CN105787560A - Dialogue data interaction processing method and device based on recurrent neural network - Google Patents
Dialogue data interaction processing method and device based on recurrent neural network Download PDFInfo
- Publication number
- CN105787560A CN105787560A CN201610158385.3A CN201610158385A CN105787560A CN 105787560 A CN105787560 A CN 105787560A CN 201610158385 A CN201610158385 A CN 201610158385A CN 105787560 A CN105787560 A CN 105787560A
- Authority
- CN
- China
- Prior art keywords
- neural network
- recurrent neural
- recognition
- dialogue
- answer
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Abstract
The invention provides a dialogue data interaction processing method based on a recurrent neural network. The method comprises the following steps: receiving a dialogue input statement of a user; carrying out knowledge base matching calculation, and judging whether a knowledge base has a problem statement, the matching degree between which and the dialogue input statement reaches a preset value; if not, requesting a dialogue generation model to give an answer to the dialogue input statement, wherein a coding layer of the dialogue generation model is constructed into the recurrent neural network, analyzing the dialogue input statement in the recurrent neural network to obtain a middle vector expressing problem meanings, and a decoding layer of the dialogue generation model is also constructed into the recurrent neural network, analyzing the middle vector in the recurrent neural network to obtain an answer vector group expressing answer meanings; and taking the answer vector group as an answer output statement and outputting the answer output statement. The method can enable human-machine interaction to be smoother; and answers given by the dialogue generation model can further enable the knowledge base to be expanded and updated.
Description
Technical field
The present invention relates to field in intelligent robotics, specifically, relate to a kind of dialogue data interaction processing method based on Recognition with Recurrent Neural Network and device.
Background technology
Chat robots (chatterbot) is a program for simulating human dialogue or chat.Chat robots Producing reason is, developer is put into the answer that oneself is interested in data base, when a problem is thrown to chat robots, it passes through similarity mode algorithm, the most close problem is found from data base, then the corresponding relation according to question and answer, provides the properest answer, and what reply to it chats companion.
But, in the scene of current robot chat, when the same or like problem that the problem with user's request can not be found in robot knowledge base to match, robot cannot return correct answer suitable in other words to user.
Limiting to except the problem except causing robot knowledge base limited of prior art this respect, also result in the mistake of semantic understanding, so that user is poor with the experience of robot communication process.Additionally, the knowledge reasoning process of prior art also has certain limitation, traditional knowledge reasoning to be all write some rules to solve knowledge reasoning problem by application developer in knowledge reasoning.But, exhaustive and formulate these rules cannot imagine for developer.Because forever having in natural language processing field and can not writing not complete rule.At this moment it is accomplished by robot and has the learning capacity of oneself, and make inferences.
To sum up, in the mutual technical field of dialogue data, it is desirable to provide a kind of can the image training robot dialogue method that generates model, utilize the model that training obtains, robot can talk with user freely, thus improving the experience of user.
Summary of the invention
For solving the problems referred to above of prior art, the invention provides a kind of dialogue data interaction processing method based on Recognition with Recurrent Neural Network, the treating method comprises following steps:
Receive the dialogue read statement of user；
Carry out knowledge base matching primitives, it is judged that whether knowledge base exists the matching degree with described dialogue read statement and reaches the problem statement of predetermined value；
If it does not exist, then request dialogue generates model provides the answer for described dialogue read statement, wherein:
The coding layer building that described dialogue generates model is the first Recognition with Recurrent Neural Network, in described first Recognition with Recurrent Neural Network, described dialogue read statement is resolved, and obtains the intermediate vector that problem of representation is semantic；
Described dialogue generates solution to model code layer and is configured to the second Recognition with Recurrent Neural Network, in described second Recognition with Recurrent Neural Network, described intermediate vector is resolved, and obtains representing the answer vector group that answer is semantic；
Described answer vector group is exported as answer output statement.
According to one embodiment of present invention, when described dialogue read statement being resolved in described first Recognition with Recurrent Neural Network, described method is further comprising the steps of:
At coding layer, the conversation sentence of input is split into the minimum word unit with semanteme, and it is input to as problem vector group in vector form the input layer of described first Recognition with Recurrent Neural Network；
The output of the hidden layer of the first Recognition with Recurrent Neural Network described in the vector sum previous moment that the input layer of described first Recognition with Recurrent Neural Network is exported by the hidden layer of described first Recognition with Recurrent Neural Network carries out semantic parsing, go forward side by side line linearity weighted array, form the described intermediate vector of representative sentences justice；And
Described intermediate vector is exported by the output layer at described first Recognition with Recurrent Neural Network.
According to one embodiment of present invention, when described intermediate vector being resolved in described second Recognition with Recurrent Neural Network, described dialogue data interaction processing method is further comprising the steps of:
Described intermediate vector is received at decoding layer, and using the input as the input layer of described second Recognition with Recurrent Neural Network of the described intermediate vector；
Hidden layer at described second Recognition with Recurrent Neural Network carries out semantic parsing to from the output of the hidden layer of the second Recognition with Recurrent Neural Network described in the described intermediate vector of input layer and previous moment, sequentially generate some single vectors, to form described answer vector group, the semanteme of each single vector in wherein said answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；And
Described answer vector group is exported by the output layer at described second Recognition with Recurrent Neural Network.
Dialogue data interaction processing method according to the present invention, this answer output statement, after described answer vector group being exported as answer output statement, is saved in accordingly in knowledge base, knowledge base to be updated and expands by system with dialogue read statement.
According to one embodiment of present invention, after carrying out knowledge base matching primitives, the conversation sentence that system reaches predetermined value according to the matching degree whether existed in knowledge base with described dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
According to another aspect of the present invention, additionally providing a kind of dialogue data interaction process device based on Recognition with Recurrent Neural Network, this device includes with lower module:
Receiver module, it is for receiving the dialogue read statement of user；
Matching primitives module, it is used for carrying out knowledge base matching primitives, reaches the problem statement of predetermined value with the matching degree whether existed in knowledge base with described dialogue read statement；
Dialogue generates model calling module, and it is for when being absent from the problem statement of coupling, providing the answer for described dialogue read statement for asking dialogue to generate model, wherein, in this module:
The coding layer building that described dialogue generates model is the first Recognition with Recurrent Neural Network, in described first Recognition with Recurrent Neural Network, described dialogue read statement is resolved, and obtains the intermediate vector that problem of representation is semantic；
Described dialogue generates solution to model code layer and is configured to the second Recognition with Recurrent Neural Network, in described second Recognition with Recurrent Neural Network, described intermediate vector is resolved, and obtains representing the answer vector group that answer is semantic；And
Answer output module, it is for exporting described answer vector group as answer output statement.
According to a preferred embodiment of the invention, when described dialogue read statement being resolved in described first Recognition with Recurrent Neural Network, described device also includes with lower module:
First Recognition with Recurrent Neural Network input module, it for splitting into the minimum word unit with semanteme at coding layer by the conversation sentence of input, and it is input to the input layer of described first Recognition with Recurrent Neural Network in vector form；
First Recognition with Recurrent Neural Network hides module, its output of the hidden layer of the first Recognition with Recurrent Neural Network described in the vector sum previous moment that the input layer of described first Recognition with Recurrent Neural Network export at the hidden layer of described first Recognition with Recurrent Neural Network carries out semantic parsing, go forward side by side line linearity weighted array, form the described intermediate vector of representative sentences justice；And
First Recognition with Recurrent Neural Network output module, the output layer at described first Recognition with Recurrent Neural Network exports described intermediate vector.
According to one embodiment of present invention, when described intermediate vector being resolved in described second Recognition with Recurrent Neural Network, described device also includes with lower module:
Second Recognition with Recurrent Neural Network input module, it is for receiving described intermediate vector at decoding layer, and using the input as the input layer of described second Recognition with Recurrent Neural Network of the described intermediate vector；
Second Recognition with Recurrent Neural Network hides module, it is for carrying out semantic parsing at the hidden layer of described second Recognition with Recurrent Neural Network to from the output of the hidden layer of the second Recognition with Recurrent Neural Network described in the described intermediate vector of input layer and previous moment, sequentially generate some single vectors, to form described answer vector group, the semanteme of each single vector in wherein said answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；And
Second Recognition with Recurrent Neural Network output module, described answer vector group is exported by it for the output layer at described second Recognition with Recurrent Neural Network.
According to one embodiment of present invention, described device also includes more new module, it is for after exporting described answer vector group as answer output statement by output module, this answer output statement is saved in knowledge base accordingly with dialogue read statement, knowledge base is updated and expands.
According to one embodiment of present invention, after carrying out knowledge base matching primitives, the problem statement reaching predetermined value according to the matching degree whether existed in knowledge base with described dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
Dialogue data exchange method according to the present invention, during owing to can not find the problem statement of coupling in knowledge base, it is also possible to the mode being generated model by the dialogue trained is furnished an answer so that people is mutual with machine more smooth.Knowledge base can also be expanded and update by answer further that provide by talking with generation model, thus improving the level of intelligence of machine further.
Other features and advantages of the present invention will be set forth in the following description, and, partly become apparent from description, or understand by implementing the present invention.The purpose of the present invention and other advantages can be realized by structure specifically noted in description, claims and accompanying drawing and be obtained.
Accompanying drawing explanation
Accompanying drawing is for providing a further understanding of the present invention, and constitutes a part for description, is provided commonly for explaining the present invention with embodiments of the invention, is not intended that limitation of the present invention.In the accompanying drawings:
Fig. 1 is the flow chart of the problem of chat module process user's proposition of chat robots in prior art；
Fig. 2 is showing the schematic diagram of typical Recognition with Recurrent Neural Network；
Fig. 3 is the flow chart that all should use Recognition with Recurrent Neural Network algorithm according to one embodiment of present invention at coding layer and decoding layer；
Fig. 4 shows the theory diagram for training dialogue generation model according to the present invention；And
Fig. 5 shows that the dialogue of application present invention training generates the chat module of model and the example that other module is mutual.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, the embodiment of the present invention is described in further detail.
Prior art adopt problem matching primitives method come the flow chart finding answer in knowledge base as it is shown in figure 1, which show.In FIG, first, the chat module of robot receives the dialogue read statement of user, step S101.In one embodiment, this dialogue read statement is exactly enquirement or perhaps the problem that user actively initiates in fact.
It follows that interactive system by finding, according to matching degree computational methods, the problem statement that the problem proposed with user matches in the knowledge base of robot, referring to step S102.In the process found, it is possible to preset a threshold value about matching degree.When the matching degree calculated is less than this threshold value, it was shown that do not find the problem statement of coupling in data base.In other words, the statement Incomplete matching that the problem statement of data base inputs with user.So, system will carry out Similarity Measure, and the statement of input is carried out similarity transformation, then finds corresponding similar Railway Project in data base, referring to step S103.After have found similar problem, system will export the several answers corresponding with these Similar Problems statements, referring to step S105.
After Similarity Measure, the problem through similarity transformation is still absent from data base, then directly export the result of " can not export answer " in step s 106.
When the matching degree calculated is more than or equal to this threshold value, it was shown that have the problem statement that the problem statement proposed with user mates completely in data base.In this case, the answer that in the direct output database of system, the problem statement mated completely of storage is corresponding.
Such as, if storing the problem being similar to " what is your name " and corresponding answer in knowledge base, when the problem so putd question to as user is " what you cry ", robot is possible to can not find in existing knowledge base the appropriate problem of coupling according to above-mentioned similarity calculating method.Therefore, also correct answer cannot be returned to user.
In this case, although user proposes two problems literally entirely different but equivalent in meaning, but can not correct understanding semanteme according to the robot of prior art.Cannot answer correctly hence for this problem, cause that the process of chat can be unsatisfactory.Obviously, this is undesirable.
The present invention proposes a kind of method for designing, dialogue is generated the method supplementary mode as knowledge base matching process of model, the mode combining dialogue generation model is adopted to furnish an answer in the unconformable situation of knowledge base matching process, thus improving the interactive experience of user.
According to the present invention, when user puts question to, being satisfied with answer if can not find in knowledge base, system can generate model according to the problem of user by the dialogue trained and provide a user with answer.
Dialogue described in the present invention generates the mode of model, it is intended that: when user's asked questions, this model can according to the problem of user based on the model generation answer trained.And the method return answer of knowledge base coupling it is based on unlike original question answering system.Further, in the process that dialog model generates, word for word or answer is generated by word based on the problem putd question to.This mode mainly solves limited when the problem in knowledge base and without answer return technical problem.
It is said that in general, it is coding-decoding framework that dialogue generates the structure of model.In this framework, model is mainly made up of coding layer and decoding layer two parts.Wherein problem is reached the purpose of semantic understanding by coding layer primary responsibility, and one vector of problem representation, this vector is exactly the semantic expressiveness of problem.And the vector that decoding layer primary responsibility generates based on coding layer generates answer.
It is based on Recognition with Recurrent Neural Network (RecurrentNeuralNetworks owing to the training of the dialog model of the present invention generates, RNNs) algorithm, therefore, according to the present invention it is possible to coding layer and the decoding layer of dialog model are all configured to the form of Recognition with Recurrent Neural Network.
It is known that the purpose of RNNs is used to process sequence data.In traditional neural network model, being again to output layer from input layer to hidden layer, be full connection between layers, the node between every layer is connectionless.But this common neutral net is for a lot of problems but helpless.Such as, you to predict what the next word of sentence is, it is generally required to use word above, because word is not independent before and after in a sentence.
Why RNNs is called circulation neural network, and the output that namely sequence is current is also relevant with output above.The concrete form of expression is in the calculating that information above can be remembered and be applied to current output by network, namely the node between hidden layer no longer has connection without connecting, and the input of hidden layer not only includes the output of input layer and also includes the output of a moment hidden layer.In theory, the sequence data of any length can be processed by RNNs.
But in practice, the complexity in order to reduce network often assumes that current state is only relevant to several states above.If Fig. 2 is the schematic diagram of a typical Recognition with Recurrent Neural Network RNNs.
In fig. 2, Recognition with Recurrent Neural Network includes three layers from top to bottom: input layer, hidden layer and output layer.Input layer is used for receiving initial data sequence, and for dialogue, this initial data sequence is carry out, according to minimum semantic primitive, the word one by one that splits.Hidden layer completes topmost work, it is for receiving the vector that these words are constituted, and these words one by one are carried out semantic parsing by the vector of the output of the hidden layer before providing according to mnemon therein under the model trained and currently received input layer, obtain another vector.Output layer is used for exporting this vector.
Therefore, the training based on the dialogue generation model of Recognition with Recurrent Neural Network is critically important.That trains is good, and what it reflected is exactly objective data.But, for the purpose of the fuzzy present invention, the process trained will not be described in detail here.
As it is shown on figure 3, which show a kind of dialogue data interaction processing method based on Recognition with Recurrent Neural Network according to the present invention.Wherein owing to part steps repeats with knowledge base matching process of the prior art, therefore, figure does not show.
In figure 3, method starts from step S301, as it was previously stated, system needs first to receive the dialogue read statement of user.Owing to the method for the present invention is the auxiliary as knowledge base matching process, therefore before utilizing dialog model to provide answer, need to carry out the judgement of matching degree, carry out knowledge base matching primitives, it is judged that whether knowledge base exists the matching degree with dialogue read statement and reaches the problem statement of predetermined value.
After carrying out knowledge base matching primitives, the conversation sentence that system reaches predetermined value according to the matching degree whether existed in knowledge base with dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
If it does not, namely request flag signal position is effective, then request dialogue generates model and provides the answer for dialogue read statement, wherein:
It is the first Recognition with Recurrent Neural Network that dialogue generates the coding layer building of model, such as the coding layer RNN of Fig. 4.Dialogue read statement (ABC) is resolved by the first Recognition with Recurrent Neural Network, line linearity weighted array of going forward side by side, form the intermediate vector V representing problem sentence justice.Meanwhile, dialogue generates solution to model code layer and is also configured to the second Recognition with Recurrent Neural Network, decoding layer RNN as shown in Figure 4.Intermediate vector V is resolved by the second Recognition with Recurrent Neural Network, obtains successively representing the answer vector group (" D ", " E ", " F ") that answer is semantic.Finally answer vector group is exported as answer output statement (DEF).
Foregoing can be embodied as, in step s 302, at the coding layer of dialog model, the conversation sentence of input is split into minimum word unit (A, the B with semanteme, and successively it is input in vector form the input layer of the first Recognition with Recurrent Neural Network C),.The output of the hidden layer of vector sum previous moment the first Recognition with Recurrent Neural Network that the input layer of the first Recognition with Recurrent Neural Network exports is carried out semantic parsing, line linearity weighted array of going forward side by side by the hidden layer at the first Recognition with Recurrent Neural Network, forms the intermediate vector V of representative sentences justice.Finally, at the output layer of the first Recognition with Recurrent Neural Network, intermediate vector is exported.
Foregoing can also be further detailed as, in step S303, when intermediate vector V being resolved in the second Recognition with Recurrent Neural Network, dialogue data interaction processing method is further comprising the steps of: receive intermediate vector V at decoding layer, and using the intermediate vector input as the input layer of the second Recognition with Recurrent Neural Network.The output of the hidden layer from the intermediate vector of input layer and previous moment the second Recognition with Recurrent Neural Network is carried out analytical Calculation by the hidden layer at the second Recognition with Recurrent Neural Network, sequentially generate answer vector group (" D ", " E ", " F "), the semanteme of each vector in wherein answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；And at the output layer of the second Recognition with Recurrent Neural Network, described answer vector group is exported.
Dialogue data interaction processing method according to the present invention, system is after exporting answer vector group as answer output statement, this answer output statement can also be saved in knowledge base accordingly with dialogue read statement, knowledge base is updated and expands, step S304.
When the method flow of Fig. 3 is described with an instantiation, for example it may be that such:
Such as, when user asks " what you cry " problem, first it is that knowledge base matching algorithm finds Similar Problems in knowledge base, when finding the higher problem of similarity such as " what name you are ", have a marking signal 0, be used for representing and dialogue need not be asked to generate model.Meanwhile, return answer (" I is intelligence the baby ") conduct that in knowledge base, this problem (" what name you are ") is corresponding to reply.And when the problem using knowledge base matching algorithm that similarity cannot be found high, marking signal 1 can be produced, it is used for representing that needs request dialogue generates model.And using the problem (" what you cry ") after participle as talking with the input generating model, afterwards dialogue is generated the answer (" I is intelligence baby ") of model generation as replying.
After knowledge base matching algorithm end of run, dialog model can judge whether to use dialogue to generate model generation answer according to marking signal (1 or 0), if 1 uses this model, otherwise need not.When marking signal is 1, first to the problematic character string such as " what you cry " after the participle transmitted.At coding layer, successively " you ", " crying ", " what " is input to the Recognition with Recurrent Neural Network of coding layer.Can generate afterwards one represent semanteme vectorial D, then at decoding layer, coding layer generate vectorial D as the input of decoding layer Recognition with Recurrent Neural Network, sequentially generate " I ", " cry ", " intelligence baby " answer.
As it is shown in figure 5, which show the dialogue containing the present invention to generate chat module and other module example that such as Authority Verification module and decision-making module use alternately of model.As it can be seen, the module containing dialogue generation model is under the jurisdiction of chat module.
It should be strongly noted that the present invention method describe realize in computer systems.This computer system such as can be arranged in the control core processor of robot.Such as, method described herein can be implemented as can to control the software that logic performs, and it is performed by the CPU in robot control system.Function as herein described can be implemented as the programmed instruction set being stored in non-transitory tangible computer computer-readable recording medium.When implemented in this fashion, this computer program includes one group of instruction, and when this group instruction is run by computer, it promotes the method that computer performs to implement above-mentioned functions.FPGA can temporarily or permanently be arranged in non-transitory tangible computer computer-readable recording medium, for instance ROM chip, computer storage, disk or other storage mediums.Except realizing with software, logic as herein described may utilize discrete parts, integrated circuit and programmable logic device (such as, field programmable gate array (FPGA) or microprocessor) combine the FPGA used, or any other equipment including they combination in any embodies.These type of embodiments all are intended to fall under within the scope of the present invention.
Therefore, according to another aspect of the present invention, additionally providing a kind of dialogue data interaction process device based on Recognition with Recurrent Neural Network, this device includes with lower module:
Receiver module, it is for receiving the dialogue read statement of user；
Matching primitives module, it is used for carrying out knowledge base matching primitives, reaches the problem statement of predetermined value with the matching degree whether existed in knowledge base with described dialogue read statement；
Dialogue generates model calling module, and it is for when being absent from the problem statement of coupling, providing the answer for described dialogue read statement for asking dialogue to generate model, wherein, in this module:
The coding layer building that described dialogue generates model is the first Recognition with Recurrent Neural Network, in described first Recognition with Recurrent Neural Network, described dialogue read statement is resolved, and obtains the intermediate vector that problem of representation is semantic；
Described dialogue generates solution to model code layer and is configured to the second Recognition with Recurrent Neural Network, in described second Recognition with Recurrent Neural Network, described intermediate vector is resolved, and obtains representing the answer vector group that answer is semantic；And
Answer output module, it is for exporting described answer vector group as answer output statement.
According to a preferred embodiment of the invention, when described dialogue read statement being resolved in described first Recognition with Recurrent Neural Network, described device also includes with lower module:
First Recognition with Recurrent Neural Network input module, it for splitting into the minimum word unit with semanteme at coding layer by the conversation sentence of input, and it is input to the input layer of described first Recognition with Recurrent Neural Network successively in vector form；
First Recognition with Recurrent Neural Network hides module, its output of the hidden layer of the first Recognition with Recurrent Neural Network described in the vector sum previous moment that the input layer of described first Recognition with Recurrent Neural Network export at the hidden layer of described first Recognition with Recurrent Neural Network carries out semantic parsing, go forward side by side line linearity weighted array, form the described intermediate vector of representative sentences justice；And
First Recognition with Recurrent Neural Network output module, the output layer at described first Recognition with Recurrent Neural Network exports described intermediate vector.
According to one embodiment of present invention, when described intermediate vector being resolved in described second Recognition with Recurrent Neural Network, described device also includes with lower module:
Second Recognition with Recurrent Neural Network input module, it is for receiving described intermediate vector at decoding layer, and using the input as the input layer of described second Recognition with Recurrent Neural Network of the described intermediate vector；
Second Recognition with Recurrent Neural Network hides module, it is for carrying out semantic parsing at the hidden layer of described second Recognition with Recurrent Neural Network to from the output of the hidden layer of the second Recognition with Recurrent Neural Network described in the described intermediate vector of input layer and previous moment, sequentially generate some single vectors, to form described answer vector group, the semanteme of each single vector in wherein said answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；And
Second Recognition with Recurrent Neural Network output module, described answer vector group is exported by it for the output layer at described second Recognition with Recurrent Neural Network.
According to one embodiment of present invention, described device also includes more new module, it is for after exporting described answer vector group as answer output statement by output module, this answer output statement is saved in knowledge base accordingly with dialogue read statement, knowledge base is updated and expands.
According to one embodiment of present invention, after carrying out knowledge base matching primitives, the problem statement reaching predetermined value according to the matching degree whether existed in knowledge base with described dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
As mentioned above, when the problem of user's request is mated less than same or similar problem in our knowledge base, use the dialogue based on Recognition with Recurrent Neural Network trained to generate model and can generate a suitable answer, return to user, be greatly enriched the experience of user.Such as there is the problem being similar to " what is your name " in our knowledge base, but when user puts question to the problem of similar " what you cry ", use existing similarity calculating method to be likely to mate less than suitable problem in existing knowledge base.Therefore, cannot correctly return to one suitable answer of user, although that is user has said that two from literal diverse sentence, but look like equally, this be exactly basically robot can not correct understanding semantic, and the dialogue of the present invention generates model and just can solve this problem, it can correct understanding semanteme.
In the long run/term, this dialogue generates model and can also solve problems with:
When the language material that training dialogue generates model is sufficiently large, it is also popular knowledge base matching process that this model just can replace main in present question answering system.Traditional knowledge base matching process is all the similarity from literal judgement two word.And this is for understanding natural language, especially Chinese is far from being enough, and the dialogue of the present invention generates model and just can solve this problem.
Such as following two words " I be you most dear friends " and " I be not you most dear friends ", if very likely thinking this two words semantic similitude by traditional knowledge base matching algorithm, but reality is but contrary, therefore this is excessively poor for Consumer's Experience.
On the other hand, this dialogue generates the problem that model can solve some knowledge reasonings.Traditional knowledge reasoning problem is all write some rules by application developer to solve knowledge reasoning problem, but this is inconceivable for application developer, because having in natural language processing field and can not writing not complete rule, therefore no matter for company, developer, or user cost is all huge, and use the dialogue trained to generate model and can perfectly solve this problem, because in a collection of special language material, this model can automatically learn some inferenctial knowledges.Such as, " I has three red Fructus Mali pumilaes and four green Fructus Mali pumilaes; may I ask my total several Fructus Mali pumilaes " it is similar to when user puts question to, traditional method writes the rule of some similar " numeral+entity "+" numeral+entity "=" numeral+entities " exactly, solve this problem, but sentence somewhat changes " I has three red Fructus Mali pumilaes; eaten one; also have several Fructus Mali pumilaes ", and above rule is just inapplicable, and uses the dialog model based on specific corpus to may learn this knowledge completely.
It should be understood that disclosed embodiment of this invention is not limited to ad hoc structure disclosed herein, processes step or material, and the equivalent replacement of these features that those of ordinary skill in the related art understand should be extended to.It is to be further understood that term is only for the purpose of describing particular embodiments as used herein, and be not intended to limit.
Special characteristic, structure or characteristic that " embodiment " mentioned in description or " embodiment " mean to describe in conjunction with the embodiments include at least one embodiment of the present invention.Therefore, description various places throughout occurs phrase " embodiment " or " embodiment " might not refer both to same embodiment.
While it is disclosed that embodiment as above, but described content is only to facilitate the embodiment understanding the present invention and adopt, is not limited to the present invention.Technical staff in any the technical field of the invention; under the premise without departing from spirit and scope disclosed in this invention; any amendment and change can be done in the formal and details implemented; but the scope of patent protection of the present invention, still must be as the criterion with the scope that appending claims defines.
Claims (10)
1. the dialogue data interaction processing method based on Recognition with Recurrent Neural Network, it is characterised in that the treating method comprises following steps:
Receive the dialogue read statement of user；
Carry out knowledge base matching primitives, it is judged that whether knowledge base exists the matching degree with described dialogue read statement and reaches the problem statement of predetermined value；
If it does not exist, then request dialogue generates model provides the answer for described dialogue read statement, wherein:
The coding layer building that described dialogue generates model is the first Recognition with Recurrent Neural Network, in described first Recognition with Recurrent Neural Network, described dialogue read statement is resolved, and obtains the intermediate vector that problem of representation is semantic；
Described dialogue generates solution to model code layer and is configured to the second Recognition with Recurrent Neural Network, in described second Recognition with Recurrent Neural Network, described intermediate vector is resolved, and obtains representing the answer vector group that answer is semantic；And
Described answer vector group is exported as answer output statement.
2. dialogue data interaction processing method according to claim 1, it is characterised in that when described dialogue read statement being resolved in described first Recognition with Recurrent Neural Network, described method is further comprising the steps of:
At coding layer, the conversation sentence of input is split into the minimum word unit with semanteme, and it is input to as problem vector group in vector form the input layer of described first Recognition with Recurrent Neural Network；
The output of the hidden layer of the first Recognition with Recurrent Neural Network described in the output of the input layer of described first Recognition with Recurrent Neural Network and previous moment is carried out semantic parsing by the hidden layer at described first Recognition with Recurrent Neural Network, go forward side by side line linearity weighted array, form the intermediate vector of representative sentences justice；
Described intermediate vector is exported by the output layer at described first Recognition with Recurrent Neural Network.
3. dialogue data interaction processing method according to claim 1, it is characterised in that when described intermediate vector being resolved in described second Recognition with Recurrent Neural Network, described method is further comprising the steps of:
Described intermediate vector is received at decoding layer, and using the input as the input layer of described second Recognition with Recurrent Neural Network of the described intermediate vector；
Hidden layer at described second Recognition with Recurrent Neural Network carries out semantic parsing to from the output of the hidden layer of the second Recognition with Recurrent Neural Network described in the described intermediate vector of input layer and previous moment, sequentially generate some single vectors, to form described answer vector group, the semanteme of each single vector in wherein said answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；
Described answer vector group is exported by the output layer at described second Recognition with Recurrent Neural Network.
4. the dialogue data interaction processing method according to any one of claim 1-3, it is characterized in that, after described answer vector group is exported as answer output statement, this answer output statement is saved in knowledge base accordingly with dialogue read statement, knowledge base is updated and expands.
5. the dialogue data interaction processing method according to any one of claim 1-3, it is characterized in that, after carrying out knowledge base matching primitives, the conversation sentence reaching predetermined value according to the matching degree whether existed in knowledge base with described dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
6. the dialogue data interaction process device based on Recognition with Recurrent Neural Network, it is characterised in that described device includes with lower module:
Receiver module, it is for receiving the dialogue read statement of user；
Matching primitives module, it is used for carrying out knowledge base matching primitives, reaches the problem statement of predetermined value with the matching degree whether existed in knowledge base with described dialogue read statement；
Dialogue generates model calling module, and it is for when being absent from the problem statement of coupling, providing the answer for described dialogue read statement for asking dialogue to generate model, wherein, in this module:
The coding layer building that described dialogue generates model is the first Recognition with Recurrent Neural Network, in described first Recognition with Recurrent Neural Network, described dialogue read statement is resolved, and obtains the intermediate vector that problem of representation is semantic；
Described dialogue generates solution to model code layer and is configured to the second Recognition with Recurrent Neural Network, in described second Recognition with Recurrent Neural Network, described intermediate vector is resolved, and obtains representing the answer vector group that answer is semantic；
Answer output module, it is for exporting described answer vector group as answer output statement.
7. dialogue data interaction process device according to claim 6, it is characterised in that when described dialogue read statement being resolved in described first Recognition with Recurrent Neural Network, described device also includes with lower module:
First Recognition with Recurrent Neural Network input module, it for splitting into the minimum word unit with semanteme at coding layer by the conversation sentence of input, and it is input to the input layer of described first Recognition with Recurrent Neural Network in vector form as problem vector group；
First Recognition with Recurrent Neural Network hides module, it is for carrying out semantic parsing at the hidden layer of described first Recognition with Recurrent Neural Network to the output of the hidden layer of the first Recognition with Recurrent Neural Network described in the output of the input layer of described first Recognition with Recurrent Neural Network and previous moment, go forward side by side line linearity weighted array, form the described intermediate vector of representative sentences justice；
First Recognition with Recurrent Neural Network output module, the output layer at described first Recognition with Recurrent Neural Network exports described intermediate vector.
8. dialogue data interaction process device according to claim 6, it is characterised in that when described intermediate vector being resolved in described second Recognition with Recurrent Neural Network, described device also includes with lower module:
Second Recognition with Recurrent Neural Network input module, it is for receiving described intermediate vector at decoding layer, and using the input as the input layer of described second Recognition with Recurrent Neural Network of the described intermediate vector；
Second Recognition with Recurrent Neural Network hides module, it is for carrying out semantic parsing at the hidden layer of described second Recognition with Recurrent Neural Network to from the output of the hidden layer of the second Recognition with Recurrent Neural Network described in the described intermediate vector of input layer and previous moment, sequentially generate some single vectors, to form described answer vector group, the semanteme of each single vector in wherein said answer vector group is corresponding to the semanteme of minimum word unit in answer output statement；
Second Recognition with Recurrent Neural Network output module, described answer vector group is exported by it for the output layer at described second Recognition with Recurrent Neural Network.
9. the dialogue data interaction process device according to any one of claim 6-8, it is characterized in that, described device also includes more new module, it is for after exporting described answer vector group as answer output statement by output module, this answer output statement is saved in knowledge base accordingly with dialogue read statement, knowledge base is updated and expands.
10. the dialogue data interaction process device according to any one of claim 6-8, it is characterized in that, after carrying out knowledge base matching primitives, the problem statement reaching predetermined value according to the matching degree whether existed in knowledge base with described dialogue read statement arranges request flag signal position, and decides whether that request dialogue generates model and provides answer according to the effectiveness of request flag signal position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610158385.3A CN105787560B (en) | 2016-03-18 | 2016-03-18 | Dialogue data interaction processing method and device based on Recognition with Recurrent Neural Network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610158385.3A CN105787560B (en) | 2016-03-18 | 2016-03-18 | Dialogue data interaction processing method and device based on Recognition with Recurrent Neural Network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105787560A true CN105787560A (en) | 2016-07-20 |
| CN105787560B CN105787560B (en) | 2018-04-03 |
Family
ID=56394140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610158385.3A Active CN105787560B (en) | 2016-03-18 | 2016-03-18 | Dialogue data interaction processing method and device based on Recognition with Recurrent Neural Network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105787560B (en) |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106297789A (en) * | 2016-08-19 | 2017-01-04 | 北京光年无限科技有限公司 | The personalized interaction method of intelligent robot and interactive system |
| CN106294726A (en) * | 2016-08-09 | 2017-01-04 | 北京光年无限科技有限公司 | Based on the processing method and processing device that robot role is mutual |
| CN106295792A (en) * | 2016-08-05 | 2017-01-04 | 北京光年无限科技有限公司 | Dialogue data interaction processing method based on multi-model output and device |
| CN106446141A (en) * | 2016-09-21 | 2017-02-22 | 北京光年无限科技有限公司 | Interaction data processing method for intelligent robot system and robot system |
| CN106528530A (en) * | 2016-10-24 | 2017-03-22 | 北京光年无限科技有限公司 | Method and device for determining sentence type |
| CN106847271A (en) * | 2016-12-12 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of data processing method and device for talking with interactive system |
| CN106874406A (en) * | 2017-01-18 | 2017-06-20 | 北京光年无限科技有限公司 | A kind of interactive output intent for robot |
| CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
| CN106997375A (en) * | 2017-02-28 | 2017-08-01 | 浙江大学 | Recommendation method is replied in customer service based on deep learning |
| CN107220296A (en) * | 2017-04-28 | 2017-09-29 | 北京拓尔思信息技术股份有限公司 | The generation method of question and answer knowledge base, the training method of neutral net and equipment |
| CN107665362A (en) * | 2017-09-26 | 2018-02-06 | 联想(北京)有限公司 | For realizing the training method of robot chat, predicting the method and device of answer |
| CN107665706A (en) * | 2016-07-29 | 2018-02-06 | 科大讯飞股份有限公司 | Rapid Speech exchange method and system |
| CN107846350A (en) * | 2016-09-19 | 2018-03-27 | Tcl集团股份有限公司 | A kind of method, computer-readable medium and the system of context-aware Internet chat |
| WO2018058994A1 (en) * | 2016-09-30 | 2018-04-05 | 华为技术有限公司 | Dialogue method, apparatus and device based on deep learning |
| CN107943788A (en) * | 2017-11-17 | 2018-04-20 | 平安科技（深圳）有限公司 | Enterprise's abbreviation generation method, device and storage medium |
| CN107967302A (en) * | 2017-11-08 | 2018-04-27 | 江苏名通信息科技有限公司 | Game customer service conversational system based on deep neural network |
| CN108153913A (en) * | 2018-01-24 | 2018-06-12 | 北京神州泰岳软件股份有限公司 | Return information generates training method, return information generation method and the device of model |
| CN108509519A (en) * | 2018-03-09 | 2018-09-07 | 北京邮电大学 | World knowledge collection of illustrative plates enhancing question and answer interactive system based on deep learning and method |
| WO2018196684A1 (en) * | 2017-04-24 | 2018-11-01 | 北京京东尚科信息技术有限公司 | Method and device for generating conversational robot |
| CN108733703A (en) * | 2017-04-20 | 2018-11-02 | 北京京东尚科信息技术有限公司 | The answer prediction technique and device of question answering system, electronic equipment, storage medium |
| CN108804488A (en) * | 2017-04-26 | 2018-11-13 | 谷歌有限责任公司 | Dialog process.It is instantiated at specific child status |
| CN108804611A (en) * | 2018-05-30 | 2018-11-13 | 浙江大学 | A kind of dialogue reply generation method and system based on self comment Sequence Learning |
| CN109002500A (en) * | 2018-06-29 | 2018-12-14 | 北京百度网讯科技有限公司 | Talk with generation method, device, equipment and computer-readable medium |
| CN109241256A (en) * | 2018-08-20 | 2019-01-18 | 百度在线网络技术（北京）有限公司 | Dialog process method, apparatus, computer equipment and readable storage medium storing program for executing |
| CN109361823A (en) * | 2018-11-01 | 2019-02-19 | 深圳市号互联科技有限公司 | A kind of intelligent interaction mode that voice is mutually converted with text |
| CN109508457A (en) * | 2018-10-31 | 2019-03-22 | 浙江大学 | A kind of transfer learning method reading series model based on machine |
| CN109547320A (en) * | 2018-09-29 | 2019-03-29 | 阿里巴巴集团控股有限公司 | Social contact method, device and equipment |
| CN109657127A (en) * | 2018-12-17 | 2019-04-19 | 北京百度网讯科技有限公司 | A kind of answer acquisition methods, device, server and storage medium |
| WO2019080648A1 (en) * | 2017-10-26 | 2019-05-02 | 华为技术有限公司 | Retelling sentence generation method and apparatus |
| CN109933773A (en) * | 2017-12-15 | 2019-06-25 | 上海擎语信息科技有限公司 | A kind of multiple semantic sentence analysis system and method |
| CN109977207A (en) * | 2019-03-21 | 2019-07-05 | 网易（杭州）网络有限公司 | Talk with generation method, dialogue generating means, electronic equipment and storage medium |
| CN110222164A (en) * | 2019-06-13 | 2019-09-10 | 腾讯科技（深圳）有限公司 | A kind of Question-Answering Model training method, problem sentence processing method, device and storage medium |
| CN110489532A (en) * | 2018-05-14 | 2019-11-22 | 西门子股份公司 | The computer implemented method and self adaptive conversation system of self adaptive conversation system are provided |
| CN110728356A (en) * | 2019-09-17 | 2020-01-24 | 阿里巴巴集团控股有限公司 | Dialogue method and system based on recurrent neural network and electronic equipment |
| CN110874402A (en) * | 2018-08-29 | 2020-03-10 | 北京三星通信技术研究有限公司 | Reply generation method, device and computer readable medium based on personalized information |
| CN111090664A (en) * | 2019-07-18 | 2020-05-01 | 重庆大学 | High-imitation person multi-mode dialogue method based on neural network |
| CN113032540A (en) * | 2021-03-19 | 2021-06-25 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment and storage medium |
| CN113806509A (en) * | 2021-09-18 | 2021-12-17 | 上海景吾智能科技有限公司 | Robot dialogue system and method based on knowledge graph and RNN neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140324429A1 (en) * | 2013-04-25 | 2014-10-30 | Elektrobit Automotive Gmbh | Computer-implemented method for automatic training of a dialogue system, and dialogue system for generating semantic annotations |
| CN104951428A (en) * | 2014-03-26 | 2015-09-30 | 阿里巴巴集团控股有限公司 | User intention recognition method and device |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105183848A (en) * | 2015-09-07 | 2015-12-23 | 百度在线网络技术（北京）有限公司 | Human-computer chatting method and device based on artificial intelligence |
-
2016
- 2016-03-18 CN CN201610158385.3A patent/CN105787560B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140324429A1 (en) * | 2013-04-25 | 2014-10-30 | Elektrobit Automotive Gmbh | Computer-implemented method for automatic training of a dialogue system, and dialogue system for generating semantic annotations |
| CN104951428A (en) * | 2014-03-26 | 2015-09-30 | 阿里巴巴集团控股有限公司 | User intention recognition method and device |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105183848A (en) * | 2015-09-07 | 2015-12-23 | 百度在线网络技术（北京）有限公司 | Human-computer chatting method and device based on artificial intelligence |
Non-Patent Citations (1)
| Title |
|---|
| DONGHO KIM等: "Stochastic Language Generation in Dialogue using Recurrent Neural Networks with Convolutional Sentence...", 《COMPUTER SCIENCE》 * |
Cited By (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665706A (en) * | 2016-07-29 | 2018-02-06 | 科大讯飞股份有限公司 | Rapid Speech exchange method and system |
| CN106295792A (en) * | 2016-08-05 | 2017-01-04 | 北京光年无限科技有限公司 | Dialogue data interaction processing method based on multi-model output and device |
| CN106295792B (en) * | 2016-08-05 | 2019-08-20 | 北京光年无限科技有限公司 | Dialogue data interaction processing method and device based on multi-model output |
| CN106294726A (en) * | 2016-08-09 | 2017-01-04 | 北京光年无限科技有限公司 | Based on the processing method and processing device that robot role is mutual |
| CN106297789A (en) * | 2016-08-19 | 2017-01-04 | 北京光年无限科技有限公司 | The personalized interaction method of intelligent robot and interactive system |
| CN106297789B (en) * | 2016-08-19 | 2020-01-14 | 北京光年无限科技有限公司 | Personalized interaction method and system for intelligent robot |
| CN107846350A (en) * | 2016-09-19 | 2018-03-27 | Tcl集团股份有限公司 | A kind of method, computer-readable medium and the system of context-aware Internet chat |
| CN106446141A (en) * | 2016-09-21 | 2017-02-22 | 北京光年无限科技有限公司 | Interaction data processing method for intelligent robot system and robot system |
| CN106446141B (en) * | 2016-09-21 | 2020-05-19 | 北京光年无限科技有限公司 | Interactive data processing method for intelligent robot system and robot system |
| CN107885756B (en) * | 2016-09-30 | 2020-05-08 | 华为技术有限公司 | Deep learning-based dialogue method, device and equipment |
| US11449678B2 (en) | 2016-09-30 | 2022-09-20 | Huawei Technologies Co., Ltd. | Deep learning based dialog method, apparatus, and device |
| WO2018058994A1 (en) * | 2016-09-30 | 2018-04-05 | 华为技术有限公司 | Dialogue method, apparatus and device based on deep learning |
| CN107885756A (en) * | 2016-09-30 | 2018-04-06 | 华为技术有限公司 | Dialogue method, device and equipment based on deep learning |
| CN106528530A (en) * | 2016-10-24 | 2017-03-22 | 北京光年无限科技有限公司 | Method and device for determining sentence type |
| CN106847271A (en) * | 2016-12-12 | 2017-06-13 | 北京光年无限科技有限公司 | A kind of data processing method and device for talking with interactive system |
| CN106874406A (en) * | 2017-01-18 | 2017-06-20 | 北京光年无限科技有限公司 | A kind of interactive output intent for robot |
| CN106997375A (en) * | 2017-02-28 | 2017-08-01 | 浙江大学 | Recommendation method is replied in customer service based on deep learning |
| CN106997375B (en) * | 2017-02-28 | 2020-08-18 | 浙江大学 | Customer service reply recommendation method based on deep learning |
| CN106875940B (en) * | 2017-03-06 | 2020-08-14 | 吉林省盛创科技有限公司 | Machine self-learning construction knowledge graph training method based on neural network |
| CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
| CN108733703A (en) * | 2017-04-20 | 2018-11-02 | 北京京东尚科信息技术有限公司 | The answer prediction technique and device of question answering system, electronic equipment, storage medium |
| WO2018196684A1 (en) * | 2017-04-24 | 2018-11-01 | 北京京东尚科信息技术有限公司 | Method and device for generating conversational robot |
| CN108733722A (en) * | 2017-04-24 | 2018-11-02 | 北京京东尚科信息技术有限公司 | A kind of dialogue robot automatic generation method and device |
| CN108733722B (en) * | 2017-04-24 | 2020-07-31 | 北京京东尚科信息技术有限公司 | Automatic generation method and device for conversation robot |
| CN108804488A (en) * | 2017-04-26 | 2018-11-13 | 谷歌有限责任公司 | Dialog process.It is instantiated at specific child status |
| US11381530B2 (en) | 2017-04-26 | 2022-07-05 | Google Llc | Instantiation of dialog process at a particular child node state |
| CN107220296B (en) * | 2017-04-28 | 2020-01-17 | 北京拓尔思信息技术股份有限公司 | Method for generating question-answer knowledge base, method and equipment for training neural network |
| CN107220296A (en) * | 2017-04-28 | 2017-09-29 | 北京拓尔思信息技术股份有限公司 | The generation method of question and answer knowledge base, the training method of neutral net and equipment |
| CN107665362A (en) * | 2017-09-26 | 2018-02-06 | 联想(北京)有限公司 | For realizing the training method of robot chat, predicting the method and device of answer |
| US11586814B2 (en) | 2017-10-26 | 2023-02-21 | Huawei Technologies Co., Ltd. | Paraphrase sentence generation method and apparatus |
| WO2019080648A1 (en) * | 2017-10-26 | 2019-05-02 | 华为技术有限公司 | Retelling sentence generation method and apparatus |
| CN107967302A (en) * | 2017-11-08 | 2018-04-27 | 江苏名通信息科技有限公司 | Game customer service conversational system based on deep neural network |
| CN107943788A (en) * | 2017-11-17 | 2018-04-20 | 平安科技（深圳）有限公司 | Enterprise's abbreviation generation method, device and storage medium |
| CN109933773A (en) * | 2017-12-15 | 2019-06-25 | 上海擎语信息科技有限公司 | A kind of multiple semantic sentence analysis system and method |
| CN109933773B (en) * | 2017-12-15 | 2023-05-26 | 上海擎语信息科技有限公司 | Multiple semantic statement analysis system and method |
| CN108153913B (en) * | 2018-01-24 | 2020-08-07 | 鼎富智能科技有限公司 | Training method of reply information generation model, reply information generation method and device |
| CN108153913A (en) * | 2018-01-24 | 2018-06-12 | 北京神州泰岳软件股份有限公司 | Return information generates training method, return information generation method and the device of model |
| CN108509519B (en) * | 2018-03-09 | 2021-03-09 | 北京邮电大学 | General knowledge graph enhanced question-answer interaction system and method based on deep learning |
| CN108509519A (en) * | 2018-03-09 | 2018-09-07 | 北京邮电大学 | World knowledge collection of illustrative plates enhancing question and answer interactive system based on deep learning and method |
| CN110489532B (en) * | 2018-05-14 | 2023-11-17 | 西门子股份公司 | Computer-implemented method for providing an adaptive session system and adaptive session system |
| CN110489532A (en) * | 2018-05-14 | 2019-11-22 | 西门子股份公司 | The computer implemented method and self adaptive conversation system of self adaptive conversation system are provided |
| CN108804611A (en) * | 2018-05-30 | 2018-11-13 | 浙江大学 | A kind of dialogue reply generation method and system based on self comment Sequence Learning |
| CN109002500A (en) * | 2018-06-29 | 2018-12-14 | 北京百度网讯科技有限公司 | Talk with generation method, device, equipment and computer-readable medium |
| CN109241256A (en) * | 2018-08-20 | 2019-01-18 | 百度在线网络技术（北京）有限公司 | Dialog process method, apparatus, computer equipment and readable storage medium storing program for executing |
| CN110874402A (en) * | 2018-08-29 | 2020-03-10 | 北京三星通信技术研究有限公司 | Reply generation method, device and computer readable medium based on personalized information |
| CN109547320A (en) * | 2018-09-29 | 2019-03-29 | 阿里巴巴集团控股有限公司 | Social contact method, device and equipment |
| CN109508457A (en) * | 2018-10-31 | 2019-03-22 | 浙江大学 | A kind of transfer learning method reading series model based on machine |
| CN109361823A (en) * | 2018-11-01 | 2019-02-19 | 深圳市号互联科技有限公司 | A kind of intelligent interaction mode that voice is mutually converted with text |
| CN109657127B (en) * | 2018-12-17 | 2021-04-20 | 北京百度网讯科技有限公司 | Answer obtaining method, device, server and storage medium |
| CN109657127A (en) * | 2018-12-17 | 2019-04-19 | 北京百度网讯科技有限公司 | A kind of answer acquisition methods, device, server and storage medium |
| CN109977207A (en) * | 2019-03-21 | 2019-07-05 | 网易（杭州）网络有限公司 | Talk with generation method, dialogue generating means, electronic equipment and storage medium |
| CN110222164A (en) * | 2019-06-13 | 2019-09-10 | 腾讯科技（深圳）有限公司 | A kind of Question-Answering Model training method, problem sentence processing method, device and storage medium |
| CN110222164B (en) * | 2019-06-13 | 2022-11-29 | 腾讯科技（深圳）有限公司 | Question-answer model training method, question and sentence processing device and storage medium |
| CN111090664A (en) * | 2019-07-18 | 2020-05-01 | 重庆大学 | High-imitation person multi-mode dialogue method based on neural network |
| CN110728356A (en) * | 2019-09-17 | 2020-01-24 | 阿里巴巴集团控股有限公司 | Dialogue method and system based on recurrent neural network and electronic equipment |
| CN110728356B (en) * | 2019-09-17 | 2023-08-04 | 创新先进技术有限公司 | Dialogue method and system based on cyclic neural network and electronic equipment |
| CN113032540A (en) * | 2021-03-19 | 2021-06-25 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment and storage medium |
| CN113032540B (en) * | 2021-03-19 | 2023-06-23 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment and storage medium |
| CN113806509A (en) * | 2021-09-18 | 2021-12-17 | 上海景吾智能科技有限公司 | Robot dialogue system and method based on knowledge graph and RNN neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105787560B (en) | 2018-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105787560A (en) | Dialogue data interaction processing method and device based on recurrent neural network | |
| CN106295792B (en) | Dialogue data interaction processing method and device based on multi-model output | |
| KR102213476B1 (en) | Learning contents recommendation system based on artificial intelligence learning and operation method thereof | |
| Rothe et al. | Question asking as program generation | |
| CN110059169B (en) | Intelligent robot chat context implementation method and system based on corpus labeling | |
| Ammanabrolu et al. | Learning knowledge graph-based world models of textual environments | |
| CN110851575B (en) | Dialogue generating system and dialogue realizing method | |
| CN106875940A (en) | A kind of Machine self-learning based on neutral net builds knowledge mapping training method | |
| CN107451230A (en) | A kind of answering method and question answering system | |
| CN113408284A (en) | Training method and device of text processing model, electronic equipment and storage medium | |
| CN112530218A (en) | Many-to-one accompanying intelligent teaching system and teaching method | |
| CN115964459B (en) | Multi-hop reasoning question-answering method and system based on food safety cognition spectrum | |
| Shastri et al. | Semantic networks and neural nets | |
| CN114817564A (en) | Attribute extraction method and device and storage medium | |
| Yang et al. | PsyCoT: Psychological Questionnaire as Powerful Chain-of-Thought for Personality Detection | |
| CN116956116A (en) | Text processing method and device, storage medium and electronic equipment | |
| CN115455985A (en) | Natural language system processing method based on machine reading understanding | |
| CN114048301B (en) | Satisfaction-based user simulation method and system | |
| Aggarwal et al. | Can I say, now machines can think? | |
| Steinert-Threlkeld | Communication and computation: New questions about compositionality | |
| Nishimoto et al. | Dialogue management with deep reinforcement learning: Balancing exploration and exploitation | |
| Bunt | Towards a dynamic interopretation theory of utterances in dialogue | |
| CN112487761A (en) | Question generating method and device based on chart sign fusion | |
| Ishizaki et al. | Exploring mixed-initiative dialogue using computer dialogue simulation | |
| CN116882450B (en) | Question-answering model editing method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |