the corresponding relationship between multiple system actions and multiple template can be pre-
First it is arranged.
In a specific embodiment, conversational system obtain according to user's sentence identify user be intended to can be by with
What under type was realized: user's input by sentence intention assessment model can be identified that user is intended to by conversational system.Wherein, it is intended that
Identification model can be classifier, for example, k nearest neighbor classifier, support vector machines, decision tree and random forest etc., herein
It is not especially limited.User can be intended to be divided into related to conversational system support scene and normal in life by intention assessment model
Classification, for example, order high guaranteed votes, photos and sending messages, open application, search cuisines, go by taxi etc..Intention assessment model according to
User's sentence " first looking at the day after tomorrow either with or without high guaranteed votes back " of family input, can be it is determined that be included into " ordering high guaranteed votes " point
Class, to identify that the user that user's sentence includes is intended that " ordering high guaranteed votes ".
In a specific embodiment, it is intended that identification model can be conversational system by a large amount of known users sentences with
And a large amount of known users are intended to be trained.Training below by taking deep neural network as an example, to intention assessment model
Process is described in detail.
Essence to the training of intention assessment model is the weight square determined in deep neural network by trained process
Battle array.Specifically: each layer of work in deep neural network can use mathematic(al) representationTo describe: from object
Each layer of work in reason level deep neural network can be understood as by five kinds to the input space (collection of input vector
Close) operation, complete the transformation (i.e. Row rank to column space) of the input space to output space, this five kinds operations are wrapped
It includes: 1, rising dimension/dimensionality reduction;2, amplification/diminution;3, it rotates;4, it translates;5, " bending ".Wherein 1,2,3 operation byIt completes, 4
Operation by+b complete, 5 operation is then by a () Lai Shixian.Here why stated with " space " two word is because being classified
Object be not single things, but a kind of things, space refers to all groups of individuals of this kind of things.Wherein, W is weight
Vector, each of vector value indicate the weighted value of a neuron in this layer of neural network.Vector W decides
The input space described in text to output space spatial alternation, i.e., each layer of weight W control how transformation space.Training is deep
The purpose of neural network is spent, that is, finally obtains all layers of weight matrix of trained neural network (by plurality of layers
Vector it is W-shaped at weight matrix).Therefore, the training process of neural network is substantially exactly the mode of study control spatial alternation,
It is exactly more specifically to learn weight matrix.
Process to the training of intention assessment model is exactly constantly to reduce the pre- of deep neural network by updating weight
The process of difference between measured value and target value.Since it is desired that the output of deep neural network is as close as really desired prediction
Value, it is possible to by comparing the predicted value and really desired target value of current network, further according to difference between the two
Situation come update each layer of neural network weight vectors (process of initialization certainly, is usually had before first time updates,
Each layer as in deep neural network is pre-configured with parameter), for example, if the predicted value of network is high, just adjust weight to
It is lower that amount allows it to predict, continuous to adjust, until deep neural network can predict really desired target value.Therefore,
With regard to needing pre-defined " how the difference between comparison prediction value and target value ", this is loss function (loss
Function) or objective function (objective function), they are the difference for measuring predicted value and target value
Important equation.Wherein, it is illustrated with loss function, the higher expression difference of output valve (loss) of loss function is bigger, then depth
The training of neural network has reformed into the process for reducing this loss as far as possible.Specifically, by a known users input by sentence
Deep neural network can be obtained by a predicted value by forward-propagating.Then, by the way that the known users sentence is corresponding
Known users intention is compared, to obtain difference value between the two, carries out backpropagation according to the difference value, so that it may
Update the weight vectors W of each layer of neural network.Constantly using a large amount of known users sentences and its is corresponding a large amount of known
User is intended to adjustment weight vectors W, and through deep neural network can predict really desired target value.
In a specific embodiment, a large amount of known users sentences that training deep neural network uses be can be manually
Input dialogue system, it can be conversational system and voluntarily obtain what the dialog history of user and conversational system obtained, be also possible to
Pass through what is obtained from by way of production method (self-play).It wherein, from production method is able to achieve by engineer one
The conversational system of Task dialog logic and one can cover user's simulator of status information as much as possible.Allow user
Simulator and dialog engine carry out automatic interaction, generate corresponding dialogue data (comprising user's sentence).Training depth nerve net
A large amount of known users that network uses are intended to can be by carrying out what artificial mark obtained to a large amount of known users sentences.
It is a kind of structural schematic diagram of conversational system provided by the present application referring to Fig. 8, Fig. 8.It is provided by the embodiments of the present application
Conversational system includes: input module 110, determining module 120 and reply module 130.
The input module 110 is used to obtain first user's sentence of user's input, wherein the first user sentence refers to
The first user behavior is shown;
The determining module 120 is used to determine that described first uses by Activity recognition model according to the first user sentence
Family behavior is abnormal behaviour, wherein the abnormal behaviour is the behavior that the conversational system can not be handled normally;
The reply module 130 is used in the abnormal behaviour be that the conversational system can identify, but can not execute
Behavior in the case where, identify the reason of abnormal behaviour can not be performed, and generate and reply the abnormal behaviour
The first revert statement, wherein first revert statement for prompt first user behavior that cannot be performed the reason of;
Or
The reply module 130 is also used to the case where the abnormal behaviour is the conversational system unrecognized behavior
Under, identify that user is intended to according to the first user sentence, and generate second replied the abnormal behaviour and reply language
Sentence, wherein second revert statement is for prompting the conversational system that the user is not supported to be intended to.
It states for simplicity, there is no be how to generate to the first revert statement, the second revert statement etc. to the present embodiment
Carry out expansion description, specifically referring to Figure 1-5 and relevant description.The present embodiment also not to third revert statement and
4th revert statement is being introduced of how generating, and specifically refers to Fig. 6, Fig. 7 and associated description.
It is a kind of structural schematic diagram of computing cluster provided by the present application referring to Fig. 9, Fig. 9.Conversational system shown in Fig. 8 can
To be realized in computing cluster as shown in Figure 9.Computing cluster include including at least one calculate node 210 and at least one
Memory node 220.
Calculate node 210 includes one or more processors 211, communication interface 212 and memory 213.Wherein, processor
211, it can be connected by bus 224 between communication interface 212 and memory 213.
Processor 211 includes one or more general processor, wherein general processor, which can be, is capable of handling electronics
Any kind of equipment of instruction, including it is central processing unit (Central Processing Unit, CPU), microprocessor, micro-
Controller, primary processor, controller and ASIC (Application Specific Integrated Circuit, dedicated collection
At circuit) etc..It can be only for the application specific processor of calculate node 210 or can be total to other calculate nodes 210
It enjoys.Processor 211 executes various types of stored digital instructions, such as the software or firmware journey that are stored in memory 213
Sequence, it can make calculate node 210 provide wider a variety of services.For example, processor 211 is able to carry out program or processing number
According to execute at least part for the method being discussed herein.Can be run in processor 211 input module as shown in Figure 8, really
Cover half block and reply module.
Communication interface 212 can be wireline interface (such as Ethernet interface), for other calculate nodes or user into
Row communication.
Memory 213 may include volatile memory (Volatile Memory), such as random access memory
(Random Access Memory, RAM);Memory also may include nonvolatile memory (Non-Volatile
), such as read-only memory (Read-Only Memory, ROM), flash memory (Flash Memory), hard disk Memory
(Hard Disk Drive, HDD) or solid state hard disk (Solid-State Drive, SSD) memory can also include above-mentioned kind
The combination of the memory of class.
Memory node 220 includes one or more processors 221, communication interface 222 and memory 223.Wherein, processor
221, it can be connected by bus 224 between communication interface 222 and memory 223.
Processor 221 includes one or more general processor, wherein general processor, which can be, is capable of handling electronics
Any kind of equipment, including CPU, microprocessor, microcontroller, primary processor, controller and ASIC etc. of instruction.It
It can be only for the application specific processor of memory node 220 or can be shared with other memory nodes 220.Processor 221 is held
The various types of stored digital instructions of row, such as the software or firmware program that are stored in memory 223, it can be such that storage saves
Point 220 provides wider a variety of services.For example, processor 221 is able to carry out program or processing data, it is discussed herein with executing
Method at least part.
Communication interface 222 can be wireline interface (such as Ethernet interface), for other calculate equipment or user into
Row communication.
Memory node 220 includes one or more storage controls 221 and storage array 225.Wherein, storage control
It can be connected by bus 226 between 221 and storage array 225.
Storage control 221 includes one or more general processor, wherein general processor, which can be, to be capable of handling
Any kind of equipment, including CPU, microprocessor, microcontroller, primary processor, controller and ASIC of e-command etc.
Deng.It can be only for the application specific processor of single memory node 220 or can save with calculate node or other storages
Point 220 is shared.It is appreciated that in the present embodiment, each memory node includes a storage control, in other embodiments
In, a storage control can also be shared with multiple memory nodes, be not especially limited herein.
Memory array 225 may include multiple memories.Memory can be nonvolatile memory, such as ROM, fast
Flash memory, HDD or SSD memory can also include the combination of the memory of mentioned kind.For example, storage array can be by
Multiple HDD or multiple SDD composition, is made of alternatively, storage array can be HDD and SDD.Wherein, multiple memories exist
Storage control 221 by assistance under be combined in different ways and form memory group, compare single memory to provide
Higher storage performance and offer technology of data copy.Optionally, memory array 225 may include one or more data
Center.Multiple data centers, which can be set, to be in the same localities, alternatively, be not especially limited in different places respectively herein.
Memory array 225 can store program code and database.Wherein, program code includes obtaining block code, determination
Block code and reply block code.Database includes: Activity recognition model, slot position identification model and intention assessment mould
Type.
Wherein, the calculate node 211 is by calling the program code in memory node 213, for executing following steps:
Obtain first user's sentence of user's input, wherein the first user sentence indicates the first user behavior;
Determine that first user behavior is abnormal behaviour by Activity recognition model according to the first user sentence,
In, the abnormal behaviour is the behavior that the conversational system can not be handled normally;
It is that the conversational system can identify in the abnormal behaviour, still, in the case where the behavior that can not be executed, identification
The reason of abnormal behaviour can not be performed, and generate the first revert statement replied the abnormal behaviour, wherein
First revert statement is for the reason of prompting first user behavior that cannot be performed;Or
In the case where the abnormal behaviour is the unrecognized behavior of the conversational system, according to the first user language
Sentence identification user is intended to, and generates the second revert statement replied the abnormal behaviour, wherein described second replys language
Sentence is for prompting the conversational system that the user is not supported to be intended to.
It states for simplicity, there is no be how to generate to the first revert statement, the second revert statement etc. to the present embodiment
Carry out expansion description, specifically referring to Figure 1-5 and relevant description.The present embodiment also not to third revert statement and
4th revert statement is being introduced of how generating, and specifically refers to Fig. 6, Fig. 7 and associated description.
It is a kind of structural schematic diagram for calculating equipment provided by the present application referring to Figure 10, Figure 10.Conversational system shown in Fig. 8
It can be realized in calculating equipment as shown in Figure 10.In a specific embodiment, calculating equipment can be intelligent terminal, this
The calculating equipment of embodiment includes one or more processors 311, communication interface 312 and memory 313.Wherein, processor
311, it can be connected by bus 324 between communication interface 312 and memory 313.
Processor 311 includes one or more general processor, wherein general processor, which can be, is capable of handling electronics
Any kind of equipment of instruction, including it is central processing unit (Central Processing Unit, CPU), microprocessor, micro-
Controller, primary processor, controller and ASIC (Application Specific Integrated Circuit, dedicated collection
At circuit) etc..Processor 311 executes the instruction of various types of stored digitals, for example, the software that is stored in memory 313 or
Person's firmware program, it can make to calculate the wider a variety of services of equipment offer.For example, processor 311 is able to carry out program or place
Data are managed, to execute at least part for the method being discussed herein.Input mould as shown in Figure 8 can be run in processor 311
Block, determining module and reply module.
Communication interface 312 can be wireline interface (such as Ethernet interface), for other calculate nodes or user into
Row communication.
Memory 313 may include volatile memory (Volatile Memory), such as random access memory
(Random Access Memory, RAM);Memory also may include nonvolatile memory (Non-Volatile
), such as read-only memory (Read-Only Memory, ROM), flash memory (Flash Memory), hard disk Memory
(Hard Disk Drive, HDD) or solid state hard disk (Solid-State Drive, SSD) memory can also include above-mentioned kind
The combination of the memory of class.Memory 313 can store program code and program data.Wherein, program code includes obtaining
Modulus block code, determining module code and reply block code.Database includes: Activity recognition model, slot position identification model
And intention assessment model.
Wherein, the processor 311 is by calling the program code in memory 313, for executing following steps:
Obtain first user's sentence of user's input, wherein the first user sentence indicates the first user behavior;
Determine that first user behavior is abnormal behaviour by Activity recognition model according to the first user sentence,
In, the abnormal behaviour is the behavior that the conversational system can not be handled normally;
It is that the conversational system can identify in the abnormal behaviour, still, in the case where the behavior that can not be executed, identification
The reason of abnormal behaviour can not be performed, and generate the first revert statement replied the abnormal behaviour, wherein
First revert statement is for the reason of prompting first user behavior that cannot be performed;Or
In the case where the abnormal behaviour is the unrecognized behavior of the conversational system, according to the first user language
Sentence identification user is intended to, and generates the second revert statement replied the abnormal behaviour, wherein described second replys language
Sentence is for prompting the conversational system that the user is not supported to be intended to.
It states for simplicity, there is no be how to generate to the first revert statement, the second revert statement etc. to the present embodiment
Carry out expansion description, specifically referring to Figure 1-5 and relevant description.The present embodiment also not to third revert statement and
4th revert statement is being introduced of how generating, and specifically refers to Fig. 6, Fig. 7 and associated description.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When implemented in software, it can entirely or partly realize in the form of a computer program product.The computer program
Product includes one or more computer instructions.When loading on computers and executing the computer program instructions, all or
It partly generates according to process or function described in the embodiment of the present application.The computer can be general purpose computer, dedicated meter
Calculation machine, computer network or other programmable devices.The computer instruction can store in computer readable storage medium
In, or from a computer readable storage medium to the transmission of another computer readable storage medium, for example, the computer
Instruction can pass through wired (such as coaxial cable, optical fiber, number from a web-site, computer, server or data center
User's line) or wireless (such as infrared, wireless, microwave etc.) mode into another web-site, computer, server or data
The heart is transmitted.The computer readable storage medium can be any usable medium or include that computer can access
The data storage devices such as one or more usable mediums integrated server, data center.The usable medium can be magnetism
Medium, (for example, floppy disk, storage dish, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state memory disc
Solid State Disk (SSD)) etc..
Claims (17)
1. a kind of interactive method characterized by comprising
Conversational system obtains first user's sentence of user's input, wherein the first user sentence indicates first user's row
For;Determine that first user behavior is abnormal behaviour by Activity recognition model according to the first user sentence, wherein institute
Stating abnormal behaviour is the behavior that the conversational system can not be handled normally;
The abnormal behaviour be the conversational system can identify but in the case where the behavior that can not execute, described in identification
The reason of abnormal behaviour can not be performed, and generate the first revert statement replied the abnormal behaviour, wherein it is described
First revert statement is for the reason of prompting first user behavior that cannot be performed;Or
In the case where the abnormal behaviour is the unrecognized behavior of the conversational system, known according to the first user sentence
Other user is intended to, and generates the second revert statement replied the abnormal behaviour, wherein second revert statement is used
The user is not supported to be intended in the prompt conversational system.
2. being replied the method according to claim 1, wherein generating first replied the abnormal behaviour
Sentence, comprising:
The reason of can not being performed according to the abnormal behaviour, searches corresponding first template, wherein first template includes
First user behavior associated first slot position;
Obtain the first slot position information for filling first slot position;
The first slot position information is inserted into first slot position, obtains first revert statement.
3. according to the method described in claim 2, it is characterized in that, the method also includes:
Search the second template, wherein second template includes the second slot position;
Obtain the first association user behavior of first user behavior, wherein the first association user behavior be with it is described
The degree of correlation of first user behavior is greater than the user behavior of threshold value;
Second slot position is inserted into the first association user behavior, obtains third revert statement, wherein the third is replied
Sentence is for prompting the conversational system to be able to carry out the first association user behavior.
4. being replied the method according to claim 1, wherein generating second replied the abnormal behaviour
Sentence, comprising:
Search third template, wherein the third template contains third slot position;
The user is intended to insert the third slot position, obtains second revert statement.
5. according to the method described in claim 4, it is characterized in that, the method also includes:
Search the second template, wherein second template includes the second slot position;
Obtain the second user sentence of user's input, wherein the second user sentence is before the first user sentence, institute
It states second user sentence and indicates second user behavior;
Obtain the second association user behavior of the second user behavior, wherein the second association user behavior be with it is described
The degree of correlation of second user behavior is greater than the user behavior of threshold value;
Second slot position is inserted into the second association user behavior, obtains the 4th revert statement, wherein the described 4th replys
Sentence is for prompting the conversational system to be able to carry out the second association user behavior.
6. method according to any one of claims 1 to 5, which is characterized in that according to the first user sentence
Identify that user is intended to, comprising:
By the first user input by sentence user's intention assessment model, it is intended to obtain the user, wherein the user
Intention assessment model is trained by known users sentence and known users intention.
7. according to the method described in claim 6, it is characterized in that, user's intention assessment model is classifier, described point
Class device is any one in k nearest neighbor classifier, support vector machines, decision tree and random forest.
8. a kind of conversational system, which is characterized in that including input module, determining module and module is replied,
The input module is used to obtain first user's sentence of user's input, wherein the first user sentence indicates the
One user behavior;
The determining module is used to determine first user behavior by Activity recognition model according to the first user sentence
For abnormal behaviour, wherein the abnormal behaviour is the behavior that the conversational system can not be handled normally;
The reply module is used in the abnormal behaviour be the behavior that the conversational system can be identified, but can not be executed
In the case of, identify the reason of abnormal behaviour can not be performed, and generate first time replied the abnormal behaviour
Multiple sentence, wherein first revert statement is for the reason of prompting first user behavior that cannot be performed;Or
The reply module is also used in the case where the abnormal behaviour is the unrecognized behavior of the conversational system, according to
The first user sentence identification user is intended to, and generates the second revert statement replied the abnormal behaviour, wherein
Second revert statement is for prompting the conversational system that the user is not supported to be intended to.
9. system according to claim 8, which is characterized in that
The reply module is also used to search corresponding first template the reason of can not being performed according to the abnormal behaviour,
In, first template contains associated first slot position of first user behavior;It is used for by the acquisition of slot position identification model
Fill the first slot position information of first slot position;The first slot position information is inserted into first slot position, obtains described the
One revert statement.
10. system according to claim 9, which is characterized in that
The reply module is also used to search the second template, wherein second template includes the second slot position;Obtain described first
First association user behavior of user behavior, wherein the first association user behavior is the phase with first user behavior
Guan Du is greater than the user behavior of threshold value;Second slot position is inserted into the first association user behavior, third is obtained and replys language
Sentence, wherein the third revert statement is for prompting the conversational system to be able to carry out the first association user behavior.
11. system according to claim 8, which is characterized in that
The reply module is also used to search third template, wherein the third template contains third slot position;By the user
It is intended to insert the third slot position, obtains second revert statement.
12. system according to claim 11, which is characterized in that
The reply module is also used to search the second template, wherein second template includes the second slot position;Obtain user's input
Second user sentence, wherein before the first user sentence, the second user sentence refers to the second user sentence
Second user behavior is shown;Obtain the second association user behavior of the second user behavior, wherein second association user
Behavior is the user behavior for being greater than threshold value with the degree of correlation of the second user behavior;The second association user behavior is inserted
Second slot position, to obtain the 4th revert statement, wherein the 4th revert statement is for prompting the conversational system energy
The behavior relevant to the second user behavior enough executed.
13. according to system described in claim 8 to 12 any claim, which is characterized in that
The reply module is also used to by the first user input by sentence user's intention assessment model, to obtain the user
It is intended to, wherein user's intention assessment model is to be trained to obtain by known users sentence and known users intention
's.
14. system according to claim 13, which is characterized in that user's intention assessment model is classifier, described
Classifier is any one in k nearest neighbor classifier, support vector machines, decision tree and random forest.
15. a kind of computer program product, which is characterized in that when the computer program product is read and executed by calculating equipment
When, the method as described in claim 1 to 8 any claim will be performed.
16. a kind of computer non-transitory storage media, which is characterized in that including instruction, when described instruction is transported on the computing device
When row, so that the method for calculating equipment execution as described in claim 1 to 8 any claim.
17. a kind of intelligent terminal, which is characterized in that including processor and memory, the processor is executed in the memory
Code execute method as described in claim 1 to 8 any claim.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910395865.5A CN110232108B (en) | 2019-05-13 | 2019-05-13 | Man-machine conversation method and conversation system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910395865.5A CN110232108B (en) | 2019-05-13 | 2019-05-13 | Man-machine conversation method and conversation system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110232108A true CN110232108A (en) | 2019-09-13 |
CN110232108B CN110232108B (en) | 2023-02-03 |
Family
ID=67860565
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910395865.5A Active CN110232108B (en) | 2019-05-13 | 2019-05-13 | Man-machine conversation method and conversation system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232108B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110704592A (en) * | 2019-09-27 | 2020-01-17 | 北京百度网讯科技有限公司 | Statement analysis processing method and device, computer equipment and storage medium |
CN113611316A (en) * | 2021-07-30 | 2021-11-05 | 百度在线网络技术(北京)有限公司 | Man-machine interaction method, device, equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106776926A (en) * | 2016-12-01 | 2017-05-31 | 竹间智能科技(上海)有限公司 | Improve the method and system of responsibility when robot talks with |
CN107193978A (en) * | 2017-05-26 | 2017-09-22 | 武汉泰迪智慧科技有限公司 | A kind of many wheel automatic chatting dialogue methods and system based on deep learning |
CN107562816A (en) * | 2017-08-16 | 2018-01-09 | 深圳狗尾草智能科技有限公司 | User view automatic identifying method and device |
US20180268818A1 (en) * | 2017-03-20 | 2018-09-20 | Ebay Inc. | Detection of mission change in conversation |
CN109145104A (en) * | 2018-09-29 | 2019-01-04 | 北京百度网讯科技有限公司 | For talking with interactive method and apparatus |
US20190019506A1 (en) * | 2017-07-14 | 2019-01-17 | International Business Machines Corporation | Dynamic personalized multi-turn interaction of cognitive models |
-
2019
- 2019-05-13 CN CN201910395865.5A patent/CN110232108B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106776926A (en) * | 2016-12-01 | 2017-05-31 | 竹间智能科技(上海)有限公司 | Improve the method and system of responsibility when robot talks with |
US20180268818A1 (en) * | 2017-03-20 | 2018-09-20 | Ebay Inc. | Detection of mission change in conversation |
CN107193978A (en) * | 2017-05-26 | 2017-09-22 | 武汉泰迪智慧科技有限公司 | A kind of many wheel automatic chatting dialogue methods and system based on deep learning |
US20190019506A1 (en) * | 2017-07-14 | 2019-01-17 | International Business Machines Corporation | Dynamic personalized multi-turn interaction of cognitive models |
CN107562816A (en) * | 2017-08-16 | 2018-01-09 | 深圳狗尾草智能科技有限公司 | User view automatic identifying method and device |
CN109145104A (en) * | 2018-09-29 | 2019-01-04 | 北京百度网讯科技有限公司 | For talking with interactive method and apparatus |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110704592A (en) * | 2019-09-27 | 2020-01-17 | 北京百度网讯科技有限公司 | Statement analysis processing method and device, computer equipment and storage medium |
CN110704592B (en) * | 2019-09-27 | 2021-06-04 | 北京百度网讯科技有限公司 | Statement analysis processing method and device, computer equipment and storage medium |
CN113611316A (en) * | 2021-07-30 | 2021-11-05 | 百度在线网络技术(北京)有限公司 | Man-machine interaction method, device, equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110232108B (en) | 2023-02-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11468246B2 (en) | Multi-turn dialogue response generation with template generation | |
US10446148B2 (en) | Dialogue system, a dialogue method and a method of adapting a dialogue system | |
US9865257B2 (en) | Device and method for a spoken dialogue system | |
CN110032682B (en) | Information recommendation list generation method, device and equipment | |
CN110442737A (en) | The twin method and system of number based on chart database | |
US11551159B2 (en) | Schema-guided response generation | |
CN109901896A (en) | Multi-task processing method in a kind of man-machine interactive system and man-machine interactive system | |
CN114564522B (en) | Intelligent push processing method and system based on block chain and big data mining | |
CN110347776A (en) | Interest point name matching process, device, equipment and storage medium | |
CN110232108A (en) | Interactive method and conversational system | |
WO2015062623A1 (en) | Group recommendation for selection of service items | |
WO2020081949A1 (en) | System and method for auto-completion of ics flow using artificial intelligence/machine learning | |
CN113641797A (en) | Data processing method, device, equipment, storage medium and computer program product | |
CN110704668B (en) | Grid-based collaborative attention VQA method and device | |
CN111310462A (en) | User attribute determination method, device, equipment and storage medium | |
CN113129010A (en) | Fraud group mining system and method based on complex network model | |
CN110222743A (en) | The method for generating plan of power grid accident | |
KR20140133032A (en) | A system and control method for a meta-heuristic algorithm utilizing similarity for performance enhancement | |
CN111126607B (en) | Data processing method, device and system for model training | |
CN113569017A (en) | Model processing method and device, electronic equipment and storage medium | |
CN116471281A (en) | Decentralised service combination method considering node selfiness | |
CN113191527A (en) | Prediction method and device for population prediction based on prediction model | |
CN111737319B (en) | User cluster prediction method, device, computer equipment and storage medium | |
Galinier et al. | Genetic algorithm to improve diversity in MDE | |
CN110096653A (en) | Construction method, device, equipment and the storage medium of space time information service architecture |
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 |