CN109857848A - Interaction content generation method, device, computer equipment and storage medium - Google Patents
Interaction content generation method, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN109857848A CN109857848A CN201910047156.8A CN201910047156A CN109857848A CN 109857848 A CN109857848 A CN 109857848A CN 201910047156 A CN201910047156 A CN 201910047156A CN 109857848 A CN109857848 A CN 109857848A
- Authority
- CN
- China
- Prior art keywords
- information
- current
- user
- target component
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000003993 interaction Effects 0.000 title claims abstract description 65
- 238000003860 storage Methods 0.000 title claims abstract description 18
- 230000013016 learning Effects 0.000 claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims description 41
- 230000009466 transformation Effects 0.000 claims description 28
- 230000015654 memory Effects 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 13
- 230000006855 networking Effects 0.000 claims description 12
- 230000005540 biological transmission Effects 0.000 claims description 8
- 238000002224 dissection Methods 0.000 claims description 6
- 241000208340 Araliaceae Species 0.000 claims description 5
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 5
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 5
- 235000008434 ginseng Nutrition 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 description 18
- 230000033001 locomotion Effects 0.000 description 12
- 230000006870 function Effects 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000002360 preparation method Methods 0.000 description 7
- 238000003058 natural language processing Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 238000004904 shortening Methods 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 206010052804 Drug tolerance Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000026781 habituation Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000012958 reprocessing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- 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
- G06F16/35—Clustering; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Information Transfer Between Computers (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of interaction content generation method, device, computer equipment and storage mediums, wherein, the interaction content generation method includes: to receive the current round information for the carrying session identification that client is sent, and is based at least one current intent parameter of current round acquisition of information；Obtain at least one history intent parameter corresponding with session identification；At least one current intent parameter and at least one history intent parameter are analyzed using preset intensified learning model, obtain target intention；Corresponding target intention template is obtained based on target intention；Based on each target component query and search text database, retrieval text corresponding with each target component is obtained；Current return information corresponding with each retrieval text is obtained, is pushed to client according to parameter priority sequence.This method can ensure that chat robots generate accurate return information in time and reply to client.
Description
Technical field
The present invention relates to semantic analysis technology fields more particularly to a kind of interaction content generation method, device, computer to set
Standby and storage medium.
Background technique
With the continuous development of science and technology, the introducing of information technology, computer technology and artificial intelligence technology, machine
Industrial circle is gradually walked out in the research of people, gradually extends to the neck such as medical treatment, health care, family, amusement and service industry
Domain.And requirement of the people for robot also conform to the principle of simplicity single duplicate mechanical action be promoted to have anthropomorphic question and answer, independence and with
The intelligent robot that other robot interacts, human-computer interaction also just become an important factor for determining intelligent robot development.
Current chat robots, which are largely laid particular emphasis on, carries out single-wheel interaction with user, so that robot can not be well
The true intention of user is obtained, because the important information of user's chat may be in dialogue before user.It is this to be based on single-wheel
Chat mechanism, have ignored the chat theme and scene analysis of the former wheels of active user, therefore its reply content returned may
There are deviation even mistakes.How to obtain becoming urgent problem to be solved closer to the chat content of user's true intention.
Summary of the invention
The embodiment of the present invention provides a kind of interaction content generation method, device, computer equipment and storage medium, to solve
The problem of how obtaining the chat content closer to user's true intention.
A kind of interaction content generation method, comprising:
The current round information for receiving the carrying session identification that client is sent, is based on current round acquisition of information at least one
A current intent parameter；
Dialogue-based mark inquires conversation recording database, obtains at least one history corresponding with session identification and is intended to
Parameter；
Using preset intensified learning model at least one current intent parameter and at least one history intent parameter into
Row analysis, obtains target intention, and target intention includes at least one target component and parameter priority sequence；
Based on each target component query and search text database, retrieval text corresponding with each target component is obtained
This；
Recalls information transformation model converts each retrieval text, obtains corresponding with each retrieval text current
At least one current return information is pushed to client according to parameter priority sequence by return information.
A kind of interaction content generating means, comprising:
Current information module is received, the current round information of the carrying session identification for receiving client transmission is based on
At least one current intent parameter of current round acquisition of information；
History parameters module is obtained, conversation recording database is inquired for dialogue-based mark, obtains and session identification phase
At least one corresponding history intent parameter；
Target intention module is obtained, for using preset intensified learning model at least one current intent parameter and extremely
A few history intent parameter is analyzed, and obtains target intention, target intention includes that at least one target component and parameter are excellent
First grade sequence；
Retrieval text module is obtained, for being based on each target component query and search text database, is obtained and each mesh
Mark the corresponding retrieval text of parameter；
Return information module is obtained, each retrieval text is converted for recalls information transformation model, is obtained and every
At least one current return information, is pushed to by the corresponding current return information of one retrieval text according to parameter priority sequence
Client.
A kind of computer equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, the processor realize above-mentioned interaction content generation method when executing the computer program.
A kind of computer readable storage medium, the computer-readable recording medium storage have computer program, the meter
Calculation machine program realizes above-mentioned interaction content generation method when being executed by processor.
Above-mentioned interaction content generation method, device, computer equipment and storage medium, by receiving working as client transmission
Preceding round information, analysis obtain at least one current intent parameter, and current intent parameter combination conversation recording database is obtained
Target intention and corresponding return information are obtained after at least one history intent parameter, can ensure that chat robots generate standard in time
True return information replies to client, avoids only relying on current intent parameter acquisition inaccuracy or even unrelated return information,
Promote the interaction relevance and reliability between chat robots and client.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by institute in the description to the embodiment of the present invention
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the application environment schematic diagram of interaction content generation method in one embodiment of the invention；
Fig. 2 is the flow chart of interaction content generation method in one embodiment of the invention；
Fig. 3 is the realization process schematic that target intention is obtained in one embodiment of the invention；
Fig. 4 is another flow chart of interaction content generation method in one embodiment of the invention；
Fig. 5 is another flow chart of interaction content generation method in one embodiment of the invention；
Fig. 6 is another flow chart of interaction content generation method in one embodiment of the invention；
Fig. 7 is another flow chart of interaction content generation method in one embodiment of the invention；
Fig. 8 is the schematic diagram of interaction content generating means in one embodiment of the invention；
Fig. 9 is the schematic diagram of computer equipment in one embodiment of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
Interaction content generation method provided in an embodiment of the present invention, can be applicable in the application environment such as Fig. 1, in the interaction
Hold generation method to apply in interaction content generation system, it includes client and server which, which generates system, wherein
Client is communicated by network with server.Client is also known as user terminal, refers to corresponding with server, mentions for client
For the program of local service.The client it is mountable but be not limited to various personal computers, laptop, smart phone,
In the computer equipments such as tablet computer and portable wearable device.Server can use independent server either multiple clothes
The server cluster of business device composition realizes that the current round information that sends for receiving user by client is generated and chatted
The corresponding return information of robot.
In one embodiment, it as shown in Fig. 2, providing a kind of interaction content generation method, applies in Fig. 1 in this way
It is illustrated, includes the following steps: for server
S10. the current round information for the carrying session identification that client is sent is received, extremely based on current round acquisition of information
A few current intent parameter.
Wherein, current round information is the information to express user's intention of the current round input client of user.Into
One step, current round information may include the intent information expressed in a variety of forms, including but not limited to lteral data, voice
Information or gesture motion information etc., are not construed as limiting herein.
Session identification is the mark to the different sessions scene of Differentiated services device starting.Because server can be with several clients
End starts several session contexts, and in order to distinguish each session context with the content analysis that conversates, server is needed to per for a moment
Talk about the corresponding session identification of scene setting.
It is intended that the purpose that user expresses in current round information, " explicit to be intended to " and " implicit to be intended to " can be divided into, it
Difference between the two it is fairly obvious:
(1) explicit to be intended to:
So-called explicit intention, that is, clearly occur a kind of vocabulary of expression intention in the language of user, such as:
" it is desirable that ", " desired ", " needs " etc..The judgement difficulty that this display is intended to for chat robots is lower, it is only necessary to
It identifies these intention vocabulary fixed, then is contacted with the other compositions in sentence.For example " I wants to make a reservation for user's input
To Pekinese's air ticket ", chat robots, which can identify, is intended to vocabulary " thinking ", so that it is predetermined for me to obtain the intent parameter of this
Air ticket, air ticket stroke include from current location to Beijing.
(2) implicit to be intended to:
It is implicit to be intended to be intended to need to chat on the contrary, the language of user does not occur directly reflecting the vocabulary being intended to explicit
Robot judges that user is intended to according to lteral data.Implicitly it is intended to compare for chat robots to be difficult to judge, at present
It is that first implicit intention is converted to explicitly to be intended to reprocessing with a kind of more method is obtained.For example user is expression " I is hungry "
When this implicit intention, it is first converted to corresponding " I wants to eat " this explicit diagram form, then allows chat robots again
It is handled according to explicit intention.
Current intent parameter is after current round information is converted to corresponding lteral data, to extract from the lteral data
Containing the verb and noun (removing stop words) being of practical significance, the customer objective expressed in most succinct mode.Wherein, stop words
It is main including English character, number, mathematical character, punctuation mark and the high function word of frequency of use etc..For example, lteral data is
" today, weather was pretty good ", the current intent parameter extracted from the lteral data include: today, weather and good (removal
Stop words "true" and " ").
In step S10, server passes through extractable at least one current intention out of current round information that client is sent
Parameter filters out the function word without practical significance in current round information, can directly acquire and effectively work as in current round information
Preceding intent parameter obtains the true intention preparation techniques basis of user for subsequent combination history intent parameter.
S20. dialogue-based mark inquires conversation recording database, obtains at least one history corresponding with session identification
Intent parameter.
Wherein, conversation recording database is the meeting that server is based on that each session context (namely each session identification) saves
Talk about set of records ends.
History intent parameter be it is corresponding with current intent parameter, based on same session identification current round session it
The intent parameter of all rounds of preceding preservation.
Specifically, server is based on the current round dialogue of same session identification corresponding record all in conversation recording database
The current intent parameter generated.It is formed it is to be appreciated that each current intent parameter is stored in conversation recording database
History intent parameter, in case subsequent server can be matched corresponding all history intent parameters based on same session identification and be carried out
True intention analysis.
In step S20, server can dialogue-based mark corresponding all go through is directly acquired in conversation recording database
History intent parameter extracts the history intent parameter in all historical sessions without server again, accelerates server analysis user
The processing speed of true intention.
S30. ginseng is intended to at least one current intent parameter and at least one history using preset intensified learning model
Number is analyzed, and target intention is obtained, and target intention includes at least one target component and parameter priority sequence.
Wherein, intensified learning model defaults in server, will input all history intent parameters of the model
It is analyzed with current intent parameter, obtains model of the analysis result as target intention.In the present embodiment, server be can be used
The combination of LSTM (Long Short-Term Memory, shot and long term memory network) model and softmax classifier is as reinforcing
Learning model.
Multi-object can be set for chat scenario in the present embodiment to be intended to, for example primary target is intended to include chatting and mentioning
It asks；Primary target, which is intended to can to continue to divide multiple Secondary objectives again, to be intended to, for primary target be intended in chat be intended to can be after
It is continuous to be divided into life, work and leisure；It is intended to based on Secondary objective, can continues to be divided into three-level target intention, according to scene need
Continue refinement etc..
Specifically, as shown in figure 3, server is as follows by the realization process that lteral data obtains target intention:
1. lteral data is pre-processed, the punctuation mark including removing corpus, removal stop words be (no physical meaning
Word) etc..
2. pretreated lteral data is generated term vector using word2vec tool.
Wherein, word2vec tool is a NLP (Natural Language Processing, natural language processing)
Tool, the words vectorization in all natural languages can be switched to the dense vector (Dense that computer is understood that by it
Vector), quantitatively to measure the relationship between word and word, contacting between word and word is excavated.It is to be appreciated that for phase
As word, corresponding term vector is also close.
3. carrying out feature extraction to term vector using LSTM model.
Wherein, LSTM model, which can solve natural language processing statistical method, can only consider nearest n word and ignore more long
The problem of preceding word, it is intended to find the correlativity between word and word, increase time content of text in data analysis, remember it
Before what has occurred, be then applied to neural network, contacting between next thing that observation and neural network are occurred, from
And obtain target intention.
The characteristics of LSTM is exactly to add other than RNN (Recurrent Neural Network, Recognition with Recurrent Neural Network) model
The valve node of each layer is added, as shown in Figure 4.Valve node has 3 classes: forgeing valve (forget gate), inputs valve
(input gate) and output valve (output gate).These valve nodes can open or close, for that will judge RNN
Whether the memory state (state of network before) of model reaches threshold value in the result of a upper output layer to be added to the current layer
Calculating in.
Valve node using sigmoid function will in RNN model the corresponding upper output layer operation knot of the valve node
Fruit calculates as input；The valve node is exported into the work that is multiplied with the calculated result of current layer if calculated result reaches threshold value
For next layer of input；The corresponding upper output layer operation result of the valve node is forgotten if not reaching threshold value.
The memory function of LSTM model is exactly to be realized by these valve nodes.When valve node is opened, before
The training result of RNN model will be associated with current RNN model and be calculated, and before when valve node is closed
Calculated result just no longer influences current calculating.Therefore, history may be implemented by the switch of control valve node to be intended to most
The influence of target intention is obtained eventually.When be not intended to history be intended to after intention analysis have an impact, such as at natural language
Start to analyze new paragraph or new chapters and sections in reason, then turns off valve node.
4. completing intent classifier work using softmax.
Finally, introducing softmax classifier in neural network can after LSTM model is carried out feature extraction to term vector
Define a new output layer (target intention probability distribution).The output layer of LSTM model neural network based is not necessarily one
A probability distribution layer, so softmax classifier can be finally introducing to LSTM model, softmax classifier can be used as one
The output (the last output layer of LSTM model) of neural network is become a probability distribution by a additional process layer, that is, often
One output is all the decimal between 0 to 1, and the sum of result of all outputs is 1.For example, if LSTM model most
Output layer is y1, y2, y3 and y4 afterwards, and server can be used softmax classifier and carry out to the last output layer of the LSTM model
Processing, using following formula:
Y ' can be calculated by softmax classifier1y’2y’3y’4Value, wherein y 'iIt is the defeated of each target intention
Probability out.It is to be appreciated that y 'iThe corresponding target intention of the maximum output of middle numerical value is exactly the corresponding target of current round information
It is intended to.For example, step S10 obtain lteral data " I seen recently Jiang Wen director New cinema evil can never prevail over good ", through in step S20
The target intention that preset intention assessment model obtains after being identified is: user discusses film, wherein cinematographic parameters include:
Film is entitled, and evil can never prevail over good.
Further, parameter priority sequence can be arranged in the present embodiment also for target component, namely to specified after verb
Higher parameter priority sequence is arranged in the corresponding noun of the object of movement, and it is suitable that secondary parameter priority is arranged to movement executor
Sequence.Scene is discussed for majority, people talk about the object for focusing on movement and executing of event, namely act the object executed
Parameter priority sequence is generally greater than movement executor.To different target component setting parameter priority sequences, it is conducive to subsequent
Concept extension is carried out based on different parameter priority sequences.
Alternatively, the also settable value by the parameter priority including the largest number of target intentions of target component of server
Highest.For example, the lteral data that is inputted by client of user is that " inquiry Long Gang rents a house information, and 20 square meter of area is cheap
".Server is that Long Gang rents a house based on the target intention that the lteral data can analyze the lteral data, which is carried out
Split analysis:
1) target component that " inquiry Long Gang rent a house information " includes is that Long Gang rents a house；
2) target component that " inquiry Long Gang rents a house information, 20 square meter of area " includes is that Long Gang rents a house and area；
3) target component that " inquiry Long Gang rents a house information, and 20 square meter of area is cheap " includes be Long Gang rent a house, face
Long-pending and price.
From the above analysis as can be seen that the third group target component that includes is most namely the parameter priority of third group is suitable
Sequence is 3, and so on, second group of parameter priority sequence is 2, and first group of parameter priority sequence is 1.Server is based on
Parameter priority sequence can be searched by the sequence from big to small of parameter priority sequence when being replied, also i.e. by the
Three group pollings to result of renting a house return to client first.
It illustrates in the present embodiment, server obtains the realization process of target component by target intention:
It, can after the identification of step S30 for example, lteral data " I seen recently the New cinema of Jiang Wen director evil can never prevail over good "
Confirm that the corresponding target intention of the lteral data talks stress-relieving activity for movement people.The target intention mould of stress-relieving activity is talked in movement people
The target component that the template can be obtained in plate is " movement people " and " stress-relieving activity ".In lteral data, " I has seen that Jiang Wen is led recently
Evil can never prevail over good for the New cinema drilled " in extract corresponding with above-mentioned target component word: movement people corresponding " I " (namely use
Family), stress-relieving activity " evil can never prevail over good for film ".Wherein, the corresponding parameter priority sequence of movement people is 1 grade, and stress-relieving activity is corresponding
Parameter priority sequence be 2 grades, then it is subsequent target component is replied into client as return information when, Selecting All Parameters are excellent
The maximum target component of value of first grade sequence first replies to client.
In step S30, server can identify the target intention of user by preset intensified learning model, accurately sentence
The demand of disconnected user out, and based on the maximum preference return of the corresponding parameter priority sequence of target component, it is based on conducive to subsequent
The target component obtains corresponding return information, accurately chat is promoted around the focus of user, to keep chat robots
Chat stickiness and practicability between user.
S40. it is based on each target component query and search text database, obtains retrieval corresponding with each target component
Text.
Wherein, networking data library is online retrieving library, for example, Baidu, search dog, Google or 360 search etc..Retrieve text
It is exactly reply text corresponding with target component, for example, target component is " weather today ", server is looked by networking data library
" weather today " corresponding weather lookup can be obtained as a result, the weather lookup result is exactly to retrieve text by asking online weather.
In step S40, server can obtain the corresponding retrieval text of target component by networking data library, provide the user with
Text information relevant to current round information, improves the accuracy of reply content.
S50. recalls information transformation model converts each retrieval text, obtains corresponding with each retrieval text
At least one current return information is pushed to client according to parameter priority sequence by current return information.
Specifically, the information transformation model in this step is to convert text information to the return information for passing to user
The model of form.For example, user inputs current round information with speech form, then this step will be retrieved using voice transformation model
Text is converted into voice messaging and exports to user, or can also be arranged by the hobby of client directly directly defeated in the form of text
Out to client.
Information transformation model is the model for mutually converting the meaning of text data expression and particular expression mode, than
Such as, corresponding lteral data is converted by voice messaging, converts corresponding sign language movement etc. for lteral data.In this implementation
Example, information transformation model can be converted based on a variety of expression ways, therefore it includes multiple transformation models.For example, by voice
Information is converted into the voice transformation model of text information, such as RNN-HMM (Recurrent Neural Network-Hidden
Markov Model, that is, Recognition with Recurrent Neural Network-Hidden Markov) model or LSTM-HMM (Long Short-Term Memory,
Shot and long term memory network-Hidden Markov) the machines acoustic model such as model, or convert gesture information in the hand of text information
Gesture transformation model, such as FLDCRFs (Fuzzy based Latent dynamic Condition Random Fields, mould
Paste hidden dynamic condition random field) etc. machines gesture identification model.It should be understood that, RNN-HMM machine acoustic mode in the present embodiment
The machine learning models such as type, LSTM-HMM machine acoustic model and FLDCRFs machine gesture identification model are well-known technique, herein
It repeats no more.
Illustrate the realization process for retrieval text being converted in step S60 current return information, in step slo,
If user passes through the current round information (voice messaging) of microphone input of client: " I has seen the new electricity of Jiang Wen director recently
Evil can never prevail over good for shadow ".Server obtains the current round information of the phonetic matrix of microphone acquisition by client, therefore can determine
Current round information is voice messaging, that is, RNN-HMM machine acoustic model can be used and convert the voice messaging to that " I sees recently
Evil can never prevail over good for the New cinema of Jiang Wen director " this lteral data.In step S60, in order to keep one with user's exchange way
Cause property, improves chat interest, and the retrieval text that step S50 is obtained can equally be used RNN-HMM machine sound by chat computer
Model is learned, voice messaging (current return information) is converted by retrieval text and returns to client.
In step S50, server can will retrieve the corresponding current return information of text according to parameter priority sequence and return
It to client, is directly replied conducive to the focus based on user, return to client accurately can currently return in time for reference
Complex information, to improve the relevance of the chat content between chat robots and user；Server can need to be arranged according to scene
Form for the return information of output is consistent with the mode of current round information that user inputs, and keeps chat habit consistency,
Improve the chat interest of user and chat robots.
In interaction content generation method provided in this embodiment, server is by receiving the current round letter that client is sent
Breath, analysis obtain at least one current intent parameter, current intent parameter combination conversation recording database are obtained at least one
Target intention and corresponding return information are obtained after history intent parameter, can ensure that chat robots generate accurate reply in time
Information-reply avoids only relying on current intent parameter acquisition inaccuracy or even unrelated return information, promotes chat to client
Interaction relevance and reliability between robot and client.
In a specific embodiment, as shown in figure 4, in step S10, i.e., the current round that reception client is sent is believed
Breath, comprising:
S11. the current round information that client is sent is received, recalls information transformation model knows current round information
Not, lteral data is obtained.
Wherein, lteral data is exactly that the current round information sent in a specific way corresponds to the text information of meaning, for example,
User sends nodding action as current round information by the camera of client, and server can convert mould by recalls information
Type (in the present embodiment be action recognition model) by nodding action correspondence be converted into lteral data " agreement ".
In step S11, server, which passes through, is converted to lteral data for the current round information that the client received is sent,
It is subsequent based on lteral data is further processed conducive to server, to obtain return information corresponding with current round information
Return to client.
S12. dissection process is carried out to lteral data using preset language processing model, obtains at least one current intention ginseng
Number.
Wherein, word2vec, a NLP (Natural can be used in the preset language processing model of the present embodiment
Language Processing, natural language processing) tool, it can turn the words vectorization in all natural languages
For computer it is understood that dense vector (Dense Vector) excavate quantitatively measuring the relationship between word and word
Contacting between word and word.It is to be appreciated that corresponding term vector is also close for similar word.
Dissection process is carried out to lteral data using preset language processing model in this step, obtains at least one current meaning
The realization process of graph parameter has been described in detail in step slo, and in order to avoid repeating, details are not described herein again.
In step S12, server can be used preset language processing model analyzing lteral data and obtain at least one current meaning
Graph parameter, the extractable most succinct current intention that user expresses in current round out are conducive to subsequent server and are based on this currently
It is intended to be combined with the true intention that all history intent parameters obtain user.
For step S11 into S12, server, which passes through, is converted to text for the current round information that the client received is sent
Data, it is subsequent based on lteral data is further processed conducive to server, to obtain and current round information corresponding time
Complex information returns to client.Server can be used preset language processing model analyzing lteral data and obtain at least one current meaning
Graph parameter, the extractable most succinct current intention that user expresses in current round out are conducive to subsequent server and are based on this currently
It is intended to be combined with the true intention that all history intent parameters obtain user.
In a specific embodiment, as shown in figure 5, in step S30, i.e., using preset intensified learning model at least
One current intent parameter and at least one history intent parameter are analyzed, and target intention is obtained, comprising:
S31. based at least one current intent parameter and at least one history intent parameter, current term vector is obtained respectively
Matrix and history term vector matrix.
Wherein, term vector is that word is mapped to a semantic space in fact, obtained Semantic mapping matrix.For example,
Centre word A maps out periphery word BCDAEFG, obtains a parameter matrix W1, centre word L is then mapped to periphery word
BCDLEFG obtains parameter matrix W2.If W2 and W1 are closely located, illustrate that A and L can be mapped to same periphery word, it may be possible to
Near synonym.
The pre-treatment step of term vector training:
1. the text (current intent parameter or history intent parameter) of pair input generates a vocabulary, count in vocabulary
The word frequency of each word sorts from high to low according to word frequency, takes V most frequent word, constitutes a vocabulary.There are one for each word
A dimension is the one-hot vector of V.Wherein, one-hot vector is associated in dictionary to characterize each element in vector
A word, the vector of specified word is expressed as: its corresponding element in vector is set as 1, and other elements are set as 0
If (word occurred in vocabulary, and corresponding position is 1 in vocabulary in vector, 0) other positions are all.If word
Do not occur in remittance table, then vector is full 0.
2. each word for inputting text is generated an one-hot vector, retain the home position of each word, because being
It is context-sensitive.
3. determining the dimension N of term vector.
Specifically, the present embodiment can be used CBOW model (continuous bag of words, using the context of word as input, come
Predict this word itself) term vector is obtained, realize that steps are as follows:
1. determining window size window, 2*window training sample, (i-window, i), (i- are generated to each word
Window+1, i) ..., (i+window-1, i), (i+window, i).
2. determining batch_size, the big palpulus of batch_size is the integral multiple of 2*window, to ensure each batch
Contain the corresponding all samples of a vocabulary.
3. there are two types of training algorithms: level Softmax and Negative Sampling.
4. the certain number of neural network repetitive exercise obtains parameter matrix of the input layer to the hidden layer that dimension is N, matrix
In every a line transposition be equivalent dimension be V term vector matrix.
In step S31, server can obtain current intent parameter and the corresponding current term vector of history intent parameter respectively
Matrix and history term vector matrix are that the word-based vector matrix of subsequent server obtains target intention preparation techniques basis.
S32. current word vector matrix and history term vector matrix are analyzed using preset intensified learning model, is obtained
Take target intention.
Specifically, LSTM (Long Short-Term Memory, the shot and long term note in step S30 can be used in this step
Recall network) combination of model and softmax classifier is as intensified learning model, to current word vector matrix and history term vector
Matrix carries out analysis and carries out dissection process, obtains target intention.
Input valve by current word vector matrix and history term vector Input matrix to preset intensified learning model, tool
The realization process of body is described in detail in above mentioned step S3 0, and in order to avoid repeating, details are not described herein again.
In step S32, server can identify the target intention of user by preset intensified learning model, accurately sentence
The demand of disconnected user out, obtains the chat content being more bonded with user demand, reduces and answers non-institute between chat robots and user
The number asked.
For step S31 into S32, server can obtain current intent parameter and the corresponding current word of history intent parameter respectively
Vector matrix and history term vector matrix are that the word-based vector matrix of subsequent server obtains target intention preparation techniques basis.
Server can identify the target intention of user by preset intensified learning model, accurately judge the demand of user, obtain
The chat content being more bonded with user demand is taken, the number given an irrelevant answer between chat robots and user is reduced.
In a specific embodiment, as shown in fig. 6, in step 40, networking data library is inquired based on each target component,
Retrieval text corresponding with each target component is obtained, is specifically comprised the following steps:
If S41. the attribute of target component is product attribute, inquired based on each target component corresponding with product attribute
Product database, obtain corresponding with each target component Products Show information as retrieving text.
Wherein, product attribute is the transaction attribute that content supplier is marked by server to each target component.Wherein,
Content supplier is to provide the provider of server where chat robots, the provider can sell simultaneously other products in kind or
Various services are provided.
It is to be appreciated that when this chat robots be applied to content supplier marketing scene, chat robots can with
The marketing product or service that the appropriate opportunity that family is linked up provides content supplier carry out promotion, are conducive to using the chat robots
Product or service sales conversion ratio improve in the content supplier of system.For example, target component is " insurance ", if insurance belongs to content
The free product of provider, content supplier can be labeled as product attribute to " insurance ".
Specifically, server can by each target intention template object name or service name and content supplier from
The material object or service type that body provides are associated, and the target component to get in server includes object name therein
Or when service name, content supplier can release corresponding product to user according to incidence relation or service is introduced.
Product database is the database for saving the available all products of content supplier or service.Products Show information
It is the product content introduction in product database to each product or service log.
Preferably, Products Show information may include that product brief and product are discussed in detail, and push away first to server
Give client production brief.When request is discussed in detail in the product that server receives client transmission, show user
It has a mind to when continuing to understand the details of product, can be further continued for being discussed in detail to the corresponding product of client transmission product.
In step S41, whether server can belong to the label of product attribute to each target component mark in advance, be conducive to clothes
Business device gets target component can be judged at once, no longer need to carry out analyzing whether it is product attribute to each target component,
Conducive to the judgement time for shortening server.Server can obtain the corresponding Products Show letter of target component based on product database
Breath is pushed to user to subsequent, promotes the practical applicability and chat content scalability for using this chat robots.
Preferably, after step S41, that is, Products Show information corresponding with each target component is being obtained as inspection
After Suo Wenben, interaction content generation method further include: Products Show information is pushed to client by pre-set product push format
End.
Wherein, to be that server is pre-set be pushed to client in chat interface for product to pre-set product push format
Format, for example, server can set push product list of file names or each product corresponding product is simply introduced.
Further, it when server receives the ProductName that client is chosen, can will be established in the ProductName and step S51
Product association is discussed in detail, product is discussed in detail and is pushed to client again, avoids being pushed to the use of client for the first time
The uninterested product in family is discussed in detail.
In this step, server can push format by pre-set product and Products Show information is sent to client, keep chatting
Its content is related to format consistency when product or service.
If S42. the attribute of target component is not product attribute, networking data library is inquired based on each target component, is obtained
Retrieval text corresponding with each target component.
Wherein, networking data library is online retrieving library, for example, Baidu, search dog, Google or 360 search etc..Retrieve text
It is exactly reply text corresponding with target component, for example, target component is " weather today ", server is looked by networking data library
" weather today " corresponding weather lookup can be obtained as a result, the weather lookup result is exactly to retrieve text by asking online weather.
In step S42, server can obtain the corresponding retrieval text of target component by networking data library, provide the user with
Retrieval text relevant to current round information, improves the accuracy of reply content；Meanwhile server be not necessarily to locally save with
The corresponding retrieval text of target component, can save the memory space of server local.
For step S41 into S42, whether server can belong to the label of product attribute to each target component mark in advance,
Getting target component conducive to server can be judged, no longer need to carry out analyzing whether it is product category to each target component
Property, conducive to the judgement time for shortening server.Server can obtain the corresponding Products Show of target component based on product database
Information is pushed to user to subsequent, promotes the practical applicability and chat content scalability for using this chat robots.Clothes
Business device can obtain the corresponding retrieval text of target component by networking data library, provide the user with relevant to current round information
Text is retrieved, the accuracy of reply content is improved；Meanwhile server is not necessarily to locally saving retrieval text corresponding with target component
This, can save the memory space of server local.
In a specific embodiment, as shown in fig. 7, before step S50, i.e., in the recalls information transformation model pair
Before each retrieval text is converted, the interaction content is generated further include:
S501. userspersonal information is obtained, user's portrait is generated based on userspersonal information.
Wherein, userspersonal information is the static data and dynamic data that server is collected, wherein static data is user
The personal information that will not change in the long-time being actively entered in server registration, for example, the region of user, the age, gender,
Culture, occupation and income etc.；Dynamic data is the number that server and user interaction in the process analyze user behavior
According to for example, living habit or consumption habit etc..Generally speaking, need to use following individual subscriber letter to obtain user's portrait
Breath:
(1) ascribed characteristics of population: including essential informations such as gender, ages；
(2) interest characteristics: browsing content, is read and seeks advice from, buys article preference etc. at collection content；
(3) consumption feature: feature relevant to consumption；
(4) position feature: city locating for user, locating residential area, user's motion track etc.；
(5) device attribute: the terminal feature etc. used；
(6) behavioral data: user behaviors log data of the users such as access time, browse path in website；
(7) social data: user social contact related data.
User's portrait is that the highly refined label that server is abstracted according to userspersonal information namely user are special
Sign mark.User is described by can use high level overview, readily comprehensible feature to user " patch ", being conducive to server
It is further processed according to label (standardized information).
Specifically, the realization process based on userspersonal information's generation user's portrait is as follows:
One, abstract classification is carried out to a certain feature in userspersonal information and summary forms label, the label of the label
Value has can classification.
Example: " male ", " female " this category feature in userspersonal information are subjected to abstract, are referred to as " gender ", " gender "
That is a label；
Two, exhaustion label values (Tag Value), so that the label includes the value of corresponding all possible situation.
Example: for label " gender ", label value can be divided into " male ", " female " and " unknown "；
For label " age ", label value can be divided into " 0-18 ", " 18-35 ", " 35-60 ", " 60-100 " etc..
Three, construct user and draw a portrait (User Profile).According to the label that step 1 and step 2 are created, user is extracted
Label value corresponding with each label in people's information.
Example: the label that user's portrait includes has gender, age, mobile phone brand, residence and hobby etc..Xiao Ming is user's picture
One example of picture, the output result of user's portrait of Xiao Ming are as follows: " male ", " 18-35 ", " iPhone ", " Beijing " and " football ".
In step S501, server can construct user's portrait based on userspersonal information, be based on the use to subsequent server
User's personality preparation techniques basis of family portrait analysis user.
S502. analysis user portrait, obtains user's personality corresponding with user's portrait.
Wherein, user's personality is stable attitude of the people to reality, and with this attitude correspondingly, habituation
Behavior in the personality characteristics that shows.Personality is just more stable once being formed, but not unalterable, but can
Plasticity.Personality is different from makings, more embodies the social property of personality, the core of the personality difference between individual is personality
Difference.
Specifically, by the way that user's portrait is analyzed and excavated, the heart demand and user's property of user can also be disclosed
The inherent natures such as lattice.For example, related shopping label includes be averaged browsing time and the similar number of mean comparisons in user's portrait, three
The shopping online label record situation of a user is as follows:
| Average browsing time (minute) | Mean comparisons' number | |
| User A | 7 | 3 |
| User B | 15 | 12 |
| User C | 25 | 20 |
It does shopping personality reference template (being generated based on actual conditions statistics):
Impulsive style: the average browsing time is less than 10 minutes, and mean comparisons' number is less than 5 times.
Rational type: the average browsing time, mean comparisons' number was between 5 times to 20 times between 10 minutes to 20 minutes.
Hesitation type: the average browsing time is greater than 20 minutes, and mean comparisons' number is greater than 20 times.
Shopping label of the server based on three users and shopping personality reference template, which compare, can obtain following result:
User A is always in short time (the average browsing time is less than 10 minutes) internal ratio less amount of commodity (mean comparisons' number
Less than 5 times) it places an order, then the shopping personality of user A is impulsive style.
Always (the average browsing time was between 10 minutes to 20 minutes) compare a small amount of similar commodity to user B in amount of time
It places an order after (mean comparisons' number is between 5 times to 20 times), then the shopping personality of user B is rational type.
Always for a long time, (the average browsing time is greater than 20 minutes) browses many commodity (mean comparisons time to user C in large quantities
Number be greater than 20 times) after place an order, then the shopping personality of user C be hesitation type.
Similarly, server can also draw a portrait to user and carry out the chat personality that analysis obtains user, and chat personality includes: emerging
It puts forth energy, is tranquil or droning etc..You need to add is that server can also pass through the sound in the current round information of analysis user's input
Speciality judges user's personality in conjunction with chat personality reference template.Wherein, the chat personality reference template of sound characteristic is (based on real
Situation statistics in border generates) as follows:
1) sound of doping breathing and fragility: if user is male, belong to young artistical type；If user is
Women, though personality is partially womanlike, and integrates beautiful, small and exquisite, optimistic, this kind of women is easier excitement.
2) powerless sound: if user is male, without specific personality；If user is women, there is sociability
By force, the characteristics such as perception, humour.
3) flat sound: user no matter men and women, character trait be it is partially manlike, the state of mind is not good enough, it is cold, shrink
Before not, current state is less positive.
4) the low and rough sound of tone: if user is male, it should possess the power that quicks observation, and real, round and smooth,
It is mature, capable and experienced, adaptable；If user is women, Sensate focus exercise, which is mainly shown as, likes lazy or sick disgruntled etc..
5) straightforward sound: if user is male, personality is capable and experienced and self-esteem is strong；If user is women, personality
Vivaciously, it is good at that social, self-esteem is strong, lacks in a sense of humour.
6) word speed is fast: user no matter men and women, the personality of such crowd is relatively active, and sociability is strong.
7) modulation in tone and resonant voice: user no matter men and women, such crowd is full of vitality.
In step S502, server can analyze based on practical experience the user's portrait obtained in step S601, obtain
The corresponding user's personality of user is obtained, different reply mode preparation techniques are used based on different user's personality for subsequent server
Basis.
S503. preset reply pattern base is inquired based on user's personality, obtains target retro mode corresponding with user's personality.
Wherein, it replys pattern base and is preset in server, for the modal sets of the return information of different user's personality
At database.
Target retro mode is the Chat mode replied for user's personality, and the present embodiment is mainly used in voice and chats
It, therefore the tone replied when mode is chat robots and user's chat replys mode.
Specifically, the synchronous both sides of the tone, are more easier to be received by other side when being chatted according to psychological study.That is
If user speaks, the tone is leisurely, if chat robots follow the rhythm of user to chat, can increase user and chat machine
People continues the wish of chat.Based on this, server can be according to the corresponding chat tone of user's personality collocation, to further enhance use
Chat interest and affective interaction between family and chat robots.
Further, server can will configure a kind of tone to each user's personality and reply mode, for example, for enthusiasm,
It is that the quick and enthusiastic tone replys mode that the tone of users' personality such as excited or positive, which replys mode,；For calm, flat or
The tone of mature user's personality replys the tone reply mode etc. that mode is middling speed and calmness.
In step S503, server can match corresponding target retro mould in preset reply pattern base according to user's personality
Formula, to further enhance chat interest and the affective interaction between user and chat robots.
S504. target retro mode is added to information transformation model, with more new information transformation model.
Specifically, the target retro mode (mode can be replied for the tone) that server obtains step S603, such as fastly
The tone reply mode etc. of speed and enthusiasm is sent to information transformation model so that information transformation model by concept text conversion be with
The current return information that target retro mode matches.Further, information transformation model by adjusting word speed and can speak
Tone realizes the current return information for obtaining and matching with target retro mode.
In step S504, server obtains the information updated in combination with target retro mode and information transformation model and converts mould
Corresponding return information is sent to client, enhances chatting machine by type to adjust the tone to match with target retro mode
Affective interaction between device people and user improves the validity of chat content.
Into S504, server can based on practical experience analyze user's portrait step S501, and it is corresponding to obtain user
User's personality, and corresponding target retro mode is matched in preset reply pattern base according to user's personality, further to increase
Chat interest and affective interaction between strong user and chat robots.Server can by after adjustment with target retro mode
The current return information to match is sent to client, enhances the affective interaction between chat robots and user, and raising is chatted
The validity of its content.
In interaction content generation method provided in this embodiment, server is by receiving the current round letter that client is sent
Breath, analysis obtain at least one current intent parameter, current intent parameter combination conversation recording database are obtained at least one
Target intention and corresponding return information are obtained after history intent parameter, can ensure that chat robots generate accurate reply in time
Information-reply avoids only relying on current intent parameter acquisition inaccuracy or even unrelated return information, promotes chat to client
Interaction relevance and reliability between robot and client.
Further, server, which passes through, is converted to lteral data for the current round information that the client received is sent,
It is subsequent based on lteral data is further processed conducive to server, to obtain return information corresponding with current round information
Return to client.Server can be used preset language processing model analyzing lteral data and obtain at least one current intention ginseng
Number, the extractable most succinct current intention that user expresses in current round out are conducive to subsequent server and are based on the current intention
The true intention of user is obtained in conjunction with all history intent parameters.
Further, server can obtain current intent parameter and the corresponding current word moment of a vector of history intent parameter respectively
Battle array and history term vector matrix are that the word-based vector matrix of subsequent server obtains target intention preparation techniques basis.Server
The target intention that user can be identified by preset intensified learning model, accurately judges the demand of user, obtains and uses
The chat content that family demand is more bonded reduces the number given an irrelevant answer between chat robots and user.
Further, whether server can belong to the label of product attribute to each target component mark in advance, be conducive to clothes
Business device gets target component can be judged at once, no longer need to carry out analyzing whether it is product attribute to each target component,
Conducive to the judgement time for shortening server.Server can obtain the corresponding Products Show letter of target component based on product database
Breath is pushed to user to subsequent, promotes the practical applicability and chat content scalability for using this chat robots.Service
Device can obtain the corresponding retrieval text of target component by networking data library, provide the user with inspection relevant to current round information
Suo Wenben improves the accuracy of reply content；Meanwhile server is not necessarily to locally saving retrieval text corresponding with target component
This, can save the memory space of server local.
Further, server can based on practical experience analyze user's portrait, obtain the corresponding user's property of user
Lattice, and corresponding target retro mode is matched in preset reply pattern base according to user's personality, with further enhance user and
Chat interest and affective interaction between chat robots.Server can will match with target retro mode after adjustment
Current return information is sent to client, enhances the affective interaction between chat robots and user, improves chat content
Validity.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
In one embodiment, a kind of interaction content generating means are provided, the interaction content generating means and above-described embodiment
Middle interaction content generation method corresponds.As shown in figure 8, the interaction content generating means include receiving current information module
10, history parameters module 20 is obtained, target intention module 30 is obtained, obtains target intention module 30, obtains retrieval text module
40 and obtain return information module 50.Detailed description are as follows for each functional module:
Current information module 10 is received, the current round information of the carrying session identification for receiving client transmission, base
In at least one current intent parameter of current round acquisition of information.
History parameters module 20 is obtained, inquires conversation recording database, acquisition and session identification for dialogue-based mark
At least one corresponding history intent parameter.
Obtain target intention module 30, for using preset intensified learning model at least one current intent parameter and
At least one history intent parameter is analyzed, and obtains target intention, target intention includes at least one target component and parameter
Priority orders.
Obtain retrieval text module 40, for be based on each target component query and search text database, obtain with it is each
The corresponding retrieval text of target component.
Obtain return information module 50, each retrieval text is converted for recalls information transformation model, obtain and
The corresponding current return information of each retrieval text pushes at least one current return information according to parameter priority sequence
To client.
Preferably, receiving current information module 10 includes receiving round information unit 11 and acquisition intent parameter unit 12.
Round information unit 11 is received, for receiving the current round information of client transmission, recalls information transformation model
Current round information is identified, lteral data is obtained.
Intent parameter unit 12 is obtained, for carrying out dissection process to lteral data using preset language processing model, is obtained
Take at least one current intent parameter.
Preferably, obtaining target intention module includes obtaining current word vector matrix and history term vector matrix list respectively
Member obtains part of speech vector matrix and obtains target intention unit.
Current word vector matrix and history term vector matrix unit are obtained respectively, for based at least one current intention ginseng
Several and at least one history intent parameter obtains current word vector matrix and history term vector matrix respectively.
Obtain part of speech vector matrix, for the part-of-speech tagging based on current intent parameter and history intent parameter, obtain with
Current word vector matrix and the corresponding current part of speech vector matrix of history term vector matrix and history part of speech vector matrix.
Obtain target intention unit, for using preset intensified learning model to current word vector matrix and history word to
Moment matrix is analyzed, and target intention is obtained.
Preferably, obtaining retrieval text module includes obtaining product information unit and obtaining retrieval text unit.
Product information unit is obtained, if the attribute for target component is product attribute, is looked into based on each target component
Product database corresponding with product attribute is ask, obtains Products Show information corresponding with each target component as retrieval
Text.
Retrieval text unit is obtained to look into if the attribute for target component is not product attribute based on each target component
Networking data library is ask, retrieval text corresponding with each target component is obtained.
Preferably, which further includes push product information module.
Product information module is pushed, for Products Show information to be pushed to client by pre-set product push format.
Preferably, which further includes push product information module and acquisition question text module.
Product information module is pushed, retrieves whether text contains at least two targets option for judging.
Question text module is obtained, if obtaining and at least two targets options for containing at least two targets option
Corresponding question text, recalls information transformation model convert question text, obtain current return information, will currently return
Complex information is pushed to client.
Specific about interaction content generating means limits the limit that may refer to above for interaction content generation method
Fixed, details are not described herein.Modules in above-mentioned interaction content generating means can fully or partially through software, hardware and its
Combination is to realize.Above-mentioned each module can be embedded in the form of hardware or independently of in the processor in computer equipment, can also be with
It is stored in the memory in computer equipment in a software form, in order to which processor calls the above modules of execution corresponding
Operation.
In one embodiment, a kind of computer equipment is provided, which can be server, internal structure
Figure can be as shown in Figure 9.The computer equipment includes processor, the memory, network interface sum number connected by system bus
According to library.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory of the computer equipment includes
Non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data
Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The database of machine equipment is used for the relevant data of interaction content generation method.The network interface of the computer equipment is used for and outside
Terminal by network connection communication.To realize a kind of interaction content generation method when the computer program is executed by processor.
In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory and can
The computer program run on a processor, processor realize above-described embodiment interaction content generation side when executing computer program
Method, such as S10 shown in Fig. 2 to step S50.Alternatively, processor is realized in above-described embodiment when executing computer program in interaction
Hold the function of each module/unit of generating means, such as module 10 shown in Fig. 8 to the function of module 50.To avoid repeating, herein
It repeats no more.
In one embodiment, a kind of computer readable storage medium is provided, computer program, computer journey are stored thereon with
Above-described embodiment interaction content generation method, such as S10 shown in Fig. 2 to step S50 are realized when sequence is executed by processor.Alternatively,
Each module/unit in interaction content generating means is realized in above-mentioned apparatus embodiment when the computer program is executed by processor
Function, such as module 10 shown in Fig. 8 is to the function of module 50.To avoid repeating, details are not described herein again.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, it is readable which can be stored in a non-volatile computer
It takes in storage medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, this Shen
Please any reference used in each embodiment to memory, storage, database or other media, may each comprise non-volatile
And/or volatile memory.Nonvolatile memory may include that read-only memory (ROM), programming ROM (PROM), electricity can be compiled
Journey ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include random access memory
(RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM
(SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.
The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although with reference to the foregoing embodiments
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these modification or
Replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of interaction content generation method characterized by comprising
The current round information for receiving the carrying session identification that client is sent, is based on the current round acquisition of information at least one
A current intent parameter；
Conversation recording database is inquired based on the session identification, obtains at least one history corresponding with the session identification
Intent parameter；
Ginseng is intended at least one described current intent parameter and at least one described history using preset intensified learning model
Number is analyzed, and target intention is obtained, and the target intention includes at least one target component and parameter priority sequence；
Based on each target component query and search text database, retrieval corresponding with each target component is obtained
Text；
Recalls information transformation model converts each retrieval text, obtains corresponding with each retrieval text
At least one described current return information is pushed to the client according to the parameter priority sequence by current return information
End.
2. interaction content generation method as described in claim 1, which is characterized in that the carrying meeting for receiving client and sending
The current round information of mark is talked about, described at least one current intent parameter of current round acquisition of information is based on, comprising:
The current round information that client is sent is received, recalls information transformation model identifies the current round information,
Obtain lteral data；
Dissection process is carried out to the lteral data using preset language processing model, obtains at least one current intent parameter.
3. interaction content generation method as described in claim 1, which is characterized in that described to use preset intensified learning model
At least one described current intent parameter and at least one described history intent parameter are analyzed, target intention, packet are obtained
It includes:
Based on current intent parameter described at least one and at least one described history intent parameter, current term vector is obtained respectively
Matrix and history term vector matrix；
The current word vector matrix and the history term vector matrix are analyzed using preset intensified learning model, obtained
Take target intention.
4. interaction content generation method as described in claim 1, which is characterized in that described to be looked into based on each target component
Retrieval text database is ask, retrieval text corresponding with each target component is obtained, comprising:
If the attribute of the target component is product attribute, based on each target component inquiry and the product attribute phase
Corresponding product database obtains Products Show information corresponding with each target component as retrieving text；
If the attribute of the target component is not product attribute, networking data library is inquired based on each target component, is obtained
Retrieval text corresponding with each target component.
5. interaction content generation method as claimed in claim 4, which is characterized in that if the attribute of the target component is product
Attribute, after acquisition retrieval text corresponding with each target component, the interaction content generation method is also
Include:
The Products Show information is pushed to the client by pre-set product push format.
6. interaction content generation method as described in claim 1, which is characterized in that in the recalls information transformation model to every
Before the one retrieval text is converted, the interaction content generation method further include:
Userspersonal information is obtained, user's portrait is generated based on the userspersonal information；
User's portrait is analyzed, user's personality corresponding with user portrait is obtained；
Preset reply pattern base is inquired based on user's personality, obtains target retro mode corresponding with user's personality；
The target retro mode is added to the information transformation model, to update the information transformation model.
7. a kind of interaction content generating means, which is characterized in that the interaction content generating means include:
Current information module is received, the current round information of the carrying session identification for receiving client transmission, based on described
At least one current intent parameter of current round acquisition of information；
History parameters module is obtained, for inquiring conversation recording database based on the session identification, is obtained and the session mark
At least one sensible corresponding history intent parameter；
Target intention module is obtained, for using preset intensified learning model at least one described current intent parameter and extremely
A few history intent parameter is analyzed, and obtains target intention, the target intention includes at least one target component
With parameter priority sequence；
Retrieval text module is obtained, for being based on each target component query and search text database, is obtained and each institute
State the corresponding retrieval text of target component；
Return information module is obtained, each retrieval text is converted for recalls information transformation model, is obtained and every
The corresponding current return information of the one retrieval text described current returns at least one according to the parameter priority sequence
Complex information is pushed to the client.
8. interaction content generating means as claimed in claim 7, which is characterized in that receiving current information module includes:
Round information unit is received, for receiving the current round information of client transmission, recalls information transformation model is to described
Current round information is identified, lteral data is obtained；
Intent parameter unit is obtained, for carrying out dissection process to the lteral data using preset language processing model, is obtained
At least one current intent parameter.
9. a kind of computer equipment, including memory, processor and storage are in the memory and can be in the processor
The computer program of upper operation, which is characterized in that the processor realized when executing the computer program as claim 1 to
Any one of 6 interaction content generation methods.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization interaction content generation method as described in any one of claim 1 to 6 when the computer program is executed by processor.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910047156.8A CN109857848A (en) | 2019-01-18 | 2019-01-18 | Interaction content generation method, device, computer equipment and storage medium |
| PCT/CN2019/120595 WO2020147428A1 (en) | 2019-01-18 | 2019-11-25 | Interactive content generation method and apparatus, computer device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910047156.8A CN109857848A (en) | 2019-01-18 | 2019-01-18 | Interaction content generation method, device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109857848A true CN109857848A (en) | 2019-06-07 |
Family
ID=66895125
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910047156.8A Pending CN109857848A (en) | 2019-01-18 | 2019-01-18 | Interaction content generation method, device, computer equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109857848A (en) |
| WO (1) | WO2020147428A1 (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222185A (en) * | 2019-06-13 | 2019-09-10 | 哈尔滨工业大学（深圳） | A kind of emotion information representation method of associated entity |
| CN110297544A (en) * | 2019-06-28 | 2019-10-01 | 联想(北京)有限公司 | Input information response's method and device, computer system and readable storage medium storing program for executing |
| CN110390108A (en) * | 2019-07-29 | 2019-10-29 | 中国工商银行股份有限公司 | Task exchange method and system based on deeply study |
| CN110413788A (en) * | 2019-07-30 | 2019-11-05 | 携程计算机技术（上海）有限公司 | Prediction technique, system, equipment and the storage medium of the scene type of session text |
| CN110489519A (en) * | 2019-07-05 | 2019-11-22 | 深圳追一科技有限公司 | The session method and Related product of dialogue-based prediction model |
| CN110555165A (en) * | 2019-07-23 | 2019-12-10 | 平安科技（深圳）有限公司 | information identification method and device, computer equipment and storage medium |
| CN110704641A (en) * | 2019-10-11 | 2020-01-17 | 零犀(北京)科技有限公司 | Ten-thousand-level intention classification method and device, storage medium and electronic equipment |
| CN110929166A (en) * | 2019-12-27 | 2020-03-27 | 咪咕文化科技有限公司 | Content recommendation method, electronic device and storage medium |
| CN111159380A (en) * | 2019-12-31 | 2020-05-15 | 腾讯科技（深圳）有限公司 | Interaction method and device, computer equipment and storage medium |
| CN111324710A (en) * | 2020-02-10 | 2020-06-23 | 深圳市医贝科技有限公司 | Virtual human-based online investigation method and device and terminal equipment |
| WO2020147428A1 (en) * | 2019-01-18 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Interactive content generation method and apparatus, computer device, and storage medium |
| CN111611365A (en) * | 2020-05-19 | 2020-09-01 | 上海鸿翼软件技术股份有限公司 | Flow control method, device, equipment and storage medium of dialog system |
| CN111694939A (en) * | 2020-04-28 | 2020-09-22 | 平安科技（深圳）有限公司 | Method, device and equipment for intelligently calling robot and storage medium |
| CN111767721A (en) * | 2020-03-26 | 2020-10-13 | 北京沃东天骏信息技术有限公司 | Information processing method, device and equipment |
| CN111859099A (en) * | 2019-12-05 | 2020-10-30 | 马上消费金融股份有限公司 | Recommendation method, device, terminal and storage medium based on reinforcement learning |
| CN112307168A (en) * | 2020-10-30 | 2021-02-02 | 康键信息技术（深圳）有限公司 | Artificial intelligence-based inquiry session processing method and device and computer equipment |
| CN112445906A (en) * | 2019-08-28 | 2021-03-05 | 北京搜狗科技发展有限公司 | Method and device for generating reply message |
| CN112632246A (en) * | 2020-12-21 | 2021-04-09 | 平安普惠企业管理有限公司 | Robot dialogue method and device based on deep learning and computer equipment |
| WO2021068321A1 (en) * | 2019-10-12 | 2021-04-15 | 平安科技（深圳）有限公司 | Information pushing method and apparatus based on human-computer interaction, and computer device |
| CN112825125A (en) * | 2019-11-21 | 2021-05-21 | 京东数字科技控股有限公司 | Sign language recognition method and device, computer storage medium and electronic equipment |
| CN112929253A (en) * | 2019-12-05 | 2021-06-08 | 北京沃东天骏信息技术有限公司 | Virtual image interaction method and device |
| CN112988956A (en) * | 2019-12-17 | 2021-06-18 | 北京搜狗科技发展有限公司 | Method and device for automatically generating conversation and method and device for detecting information recommendation effect |
| CN113157874A (en) * | 2021-02-20 | 2021-07-23 | 北京百度网讯科技有限公司 | Method, apparatus, device, medium, and program product for determining user's intention |
| CN113239146A (en) * | 2021-05-12 | 2021-08-10 | 平安科技（深圳）有限公司 | Response analysis method, device, equipment and storage medium |
| WO2021159668A1 (en) * | 2020-02-14 | 2021-08-19 | 深圳壹账通智能科技有限公司 | Robot dialogue method and apparatus, computer device and storage medium |
| CN113761176A (en) * | 2021-02-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and storage medium for extracting session abstract |
| CN114416937A (en) * | 2021-12-27 | 2022-04-29 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment, storage medium and computer program product |
| CN114661882A (en) * | 2022-03-30 | 2022-06-24 | 浪潮通用软件有限公司 | Robot chat management method, equipment and medium |
| US11651250B2 (en) | 2019-11-20 | 2023-05-16 | International Business Machines Corporation | Automatically generated conversation output |
| CN116737883A (en) * | 2023-08-15 | 2023-09-12 | 科大讯飞股份有限公司 | Man-machine interaction method, device, equipment and storage medium |
| CN116881429A (en) * | 2023-09-07 | 2023-10-13 | 四川蜀天信息技术有限公司 | Multi-tenant-based dialogue model interaction method, device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997342A (en) * | 2017-03-27 | 2017-08-01 | 上海奔影网络科技有限公司 | Intension recognizing method and device based on many wheel interactions |
| CN107918634A (en) * | 2017-06-27 | 2018-04-17 | 上海壹账通金融科技有限公司 | Intelligent answer method, apparatus and computer-readable recording medium |
| US20180129646A1 (en) * | 2011-03-31 | 2018-05-10 | Microsoft Technology Licensing, Llc | Augmented conversational understanding agent |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105068661B (en) * | 2015-09-07 | 2018-09-07 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method based on artificial intelligence and system |
| CN109857848A (en) * | 2019-01-18 | 2019-06-07 | 深圳壹账通智能科技有限公司 | Interaction content generation method, device, computer equipment and storage medium |
-
2019
- 2019-01-18 CN CN201910047156.8A patent/CN109857848A/en active Pending
- 2019-11-25 WO PCT/CN2019/120595 patent/WO2020147428A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180129646A1 (en) * | 2011-03-31 | 2018-05-10 | Microsoft Technology Licensing, Llc | Augmented conversational understanding agent |
| CN106997342A (en) * | 2017-03-27 | 2017-08-01 | 上海奔影网络科技有限公司 | Intension recognizing method and device based on many wheel interactions |
| CN107918634A (en) * | 2017-06-27 | 2018-04-17 | 上海壹账通金融科技有限公司 | Intelligent answer method, apparatus and computer-readable recording medium |
Cited By (43)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020147428A1 (en) * | 2019-01-18 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Interactive content generation method and apparatus, computer device, and storage medium |
| CN110222185A (en) * | 2019-06-13 | 2019-09-10 | 哈尔滨工业大学（深圳） | A kind of emotion information representation method of associated entity |
| CN110297544A (en) * | 2019-06-28 | 2019-10-01 | 联想(北京)有限公司 | Input information response's method and device, computer system and readable storage medium storing program for executing |
| CN110489519A (en) * | 2019-07-05 | 2019-11-22 | 深圳追一科技有限公司 | The session method and Related product of dialogue-based prediction model |
| CN110555165B (en) * | 2019-07-23 | 2023-04-07 | 平安科技（深圳）有限公司 | Information identification method and device, computer equipment and storage medium |
| CN110555165A (en) * | 2019-07-23 | 2019-12-10 | 平安科技（深圳）有限公司 | information identification method and device, computer equipment and storage medium |
| CN110390108A (en) * | 2019-07-29 | 2019-10-29 | 中国工商银行股份有限公司 | Task exchange method and system based on deeply study |
| CN110390108B (en) * | 2019-07-29 | 2023-11-21 | 中国工商银行股份有限公司 | Task type interaction method and system based on deep reinforcement learning |
| CN110413788A (en) * | 2019-07-30 | 2019-11-05 | 携程计算机技术（上海）有限公司 | Prediction technique, system, equipment and the storage medium of the scene type of session text |
| CN110413788B (en) * | 2019-07-30 | 2023-01-31 | 携程计算机技术（上海）有限公司 | Method, system, device and storage medium for predicting scene category of conversation text |
| CN112445906A (en) * | 2019-08-28 | 2021-03-05 | 北京搜狗科技发展有限公司 | Method and device for generating reply message |
| CN110704641A (en) * | 2019-10-11 | 2020-01-17 | 零犀(北京)科技有限公司 | Ten-thousand-level intention classification method and device, storage medium and electronic equipment |
| WO2021068321A1 (en) * | 2019-10-12 | 2021-04-15 | 平安科技（深圳）有限公司 | Information pushing method and apparatus based on human-computer interaction, and computer device |
| US11651250B2 (en) | 2019-11-20 | 2023-05-16 | International Business Machines Corporation | Automatically generated conversation output |
| CN112825125A (en) * | 2019-11-21 | 2021-05-21 | 京东数字科技控股有限公司 | Sign language recognition method and device, computer storage medium and electronic equipment |
| CN112929253B (en) * | 2019-12-05 | 2023-08-08 | 北京沃东天骏信息技术有限公司 | Virtual image interaction method and device |
| CN111859099A (en) * | 2019-12-05 | 2020-10-30 | 马上消费金融股份有限公司 | Recommendation method, device, terminal and storage medium based on reinforcement learning |
| CN111859099B (en) * | 2019-12-05 | 2021-08-31 | 马上消费金融股份有限公司 | Recommendation method, device, terminal and storage medium based on reinforcement learning |
| CN112929253A (en) * | 2019-12-05 | 2021-06-08 | 北京沃东天骏信息技术有限公司 | Virtual image interaction method and device |
| CN112988956A (en) * | 2019-12-17 | 2021-06-18 | 北京搜狗科技发展有限公司 | Method and device for automatically generating conversation and method and device for detecting information recommendation effect |
| CN110929166B (en) * | 2019-12-27 | 2023-10-20 | 咪咕文化科技有限公司 | Content recommendation method, electronic equipment and storage medium |
| CN110929166A (en) * | 2019-12-27 | 2020-03-27 | 咪咕文化科技有限公司 | Content recommendation method, electronic device and storage medium |
| CN111159380A (en) * | 2019-12-31 | 2020-05-15 | 腾讯科技（深圳）有限公司 | Interaction method and device, computer equipment and storage medium |
| CN111324710B (en) * | 2020-02-10 | 2024-01-12 | 深圳市医贝科技有限公司 | Online investigation method and device based on virtual person and terminal equipment |
| CN111324710A (en) * | 2020-02-10 | 2020-06-23 | 深圳市医贝科技有限公司 | Virtual human-based online investigation method and device and terminal equipment |
| WO2021159668A1 (en) * | 2020-02-14 | 2021-08-19 | 深圳壹账通智能科技有限公司 | Robot dialogue method and apparatus, computer device and storage medium |
| CN111767721A (en) * | 2020-03-26 | 2020-10-13 | 北京沃东天骏信息技术有限公司 | Information processing method, device and equipment |
| CN111694939B (en) * | 2020-04-28 | 2023-09-19 | 平安科技（深圳）有限公司 | Method, device, equipment and storage medium for intelligent robot calling |
| CN111694939A (en) * | 2020-04-28 | 2020-09-22 | 平安科技（深圳）有限公司 | Method, device and equipment for intelligently calling robot and storage medium |
| CN111611365A (en) * | 2020-05-19 | 2020-09-01 | 上海鸿翼软件技术股份有限公司 | Flow control method, device, equipment and storage medium of dialog system |
| CN112307168B (en) * | 2020-10-30 | 2023-11-07 | 康键信息技术（深圳）有限公司 | Artificial intelligence-based inquiry session processing method and device and computer equipment |
| CN112307168A (en) * | 2020-10-30 | 2021-02-02 | 康键信息技术（深圳）有限公司 | Artificial intelligence-based inquiry session processing method and device and computer equipment |
| CN112632246A (en) * | 2020-12-21 | 2021-04-09 | 平安普惠企业管理有限公司 | Robot dialogue method and device based on deep learning and computer equipment |
| CN113761176A (en) * | 2021-02-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and storage medium for extracting session abstract |
| CN113157874A (en) * | 2021-02-20 | 2021-07-23 | 北京百度网讯科技有限公司 | Method, apparatus, device, medium, and program product for determining user's intention |
| CN113239146B (en) * | 2021-05-12 | 2023-07-28 | 平安科技（深圳）有限公司 | Response analysis method, device, equipment and storage medium |
| CN113239146A (en) * | 2021-05-12 | 2021-08-10 | 平安科技（深圳）有限公司 | Response analysis method, device, equipment and storage medium |
| CN114416937B (en) * | 2021-12-27 | 2023-06-20 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment, storage medium and computer program product |
| CN114416937A (en) * | 2021-12-27 | 2022-04-29 | 北京百度网讯科技有限公司 | Man-machine interaction method, device, equipment, storage medium and computer program product |
| CN114661882A (en) * | 2022-03-30 | 2022-06-24 | 浪潮通用软件有限公司 | Robot chat management method, equipment and medium |
| CN116737883A (en) * | 2023-08-15 | 2023-09-12 | 科大讯飞股份有限公司 | Man-machine interaction method, device, equipment and storage medium |
| CN116881429A (en) * | 2023-09-07 | 2023-10-13 | 四川蜀天信息技术有限公司 | Multi-tenant-based dialogue model interaction method, device and storage medium |
| CN116881429B (en) * | 2023-09-07 | 2023-12-01 | 四川蜀天信息技术有限公司 | Multi-tenant-based dialogue model interaction method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020147428A1 (en) | 2020-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109857848A (en) | Interaction content generation method, device, computer equipment and storage medium | |
| CN109829039B (en) | Intelligent chat method, intelligent chat device, computer equipment and storage medium | |
| WO2020177282A1 (en) | Machine dialogue method and apparatus, computer device, and storage medium | |
| CN110175227A (en) | A kind of dialogue auxiliary system based on form a team study and level reasoning | |
| CN112989038B (en) | Sentence-level user portrait generation method and device and storage medium | |
| Windiatmoko et al. | Developing facebook chatbot based on deep learning using rasa framework for university enquiries | |
| CN111858940A (en) | Multi-head attention-based legal case similarity calculation method and system | |
| CN109635080A (en) | Acknowledgment strategy generation method and device | |
| CN111651571A (en) | Man-machine cooperation based session realization method, device, equipment and storage medium | |
| CN112926308B (en) | Method, device, equipment, storage medium and program product for matching text | |
| CN115481231A (en) | Session control method, device, computer equipment and storage medium | |
| CN113111157B (en) | Question-answer processing method, device, computer equipment and storage medium | |
| CN114253990A (en) | Database query method and device, computer equipment and storage medium | |
| CN117851563A (en) | Automatic question answering method and device, electronic equipment and readable storage medium | |
| CN110246494A (en) | Service request method, device and computer equipment based on speech recognition | |
| CN117131273A (en) | Resource searching method, device, computer equipment, medium and product | |
| CN112809694A (en) | Robot control method, device, storage medium and computer equipment | |
| Anamalamudi et al. | SmarTy: an intelligent voice response system model for Indian local languages | |
| CN116226320A (en) | Method and device for predicting context information, electronic equipment and storage medium | |
| CN115983269A (en) | Intelligent community data named entity identification method, terminal and computer medium | |
| CN114239604A (en) | Online consultation processing method and device and computer equipment | |
| CN113886539A (en) | Method and device for recommending dialect, customer service equipment and storage medium | |
| CN113569041A (en) | Text detection method and device, computer equipment and readable storage medium | |
| Dikshit et al. | Automating Questions and Answers of Good and Services Tax system using clustering and embeddings of queries | |
| CN111460106A (en) | Information interaction method, device and equipment |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190607 |
|
| WD01 | Invention patent application deemed withdrawn after publication |