CN108920622A - A kind of training method of intention assessment, training device and identification device - Google Patents
A kind of training method of intention assessment, training device and identification device Download PDFInfo
- Publication number
- CN108920622A CN108920622A CN201810694995.4A CN201810694995A CN108920622A CN 108920622 A CN108920622 A CN 108920622A CN 201810694995 A CN201810694995 A CN 201810694995A CN 108920622 A CN108920622 A CN 108920622A
- Authority
- CN
- China
- Prior art keywords
- model
- training
- vector
- intention assessment
- text
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
A kind of training method of intention assessment, device and identification device, method include：Obtain corpus text vector corresponding with the corpus in corpus library text；Construct the associated losses function formula of training pattern；Obtain training data；It is mapped to corresponding corpus text vector after carrying out cutting processing to training data, is denoted as training vector；Training sample vector is predicted using training pattern, according to prediction result, calculates the loss function value of each training pattern；Loss function value based on each model calculates associated losses functional value, judges whether associated losses functional value is less than given threshold, if so, training terminates, if not, updating each model parameter, continues repetitive exercise.So as to improve the generalization ability of intention assessment as far as possible, semantic ambiguity and Fault-Tolerant Problems are solved.
Description
Technical field
The present invention relates to technical field of intelligence, and in particular to a kind of training side of the intention assessment based on multi-task learning
Method, training device and identification device.
Background technique
" intention assessment " refers to one section inputted to user for expressing the information of query demand, determines affiliated intention
Classification.Current intention assessment technology is mainly used in search engine, interactive system etc., can specifically be divided into based on mould
Plate/dictionary and method based on Supervised classification excavate specific meaning based on template/dictionary method from user's history input
Artwork version/dictionary, if user's input matches with the word in template/dictionary of corresponding classification, then it is assumed that the input belongs to this
It is intended to classification；Method based on Supervised classification is then based on history input data building intent classifier model, prediction user's input
It is intended to classification.Applicant it has been investigated that, defect existing for technology essentially consists in following several respects at present：
1., generalization ability, the method based on template is limited to template and dictionary coverage rate problem, based on Supervised classification
Method is then limited to training corpus data scale and data quality problem.
2., ambiguity problem, semantic missing and Fault-Tolerant Problems, often there is imperfect, semantic missing in short text and input is wrong
Accidentally the problems such as, e.g., user's input, " play and after the night of bear ", are actually intended to search " the midnight seraglio of Teddy bear ".
In addition, some prior arts attempt the method based on multi-task learning also to classify.However this method is also deposited
In two problems：Firstly, this method be based on other tasks training obtain text vector, then with the text vector of goal task into
After row splicing, the classifier of this task of retraining, the error of other tasks is possible to bring a negative impact this task；Secondly,
If other tasks are unrelated with current task, a large amount of unrelated external informations are introduced, classification results may be brought disturb instead
It is dynamic.
Therefore, the missing of semanteme present in intention assessment, fault-tolerant, ambiguity and evolvement problem how are solved, this field is become
Technical staff's one of the technical problems that are urgent to solve.
Summary of the invention
In view of this, the embodiment of the invention provides a kind of training methods of intention assessment based on multi-task learning, instruction
Practice device and identification device, to solve model generalization ability when intention assessment, ambiguity problem and semantic Fault-Tolerant Problems.
To achieve the above object, the embodiment of the present invention provides the following technical solutions：
A kind of training method of intention assessment, including：
Corpus in corpus library text is mapped to semantic space, obtain and the low-dimensional corresponding with corpus it is dense to
Amount, is denoted as corpus text vector；
Construct the associated losses function formula of the training pattern, the training pattern includes intention assessment model, similar
Short text generates model and entity recognition model；
Obtain the training data for being directed to the training pattern；
Cutting processing is carried out to the training data, will treated training data is mapped to corresponding corpus library text to
Amount, is denoted as training vector；
The training vector is input to the training pattern, prediction result and the training based on training pattern output
The intention assessment model is calculated in the corresponding legitimate reading of vector, similar short texts generate model and entity recognition model
Loss function value；
It brings the loss function value of each model into the associated losses function formula and obtains associated losses functional value；
Judge whether the associated losses functional value is less than setting value, if not, adjusting the intention assessment model, similar
Short text generates the training parameter of model and entity recognition model, to reduce the loss function value of model, continues repetitive exercise, such as
Fruit is that training terminates.
Preferably, in the training method of above-mentioned intention assessment, the associated losses function formula of the training pattern is constructed, is wrapped
It includes：
Construct intention assessment model, similar short texts generate the associated losses function formula of model and entity recognition model：
Loss_total=α * lossintent_recognition+β*losssim_query_generation+γ*lossentity_recognition, wherein
Described α, β and γ are preset loss weight factor, the lossintent_recognitionFor the loss function of intention assessment model,
The lossentity_recognitionFor the loss function of entity recognition model, the losssim_query_generationFor Similar Text
Generate the loss function of model.
Preferably, in the training method of above-mentioned intention assessment, the corpus by corpus library text is mapped to semantic sky
Between, it obtains and the dense vector of low-dimensional corresponding with corpus, including：
Corpus in corpus data library is carried out to the cutting of the corresponding granularity of word level, word level or phonetic level；
The text after cutting is trained based on neural network model, is denoted as the dense vector of low-dimensional, it is described low
Tieing up dense vector includes word vector, term vector or phonetic vector.
Preferably, it in the training method of above-mentioned intention assessment, is modeled to obtain intention assessment for intention assessment task
Model, including：
It is modeled using LSTM model for intention assessment task, obtains intention assessment model, the intention assessment mould
The input of type is inquiry query, is exported to be intended to class label.
Preferably, it in the training method of above-mentioned intention assessment, is modeled to obtain similar short texts for similar short texts
Model is generated, including：
It is built to obtain similar short texts generation model for similar short texts using Seq2Seq model, it is described similar short
The input of text generation model is user input query query, is exported as similar short text.
Preferably, it in the training method of above-mentioned intention assessment, is modeled to obtain Entity recognition for Entity recognition task
Task model, including：
More disaggregated models are constructed based on convolutional neural networks, using more disaggregated models as Entity recognition task model
Entity recognition task is trained based on training data, the training data is：Included in text comprising required identification
Entity context text, mode input be the short text comprising entity, export as entity type label.
A kind of training device of intention assessment, including：
Corpus vector training unit, for the corpus in corpus library text to be mapped to semantic space, obtain with it is described with
The dense vector of the corresponding low-dimensional of corpus, is denoted as corpus text vector；
Model storage unit, for being stored with intention assessment model, similar short texts generate model and entity recognition model；
Associated losses function formula storage unit, for storing the associated losses function formula of training pattern, the training
Model includes intention assessment model, similar short texts generation model and entity recognition model；
Collecting training data unit, for obtaining the training data for being directed to the training pattern；
Training vector acquisition unit, for carrying out cutting processing to the training data, by treated, training data reflects
It is mapped to corresponding corpus text vector, is denoted as training vector；
Loss function value computing unit is based on training pattern for the training vector to be input to the training pattern
The prediction result of output and the corresponding legitimate reading of the training vector, are calculated the intention assessment model, similar short essay
The loss function value of this generation model and entity recognition model brings the loss function value of each model into the associated losses letter
Number formula obtains associated losses functional value；
Parameter adjustment unit, for judging whether the associated losses functional value is less than setting value, if not, described in adjustment
Intention assessment model, similar short texts generate the training parameter of model and entity recognition model, to reduce the loss function of model
Value continues repetitive exercise, if so, training terminates.
Preferably, in the training device of above-mentioned intention assessment, the associated losses function formula is：
Loss_total=α * lossintent_recognition+β*losssim_query_generation+γ*
lossentity_recognition, wherein described α, β and γ are preset loss weight factor, the lossintent_recognitionFor
The loss function of intention assessment model, the lossentity_recognitionIt is described for the loss function of entity recognition model
losssim_query_generationThe loss function of model is generated for Similar Text.
Preferably, in the training device of above-mentioned intention assessment, the corpus vector training unit is specifically used for：
Corpus in corpus data library is carried out to the cutting of the corresponding granularity of word level, word level or phonetic level；
The text after cutting is trained based on neural network model, is denoted as the dense vector of low-dimensional, it is described low
Tieing up dense vector includes word vector, term vector or phonetic vector.
Preferably, in the training device of above-mentioned intention assessment, the intention assessment model that is stored in the model storage unit
For：
It is modeled using LSTM model for intention assessment task, obtained intention assessment model, the intention assessment
The input of model is inquiry query, is exported to be intended to class label；
The similar short texts stored in the model storage unit generate model：
It is built to obtain similar short texts generation model for similar short texts using Seq2Seq model, it is described similar short
The input of text generation model is user input query query, is exported as similar short text；
The Entity recognition task model stored in the model storage unit is：
Based on more disaggregated models of convolutional neural networks building, the training data of more disaggregated models is：Comprising required
The context text of entity included in the text of identification, mode input are the short text comprising entity, are exported as entity class
Type label.
A kind of intention assessment equipment, including memory and processor；
The intention assessment mould that the training method training of above-mentioned any one intention assessment obtains is stored in the memory
Type, similar short texts generate model and entity recognition model, and the processor is used for when getting user query query, adjust
With and execute the intention assessment model, similar short texts generate model and entity recognition model.
Based on the above-mentioned technical proposal, above scheme provided in an embodiment of the present invention, can be based on the mode of multi-task learning
Input text is handled for intention assessment model, similar short texts generation model and entity recognition model using obtaining,
Relevant linguistry can more effectively be learnt, the generalization ability of intention assessment model is improved, can solve to be intended to as far as possible
Present in identification the problem of semantic ambiguity.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of flow diagram of the training method of intention assessment disclosed in the embodiment of the present application；
Fig. 2 is the example based on LSTM model construction intention assessment model；
Fig. 3 is a kind of structural schematic diagram of the training device of intention assessment provided by the present application；
Fig. 4 is a kind of structural schematic diagram of intention assessment device disclosed in the embodiment of the present application.
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 only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
When being directed in the prior art to input text progress intention assessment, often there is the problem of ambiguity, the application is public
The training method for having opened a kind of intention assessment, referring to Fig. 1, this method may include：
Step S101：Corpus in expectation database is trained to obtain text vector corresponding with corpus；
Specifically, the corpus in corpus library text is mapped to semantic space in this step, obtain and described and corpus pair
The dense vector of the low-dimensional answered, is denoted as corpus text vector, and the concrete mode of the dense vector of low-dimensional can be word vector, word
Vector, phonetic vector, in training, used training pattern can be using models such as word2vec.Wherein, the corpus number
The corpus such as information, click logs and speech recognition result information, these languages are inputted according to the history text for being stored with user in library
Material is specially textual form, wherein text entry information is the input data that user is keyed in by input equipment, the click
Log refers to that the linked contents that user is clicked, such as user scan for, when being clicked by mouse and opening a link,
The content of click and corresponding query are stored in click logs, the voice messaging be user by phonetic search or into
Voice content when other interactive voices of row, wherein voice messaging, which can first pass through speech recognition technology and be converted to, to match
It is being stored in the corpus data library after text as voice corpus.Specifically, the corpus data library can be by default
Data grabber using corresponding corpus is grabbed in log by system each in electronic equipment, for example, the corpus data library
In corpus can be and applied by data grabber from search query log, click logs, question and answer knowledge base title, microblogging, language
It is acquired in the magnanimity short message data such as sound conversation log, the history grabbed is inputted into text information, click logs information
And voice messaging etc. is saved to the corpus data library.
This step specifically includes following steps：Firstly, some text (corpus) in corpus data library is first carried out
Then corresponding word level/word level/phonetic level granularity cutting is trained based on neural network model, after cutting
Text representation is the dense real vector of low-dimensional.For example, text is " Chinese people ", at this point, based on character/word/phonetic to text
After carrying out cutting, available " in/state/people/people ", " China/people ", " zhongguo/renmin ".It is obtained in above scheme
Word vector corresponding with text is to obtain phonetic vector corresponding with text in order to solve the problems, such as unregistered word, be to solve
The false transitions problem of compliant speech identification.Certainly, it stores and extracts for the ease of data, the application can be with a matrix type
The corresponding corpus text vector of each corpus is stored, specifically, being based on after obtaining the dense vector of the corresponding low-dimensional of corpus
Word2vec model each dense vector of low-dimensional corresponding to corpus is combined, and obtains the corresponding feature vector square of the text
Battle array.For example, what is selected when specific implementation is the word vector of 100 dimensions, then after training, should " in/state/people/people " obtain one
The word vector matrix of a corresponding 4*100, the vector of the corresponding individual character of every row of matrix, that each column indicates is the every of word vector
It is one-dimensional, it is as follows：
" in " [- 0.1111,0.111 ..., 0.1233]
" state " [0.5324,0.3678 ..., -0.1111]
" people " [- 0.1111,0.1915 ..., 0.4996]
" people " [- 0.1111,0.8113 ..., 0.3267].
Step S102：The associated losses function formula of the training pattern is constructed, the training pattern includes intention assessment
Model, similar short texts generate model and entity recognition model；
Before the associated losses function formula for constructing the training pattern, it is also necessary to model for each task
To training pattern, specifically：
It is modeled respectively for intention assessment task, similar short texts task and Entity recognition task, obtains being intended to know
Other model, similar short texts generate model and entity recognition model；
For intention assessment task, when being trained to intention assessment task, training data is the inquiry of user's input
Query and intention class label corresponding with the inquiry query are based on after being segmented by inquiry query first or divide word
The training result of step S101 obtains each word or word is mapped to corresponding vector, is then mapped to each word or word corresponding
Institute's directed quantity is averaging, and is obtained the corresponding vector of inquiry query, is denoted as V1, then carries out entity knowledge to the inquiry query
Not, check whether inquiry query matches the AD HOC in preset regular pattern table, such as：" playing fellow No.9's door ", fellow No.9
Door is identified as album by the preset regular pattern table, matches the mode in the regular pattern table：Album is played, then
Inquiry query obtains corresponding K dimensional vector V2, wherein K is the number of modes in regular pattern table, if i-th dimension value is
1, then it represents inquiry query and matches i-th of mode in regular pattern table, value is that 0 representative does not match regular pattern
Mode in table.Then using V1 and V2 as the input of default neural network model.In the present embodiment, the structure packet of neural network
Include input layer, lstm layers, layer layers of dense and layer layers of skip.For details, reference can be made to Fig. 2 examples.
In the present embodiment, for example, for video search engine, the inquiry query of input is " playing fellow No.9's door ", then defeated
Intention class label out is " PLAY_VIDEO ", and the inquiry query of input is " downloading king's honor ", then the intention class exported
Distinguishing label is " DOWNLOAD_GAME ", and the inquiry query of input is " member supplements with money ", then the intention labels exported are
“PURCHASE”。
Task is generated for similar short texts, training data is similar short texts pair, specifically includes following three classes：①,
Query is inquired, the inquiry query of same document is clicked with inquiry query, 2., inquires query, same session's looks into
Query is ask, 3., inquires query, and the title that inquiry query is clicked.It is built generating task to similar short texts
When mould, modeled to obtain similar short texts generation model, the similar short texts for similar short texts based on Seq2Seq model
The input for generating model is user query query, is exported as short text similar with the inquiry query.
The target of Seq2seq model is based on given sequence X, output sequence Y, and loss function value is calculated using cross entropy
It arrives.Seq2seq model is made of two Recognition with Recurrent Neural Network of encoder and decoder, and encoder neural network will input sequence
Column are converted to the vector of a regular length, and the vector that decoder neural network is generated according to encoder neural network generates
One new sequence.When model training, the vector of encoder neural network and other task sharings.When predicting text, such as：
The input of model is：" how buying member ", encoder neural network by this section of text one by one word or one by one word sequentially input to
Network, is converted into the vector of a regular length, and the vector of the regular length is converted to new text by decoder neural network
This output, such as output " how being supplemented with money to member ".
For Entity recognition task, more disaggregated models are constructed as entity recognition model based on convolutional neural networks, are used
More disaggregated models are based on training data and are trained to Entity recognition task, the training data of the entity recognition model
For：Context text comprising the entity, mode input are the short text comprising entity, are exported as entity type label.Training
The specific source of data can be：The internets such as encyclopaedia disclose corpus resource, or by the application related personnel language that mark constructs by hand
Expect data.
In the present embodiment, convolutional neural networks framework includes following several layers of：Vector search layer, linear transformation layer 1,
Sigmoid layers, linear transformation layer 2, whole sentence analysis layer.When being trained to the convolutional neural networks, neural network model
Based on stochastic gradient descent algorithm, it is trained using cross entropy loss function value.In the entity type mark of prediction input text
When label, for example, in moment t, system Chinese character to be processed is " north " for input " Beijing ", it will be described by vector search layer
" north " is mapped as real vector, is transferred to linear transformation layer, and after linear layer and sigmoid layers of processing, system is to " north "
All possible labels of correspondence give a mark, the maximum label of Marking Probability, score value is higher, and mark probability is bigger.In subsequent time
T+1, system then handle next Chinese character " capital " in sentence.In sentence analytic sheaf, for the text generation score value net of processing
Network, wherein the node of t column is the score value of the corresponding all entity type labels of t moment Chinese character to be processed, the node of t+1 column
Line identifies transition probability between the node of t column, for describing the possibility of transformation between label, is finally based on viterbi
Algorithm finds the highest path of overall scores, as final flag sequence in a network.For example, " Beijing " corresponding reality
Body type label path is " B-LOC I-LOC I-LOC ".
After the completion of training pattern building, the loss function value of each training pattern is constructed, wherein the damage of intention assessment model
Mistake functional value is lossintent_recognition, the loss function value of entity recognition model is lossentity_recognition, similar text
The loss function value of this generation model isThe lossintent_recognition、
lossentity_recognitionAnd losssim_query_generationValue can when being trained to training pattern using mean square deviation or
Person's Cross-Entropy Method obtains.For each loss function value assign one specifically loss weight factor, will assign loss weight factor it
The sum of three loss function values afterwards generate the joint of model and entity recognition model as intention assessment model, similar short texts
Loss function value, that is, pass through associated losses function formula：
Loss_total=α * lossintent_recognition+β*losssim_query_generation+γ*
lossentity_recognitionCalculate to obtain associated losses functional value loss_total, wherein wherein α, β and γ are respectively
lossintent_recognition、lossentity_recognitionAnd losssim_query_generationLoss weight factor, described α, β and
Between (0,1), occurrence can voluntarily be adjusted the value of γ according to user demand.That is, three models respectively have loss function,
If x is equivalent to loss at this point, the loss function of intention assessment model is f1 (x) for user's inputintent_recognition, similar
The loss function of short text model is f2 (x), is equivalent to lossentity_recognition, the loss function of entity recognition model is f3
(x), it is equivalent to losssim_query_generation, then associated losses function is the linear weighted function of 3 loss functions, a*f1 (x)+c*
F2 (x)+b*f3 (x), wherein a is equivalent to α, and b is equivalent to β, and c is equivalent to γ.
Step S103：Obtain the training data for being directed to the training pattern；
Wherein, the training data includes generating model and Entity recognition mould for intention assessment model, similar short texts
The training data of type；
In this step, the training data is some data in preset training data set, in specific data
Appearance can voluntarily be selected according to user demand, using the training data as the input text of above-mentioned each model, and each instruction
Practice the corresponding actual similar short texts of data and class label is known.
Step S104：Cutting processing is carried out to the training data；
This step specifically includes：To the intention assessment model, similar short texts generation model and entity recognition model
Training data is segmented, word or words mixing cutting is divided to handle；
Step S105：By treated, training data is mapped to corresponding corpus text vector, is denoted as training vector；
Word or words mixing cutting are segmented or are divided to the input text of each model, when input text is Chinese text
This when, can choose the participle model based on hidden Markov, character or condition random field and carry out cutting to input text, when defeated
When to enter text be English text, it may be selected to carry out cutting as separator based on punctuate and space, if embodiment selection divides
Word, input text are：" I wants to see fellow No.9's door ", the input text are split as " I/thinking/sees/fellow No.9's door ".Such as inputting text is
" I want to play game ", which is split as " I/want/to/play/game "；If embodiment is selected
Select a point word, then press character cutting, such as can will input text " I want see fellow No.9 door " cutting be " I/thinking/see/it is old/nine/
Door ", in addition, also can choose words mixing cutting according to needs are applied, for example, cutting to the Chinese in input text by character
The mode of dividing carries out cutting to input text, carries out cutting to text is inputted by word segmentation mode to the English in input text, such as：
" I sees/billions at/thinking/".
After the training data to each mode input carries out cutting processing, the training data after cutting is mapped to step
Corpus text vector obtained in S101, detailed process are：Obtain corpus institute identical with training data in corpus data library
Corresponding corpus data library text vector, the corpus data library text vector are training vector corresponding with training data,
It is specifically as follows word vector, term vector or phonetic vector, if corpus text vector is phonetic vector, by training data
Son or word are converted to phonetic, then re-map to obtain corresponding phonetic vector.Above three model can share a set of character/word/spelling
Sound vector.
Step S106：The training vector is input to the training pattern, the prediction result based on training pattern output
The intention assessment model is calculated in legitimate reading corresponding with the training vector, similar short texts generate model and reality
The loss function value of body identification model；
In this step, when each model of each training, circulation is gone through all over each task, randomly selects a part training number
According to corresponding training vector as mode input, output result and the corresponding legitimate reading meter of the training data according to model
The loss function value of each model is calculated, specifically, in the present embodiment, for each model, result based on model prediction and every
The corresponding legitimate reading of a training data carries out one-hot processing, calculates cross entropy loss function, obtains the damage of each model
Lose functional value.
Step S107：Joint damage is calculated in loss function value based on the associated losses function formula and each model
Lose functional value；
In this step, each loss function value that step S107 is calculated brings the associated losses function formula into,
The associated losses functional value can be obtained.
Step S108：Judge whether the associated losses functional value is less than setting value, if so, training terminates, if not,
Execute step S109；
Step S109：The intention assessment model is updated based on Back Propagation Algorithm, similar short texts generate model and reality
The parameter of body identification model, the input vector of three model sharings continue to execute step as model parameter also synchronized update
S103 is iterated training.
In the above scheme, input text is handled by way of multi-task learning, can be more effectively learnt
Relevant linguistry, the generalization ability of raising intention assessment model, as far as possible semanteme missing present in solution intention assessment,
Fault-tolerant, ambiguity and it is extensive the problems such as, specifically, the intention assessment model, similar short essay can be updated based on back-propagation algorithm
When the parameter of this generation model and entity recognition model.
Corresponding to the above method, disclosed herein as well is a kind of training devices of intention assessment, referring to Fig. 3 comprising：
Corpus vector training unit 01, for the corpus in corpus library text to be mapped to semantic space, obtain with it is described
The dense vector of low-dimensional corresponding with corpus, is denoted as corpus text vector；
Model storage unit 02, for being stored with intention assessment model, similar short texts generate model and Entity recognition mould
Type；
Associated losses function formula storage unit 03, for storing the associated losses function formula of training pattern, the instruction
Practicing model includes intention assessment model, similar short texts generation model and entity recognition model；
Collecting training data unit 04, for obtaining the training data for being directed to the training pattern；
Training vector acquisition unit 05 will treated training data for carrying out cutting processing to the training data
It is mapped to corresponding corpus text vector, is denoted as training vector；
Loss function value computing unit 06, for the training vector to be input to the training pattern, based on training mould
The prediction result and the corresponding legitimate reading of the training vector of type output, are calculated the intention assessment model, similar short
The loss function value of text generation model and entity recognition model brings the loss function value of each model into the associated losses
Function formula obtains associated losses functional value；
Parameter adjustment unit 07, for judging whether the associated losses functional value is less than setting value, if not, adjustment institute
The training parameter of intention assessment model, similar short texts generation model and entity recognition model is stated, to reduce the loss letter of model
Numerical value continues repetitive exercise, if so, training terminates.
It corresponds to the above method, the history text input information of user is stored in the corpus data library, clicks day
Will and voice messaging can be based on information above training corpus text vector, and the training dataset of three models of construction.
It corresponds to the above method, when the corpus vector training unit training vector, is specifically used for：
Corpus in corpus data library is carried out to the cutting of the corresponding granularity of word level, word level or phonetic level；
The text after cutting is trained based on neural network model, is denoted as the dense vector of low-dimensional, it is described low
Tieing up dense vector includes word vector, term vector or phonetic vector.
It corresponds to the above method, the intention assessment model stored in the model storage unit is：
It is modeled using LSTM model for intention assessment task, obtained intention assessment model, the intention assessment
The input of model is inquiry query, is exported to be intended to class label；
The similar short texts stored in the model storage unit generate model：
It is built to obtain similar short texts generation model for similar short texts using Seq2Seq model, it is described similar short
The input of text generation model is user input query query, is exported as similar short text；
The Entity recognition task model stored in the model storage unit is：
Based on more disaggregated models of convolutional neural networks building, the training data of more disaggregated models is：Comprising required
The context text of entity included in the text of identification, mode input are the short text comprising entity, are exported as entity class
Type label.
Corresponding to the training method of above-mentioned intention assessment, disclosed herein as well is a kind of intention assessment equipment, referring to fig. 4,
For intention assessment device structure schematic diagram disclosed in the embodiment of the present application, which may include：
Memory 100 and processor 200；
The intention assessment equipment further includes communication interface 300 and communication bus 400, wherein memory 100, processing
Device 200 and the communication of communication interface 300 realize mutual communication by communication bus 400.
The memory 100 is for storing program code；Said program code includes computer operation instruction.Specifically,
The meaning that the training method training of intention assessment disclosed in the above-mentioned any one embodiment of the application obtains is stored in the memory
Figure identification model, similar short texts generate the program code of model and entity recognition model.
Memory 100 may include high speed RAM memory, it is also possible to further include nonvolatile memory (non-
Volatile memory), a for example, at least magnetic disk storage.
The processor 200 can be a central processor CPU or specific integrated circuit ASIC
(Application Specific Integrated Circuit), or be arranged to implement the one of the embodiment of the present invention
A or multiple integrated circuits.The processor 200 is for calling and executing said program code.Specifically, the processor is used
In when getting user query query, call and execute the intention assessment model, similar short texts generate model and entity
Identification model.
For convenience of description, it is divided into various modules when description system above with function to describe respectively.Certainly, implementing this
The function of each module can be realized in the same or multiple software and or hardware when application.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for system or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.System and system embodiment described above is only schematical, wherein the conduct
The unit of separate part description may or may not be physically separated, component shown as a unit can be or
Person may not be physical unit, it can and it is in one place, or may be distributed over multiple network units.It can root
According to actual need that some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel can understand and implement without creative efforts.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
It should also be noted that, herein, relational terms such as first and second and the like are used merely to one
Entity or operation are distinguished with another entity or operation, without necessarily requiring or implying between these entities or operation
There are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant are intended to contain
Lid non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (11)
1. a kind of training method of intention assessment, which is characterized in that including：
Corpus in corpus library text is mapped to semantic space, is obtained and the dense vector of low-dimensional corresponding with corpus, note
For corpus text vector；
The associated losses function formula of training pattern is constructed, the training pattern includes intention assessment model, similar short texts life
At model and entity recognition model；
Obtain the training data for being directed to the training pattern；
Cutting processing is carried out to the training data, by treated, training data is mapped to corresponding corpus text vector,
It is denoted as training vector；
The training vector is input to the training pattern, prediction result and the training vector based on training pattern output
The intention assessment model is calculated in corresponding legitimate reading, similar short texts generate the damage of model and entity recognition model
Lose functional value；
It brings the loss function value of each model into the associated losses function formula and obtains associated losses functional value；
Judge whether the associated losses functional value is less than setting value, if not, adjusting the intention assessment model, similar short essay
The training parameter of this generation model and entity recognition model continues repetitive exercise to reduce the loss function value of model, if
It is that training terminates.
2. the training method of intention assessment according to claim 1, which is characterized in that construct the joint of the training pattern
Loss function formula, including：
Construct intention assessment model, similar short texts generate the associated losses function formula of model and entity recognition model：loss_
Total=α * lossintent_recognition+β*losssim_query_generation+γ*lossentity_recognition, wherein the α,
β and γ is preset loss weight factor, the lossintent_recognitionIt is described for the loss function of intention assessment model
lossentity_recognitionFor the loss function of entity recognition model, the losssim_query_generationFor Similar Text generation
The loss function of model.
3. the training method of intention assessment according to claim 1, which is characterized in that the language by corpus library text
Material is mapped to semantic space, obtains and the dense vector of low-dimensional corresponding with corpus, including：
Corpus in corpus data library is carried out to the cutting of the corresponding granularity of word level, word level or phonetic level；
The text after cutting is trained based on neural network model, is denoted as the dense vector of low-dimensional, the low-dimensional is thick
Close vector includes word vector, term vector or phonetic vector.
4. the training method of intention assessment according to claim 1, which is characterized in that built for intention assessment task
Mould obtains intention assessment model, including：
It is modeled using LSTM model for intention assessment task, obtains intention assessment model, the intention assessment model
Input is inquiry query, is exported to be intended to class label.
5. the training method of intention assessment according to claim 1, which is characterized in that modeled for similar short texts
It obtains similar short texts and generates model, including：
It is built to obtain similar short texts generation model, the similar short texts for similar short texts using Seq2Seq model
The input for generating model is user input query query, is exported as similar short text.
6. the training method of intention assessment according to claim 1, which is characterized in that built for Entity recognition task
Mould obtains Entity recognition task model, including：
More disaggregated models are constructed based on convolutional neural networks, are based on using more disaggregated models as Entity recognition task model
Training data is trained Entity recognition task, and the training data is：It is real included in text comprising required identification
The context text of body, mode input are the short text comprising entity, are exported as entity type label.
7. a kind of training device of intention assessment, which is characterized in that including：
Corpus vector training unit obtains and described and corpus for the corpus in corpus library text to be mapped to semantic space
The corresponding dense vector of low-dimensional, is denoted as corpus text vector；
Model storage unit, for being stored with intention assessment model, similar short texts generate model and entity recognition model；
Associated losses function formula storage unit, for storing the associated losses function formula of training pattern, the training pattern
Model and entity recognition model are generated including intention assessment model, similar short texts；
Collecting training data unit, for obtaining the training data for being directed to the training pattern；
Training vector acquisition unit, for carrying out cutting processing to the training data, by treated, training data is mapped to
Corresponding corpus text vector, is denoted as training vector；
Loss function value computing unit is exported for the training vector to be input to the training pattern based on training pattern
Prediction result and the corresponding legitimate reading of the training vector, it is raw that the intention assessment model, similar short texts are calculated
At the loss function value of model and entity recognition model, it is public to bring the loss function value of each model into the associated losses function
Formula obtains associated losses functional value；
Parameter adjustment unit, for judging whether the associated losses functional value is less than setting value, if not, adjusting the intention
Identification model, similar short texts generate the training parameter of model and entity recognition model, to reduce the loss function value of model, after
Continuous repetitive exercise, if so, training terminates.
8. the training device of intention assessment according to claim 7, which is characterized in that the associated losses function formula
For：
Loss_total=α * lossintent_recognition+β*losssim_query_generation+γ*lossentity_recognition,
In, described α, β and γ are preset loss weight factor, the lossintent_recognitionFor the loss letter of intention assessment model
Number, the lossentity_recognitionFor the loss function of entity recognition model, the losssim_query_generationFor similar text
The loss function of this generation model.
9. the training device of intention assessment according to claim 7, which is characterized in that the corpus vector training unit,
It is specifically used for：
Corpus in corpus data library is carried out to the cutting of the corresponding granularity of word level, word level or phonetic level；
The text after cutting is trained based on neural network model, is denoted as the dense vector of low-dimensional, the low-dimensional is thick
Close vector includes word vector, term vector or phonetic vector.
10. the training device of intention assessment according to claim 7, which is characterized in that deposited in the model storage unit
The intention assessment model of storage is：
It is modeled using LSTM model for intention assessment task, obtained intention assessment model, the intention assessment model
Input be inquiry query, export for be intended to class label；
The similar short texts stored in the model storage unit generate model：
It is built to obtain similar short texts generation model, the similar short texts for similar short texts using Seq2Seq model
The input for generating model is user input query query, is exported as similar short text；
The Entity recognition task model stored in the model storage unit is：
Based on more disaggregated models of convolutional neural networks building, the training data of more disaggregated models is：Include required identification
Text included in entity context text, mode input be the short text comprising entity, export as entity type mark
Label.
11. a kind of intention assessment equipment, which is characterized in that including memory and processor；
The intention assessment that the training method training of claim 1-6 any one intention assessment obtains is stored in the memory
Model, similar short texts generate model and entity recognition model, and the processor is used for when getting user query query,
Call and execute the intention assessment model, similar short texts generate model and entity recognition model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810694995.4A CN108920622B (en) | 2018-06-29 | 2018-06-29 | Training method, training device and recognition device for intention recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810694995.4A CN108920622B (en) | 2018-06-29 | 2018-06-29 | Training method, training device and recognition device for intention recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108920622A true CN108920622A (en) | 2018-11-30 |
| CN108920622B CN108920622B (en) | 2021-07-20 |
Family
ID=64423645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810694995.4A Active CN108920622B (en) | 2018-06-29 | 2018-06-29 | Training method, training device and recognition device for intention recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108920622B (en) |
Cited By (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109545186A (en) * | 2018-12-16 | 2019-03-29 | 初速度（苏州）科技有限公司 | A kind of speech recognition training system and method |
| CN109741751A (en) * | 2018-12-11 | 2019-05-10 | 上海交通大学 | Intension recognizing method and device towards intelligent sound control |
| CN109816111A (en) * | 2019-01-29 | 2019-05-28 | 北京金山数字娱乐科技有限公司 | Reading understands model training method and device |
| CN109857844A (en) * | 2018-12-29 | 2019-06-07 | 北京三快在线科技有限公司 | Intension recognizing method, device, electronic equipment based on dialog text of ordering |
| CN109857868A (en) * | 2019-01-25 | 2019-06-07 | 北京奇艺世纪科技有限公司 | Model generating method, file classification method, device and computer readable storage medium |
| CN109871485A (en) * | 2019-02-13 | 2019-06-11 | 北京航空航天大学 | A kind of personalized recommendation method and device |
| CN109933663A (en) * | 2019-02-26 | 2019-06-25 | 上海凯岸信息科技有限公司 | Intention assessment algorithm based on embedding method |
| CN110085308A (en) * | 2019-04-23 | 2019-08-02 | 挂号网（杭州）科技有限公司 | A kind of diagnosis and treatment department classification method based on fusion deep learning |
| CN110134969A (en) * | 2019-05-27 | 2019-08-16 | 北京奇艺世纪科技有限公司 | A kind of entity recognition method and device |
| CN110147535A (en) * | 2019-04-18 | 2019-08-20 | 平安科技（深圳）有限公司 | Similar Text generation method, device, equipment and storage medium |
| CN110222339A (en) * | 2019-06-05 | 2019-09-10 | 深圳市思迪信息技术股份有限公司 | Based on the intension recognizing method and device for improving XGBoost algorithm |
| CN110287283A (en) * | 2019-05-22 | 2019-09-27 | 中国平安财产保险股份有限公司 | Intent model training method, intension recognizing method, device, equipment and medium |
| CN110334344A (en) * | 2019-06-13 | 2019-10-15 | 腾讯科技（深圳）有限公司 | A kind of semanteme intension recognizing method, device, equipment and storage medium |
| CN110516035A (en) * | 2019-07-05 | 2019-11-29 | 同济大学 | A kind of man-machine interaction method and system of mixing module |
| CN110532368A (en) * | 2019-09-04 | 2019-12-03 | 深圳前海达闼云端智能科技有限公司 | Question answering method, electronic equipment and computer readable storage medium |
| CN110610001A (en) * | 2019-08-12 | 2019-12-24 | 大箴(杭州)科技有限公司 | Short text integrity identification method and device, storage medium and computer equipment |
| CN110619047A (en) * | 2019-09-16 | 2019-12-27 | 出门问问信息科技有限公司 | Method and device for constructing natural language model and readable storage medium |
| CN110674253A (en) * | 2019-09-23 | 2020-01-10 | 出门问问信息科技有限公司 | Semantic processing method and equipment |
| CN110807332A (en) * | 2019-10-30 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Training method of semantic understanding model, semantic processing method, semantic processing device and storage medium |
| CN110852108A (en) * | 2019-11-11 | 2020-02-28 | 中山大学 | Joint training method, apparatus and medium for entity recognition and entity disambiguation |
| CN110956018A (en) * | 2019-11-22 | 2020-04-03 | 腾讯科技（深圳）有限公司 | Training method of text processing model, text processing method, text processing device and storage medium |
| CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN111198938A (en) * | 2019-12-26 | 2020-05-26 | 深圳市优必选科技股份有限公司 | Sample data processing method, sample data processing device and electronic equipment |
| CN111209383A (en) * | 2020-01-06 | 2020-05-29 | 广州小鹏汽车科技有限公司 | Method and device for processing multi-turn dialogue, vehicle, and storage medium |
| CN111292752A (en) * | 2018-12-06 | 2020-06-16 | 北京嘀嘀无限科技发展有限公司 | User intention identification method and device, electronic equipment and storage medium |
| CN111309915A (en) * | 2020-03-03 | 2020-06-19 | 爱驰汽车有限公司 | Method, system, device and storage medium for training natural language of joint learning |
| CN111354354A (en) * | 2018-12-20 | 2020-06-30 | 深圳市优必选科技有限公司 | Training method and device based on semantic recognition and terminal equipment |
| CN111428872A (en) * | 2019-01-10 | 2020-07-17 | 维萨国际服务协会 | Systems, methods, and computer program products for incorporating knowledge from more complex models into simpler models |
| CN111538894A (en) * | 2020-06-19 | 2020-08-14 | 腾讯科技（深圳）有限公司 | Query feedback method and device, computer equipment and storage medium |
| CN111563208A (en) * | 2019-01-29 | 2020-08-21 | 株式会社理光 | Intention identification method and device and computer readable storage medium |
| CN111597804A (en) * | 2020-05-15 | 2020-08-28 | 腾讯科技（深圳）有限公司 | Entity recognition model training method and related device |
| CN111612025A (en) * | 2019-02-25 | 2020-09-01 | 北京嘀嘀无限科技发展有限公司 | Description model training method, text description device and electronic equipment |
| CN111667066A (en) * | 2020-04-23 | 2020-09-15 | 北京旷视科技有限公司 | Network model training and character recognition method and device and electronic equipment |
| CN111680514A (en) * | 2019-02-25 | 2020-09-18 | 北京猎户星空科技有限公司 | Information processing and model training method, device, equipment and storage medium |
| CN111680517A (en) * | 2020-06-10 | 2020-09-18 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for training a model |
| CN111737416A (en) * | 2020-06-29 | 2020-10-02 | 重庆紫光华山智安科技有限公司 | Case processing model training method, case text processing method and related device |
| CN111785350A (en) * | 2020-06-30 | 2020-10-16 | 易联众信息技术股份有限公司 | Information extraction method, application, device and medium |
| WO2020211008A1 (en) * | 2019-04-17 | 2020-10-22 | 深圳市欢太科技有限公司 | Speech recognition method and apparatus, storage medium and electronic device |
| CN111949784A (en) * | 2020-08-14 | 2020-11-17 | 中国工商银行股份有限公司 | Outbound method and device based on intention recognition |
| WO2020237869A1 (en) * | 2019-05-31 | 2020-12-03 | 平安科技（深圳）有限公司 | Question intention recognition method and apparatus, computer device, and storage medium |
| CN112102832A (en) * | 2020-09-18 | 2020-12-18 | 广州小鹏汽车科技有限公司 | Speech recognition method, speech recognition device, server and computer-readable storage medium |
| CN112183631A (en) * | 2020-09-28 | 2021-01-05 | 云知声智能科技股份有限公司 | Method and terminal for establishing intention classification model |
| CN112185351A (en) * | 2019-07-05 | 2021-01-05 | 北京猎户星空科技有限公司 | Voice signal processing method and device, electronic equipment and storage medium |
| CN112256826A (en) * | 2020-10-19 | 2021-01-22 | 网易（杭州）网络有限公司 | Emotion analysis method, evaluation method and emotion analysis model training method and device |
| CN112417116A (en) * | 2020-11-18 | 2021-02-26 | 四川长虹电器股份有限公司 | Question understanding model training method and system based on few-sample corpus |
| CN112417145A (en) * | 2019-08-23 | 2021-02-26 | 武汉Tcl集团工业研究院有限公司 | Text multi-classification model generation method, text processing device and medium |
| CN112417886A (en) * | 2020-11-20 | 2021-02-26 | 平安普惠企业管理有限公司 | Intention entity information extraction method and device, computer equipment and storage medium |
| CN112528679A (en) * | 2020-12-17 | 2021-03-19 | 科大讯飞股份有限公司 | Intention understanding model training method and device and intention understanding method and device |
| CN112749565A (en) * | 2019-10-31 | 2021-05-04 | 华为终端有限公司 | Semantic recognition method and device based on artificial intelligence and semantic recognition equipment |
| CN112765356A (en) * | 2021-01-29 | 2021-05-07 | 苏州思必驰信息科技有限公司 | Training method and system of multi-intention recognition model |
| CN112906370A (en) * | 2019-12-04 | 2021-06-04 | 马上消费金融股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN112905893A (en) * | 2021-03-22 | 2021-06-04 | 北京百度网讯科技有限公司 | Training method of search intention recognition model, search intention recognition method and device |
| CN113010667A (en) * | 2019-12-20 | 2021-06-22 | 王道维 | Training method for machine learning decision model by using natural language corpus |
| CN113094475A (en) * | 2021-06-08 | 2021-07-09 | 成都晓多科技有限公司 | Dialog intention recognition system and method based on context attention flow |
| CN113254617A (en) * | 2021-06-11 | 2021-08-13 | 成都晓多科技有限公司 | Message intention identification method and system based on pre-training language model and encoder |
| WO2021164244A1 (en) * | 2020-02-18 | 2021-08-26 | 百度在线网络技术（北京）有限公司 | Voice interaction method and apparatus, device and computer storage medium |
| CN113486178A (en) * | 2021-07-12 | 2021-10-08 | 恒安嘉新（北京）科技股份公司 | Text recognition model training method, text recognition device and medium |
| CN114997154A (en) * | 2022-05-11 | 2022-09-02 | 北京科东电力控制系统有限责任公司 | Automatic construction method and system for speaker-to-speaker robot corpus |
| US11507854B2 (en) * | 2019-01-29 | 2022-11-22 | Ricoh Company, Ltd. | Method and apparatus for recognizing intention, and non-transitory computer-readable recording medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106156003A (en) * | 2016-06-30 | 2016-11-23 | 北京大学 | A kind of question sentence understanding method in question answering system |
| CN106815566A (en) * | 2016-12-29 | 2017-06-09 | 天津中科智能识别产业技术研究院有限公司 | A kind of face retrieval method based on multitask convolutional neural networks |
| US20170193099A1 (en) * | 2015-12-31 | 2017-07-06 | Quixey, Inc. | Machine Identification of Grammar Rules That Match a Search Query |
| CN107464559A (en) * | 2017-07-11 | 2017-12-12 | 中国科学院自动化研究所 | Joint forecast model construction method and system based on Chinese rhythm structure and stress |
| CN107704866A (en) * | 2017-06-15 | 2018-02-16 | 清华大学 | Multitask Scene Semantics based on new neural network understand model and its application |
-
2018
- 2018-06-29 CN CN201810694995.4A patent/CN108920622B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170193099A1 (en) * | 2015-12-31 | 2017-07-06 | Quixey, Inc. | Machine Identification of Grammar Rules That Match a Search Query |
| CN106156003A (en) * | 2016-06-30 | 2016-11-23 | 北京大学 | A kind of question sentence understanding method in question answering system |
| CN106815566A (en) * | 2016-12-29 | 2017-06-09 | 天津中科智能识别产业技术研究院有限公司 | A kind of face retrieval method based on multitask convolutional neural networks |
| CN107704866A (en) * | 2017-06-15 | 2018-02-16 | 清华大学 | Multitask Scene Semantics based on new neural network understand model and its application |
| CN107464559A (en) * | 2017-07-11 | 2017-12-12 | 中国科学院自动化研究所 | Joint forecast model construction method and system based on Chinese rhythm structure and stress |
Non-Patent Citations (2)
| Title |
|---|
| XIAODONG ZHANG等: "A Joint Model of Intent Determination and Slot Filling for Spoken Language Understanding", 《PROCEEDINGS OF THE TWENTY-FIFTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE》 * |
| 向阳: "《模型智能构造系统研究》", 31 August 2000 * |
Cited By (93)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292752B (en) * | 2018-12-06 | 2023-05-12 | 北京嘀嘀无限科技发展有限公司 | User intention recognition method and device, electronic equipment and storage medium |
| CN111292752A (en) * | 2018-12-06 | 2020-06-16 | 北京嘀嘀无限科技发展有限公司 | User intention identification method and device, electronic equipment and storage medium |
| CN109741751A (en) * | 2018-12-11 | 2019-05-10 | 上海交通大学 | Intension recognizing method and device towards intelligent sound control |
| CN109545186A (en) * | 2018-12-16 | 2019-03-29 | 初速度（苏州）科技有限公司 | A kind of speech recognition training system and method |
| CN109545186B (en) * | 2018-12-16 | 2022-05-27 | 魔门塔(苏州)科技有限公司 | Speech recognition training system and method |
| CN111354354A (en) * | 2018-12-20 | 2020-06-30 | 深圳市优必选科技有限公司 | Training method and device based on semantic recognition and terminal equipment |
| CN111354354B (en) * | 2018-12-20 | 2024-02-09 | 深圳市优必选科技有限公司 | Training method, training device and terminal equipment based on semantic recognition |
| CN109857844B (en) * | 2018-12-29 | 2022-01-14 | 北京三快在线科技有限公司 | Intent recognition method and device based on ordering dialogue text and electronic equipment |
| CN109857844A (en) * | 2018-12-29 | 2019-06-07 | 北京三快在线科技有限公司 | Intension recognizing method, device, electronic equipment based on dialog text of ordering |
| CN111428872A (en) * | 2019-01-10 | 2020-07-17 | 维萨国际服务协会 | Systems, methods, and computer program products for incorporating knowledge from more complex models into simpler models |
| US12039458B2 (en) | 2019-01-10 | 2024-07-16 | Visa International Service Association | System, method, and computer program product for incorporating knowledge from more complex models in simpler models |
| CN109857868A (en) * | 2019-01-25 | 2019-06-07 | 北京奇艺世纪科技有限公司 | Model generating method, file classification method, device and computer readable storage medium |
| US11507854B2 (en) * | 2019-01-29 | 2022-11-22 | Ricoh Company, Ltd. | Method and apparatus for recognizing intention, and non-transitory computer-readable recording medium |
| CN109816111A (en) * | 2019-01-29 | 2019-05-28 | 北京金山数字娱乐科技有限公司 | Reading understands model training method and device |
| CN111563208A (en) * | 2019-01-29 | 2020-08-21 | 株式会社理光 | Intention identification method and device and computer readable storage medium |
| CN109871485A (en) * | 2019-02-13 | 2019-06-11 | 北京航空航天大学 | A kind of personalized recommendation method and device |
| CN109871485B (en) * | 2019-02-13 | 2022-04-05 | 北京航空航天大学 | Personalized recommendation method and device |
| CN111680514A (en) * | 2019-02-25 | 2020-09-18 | 北京猎户星空科技有限公司 | Information processing and model training method, device, equipment and storage medium |
| CN111612025B (en) * | 2019-02-25 | 2023-12-12 | 北京嘀嘀无限科技发展有限公司 | Description model training method, text description device and electronic equipment |
| CN111680514B (en) * | 2019-02-25 | 2024-03-01 | 北京猎户星空科技有限公司 | Information processing and model training method, device, equipment and storage medium |
| CN111612025A (en) * | 2019-02-25 | 2020-09-01 | 北京嘀嘀无限科技发展有限公司 | Description model training method, text description device and electronic equipment |
| CN109933663A (en) * | 2019-02-26 | 2019-06-25 | 上海凯岸信息科技有限公司 | Intention assessment algorithm based on embedding method |
| WO2020211008A1 (en) * | 2019-04-17 | 2020-10-22 | 深圳市欢太科技有限公司 | Speech recognition method and apparatus, storage medium and electronic device |
| CN110147535A (en) * | 2019-04-18 | 2019-08-20 | 平安科技（深圳）有限公司 | Similar Text generation method, device, equipment and storage medium |
| CN110085308B (en) * | 2019-04-23 | 2022-02-25 | 挂号网（杭州）科技有限公司 | Diagnosis and treatment department classification method based on fusion deep learning |
| CN110085308A (en) * | 2019-04-23 | 2019-08-02 | 挂号网（杭州）科技有限公司 | A kind of diagnosis and treatment department classification method based on fusion deep learning |
| CN110287283B (en) * | 2019-05-22 | 2023-08-01 | 中国平安财产保险股份有限公司 | Intention model training method, intention recognition method, device, equipment and medium |
| CN110287283A (en) * | 2019-05-22 | 2019-09-27 | 中国平安财产保险股份有限公司 | Intent model training method, intension recognizing method, device, equipment and medium |
| CN110134969B (en) * | 2019-05-27 | 2023-07-14 | 北京奇艺世纪科技有限公司 | Entity identification method and device |
| CN110134969A (en) * | 2019-05-27 | 2019-08-16 | 北京奇艺世纪科技有限公司 | A kind of entity recognition method and device |
| WO2020237869A1 (en) * | 2019-05-31 | 2020-12-03 | 平安科技（深圳）有限公司 | Question intention recognition method and apparatus, computer device, and storage medium |
| CN110222339A (en) * | 2019-06-05 | 2019-09-10 | 深圳市思迪信息技术股份有限公司 | Based on the intension recognizing method and device for improving XGBoost algorithm |
| CN110222339B (en) * | 2019-06-05 | 2023-04-28 | 深圳市思迪信息技术股份有限公司 | Intention recognition method and device based on improved XGBoost algorithm |
| CN110334344A (en) * | 2019-06-13 | 2019-10-15 | 腾讯科技（深圳）有限公司 | A kind of semanteme intension recognizing method, device, equipment and storage medium |
| CN110334344B (en) * | 2019-06-13 | 2024-05-14 | 腾讯科技（深圳）有限公司 | Semantic intention recognition method, device, equipment and storage medium |
| CN112185351A (en) * | 2019-07-05 | 2021-01-05 | 北京猎户星空科技有限公司 | Voice signal processing method and device, electronic equipment and storage medium |
| CN112185351B (en) * | 2019-07-05 | 2024-05-24 | 北京猎户星空科技有限公司 | Voice signal processing method and device, electronic equipment and storage medium |
| CN110516035A (en) * | 2019-07-05 | 2019-11-29 | 同济大学 | A kind of man-machine interaction method and system of mixing module |
| CN110610001B (en) * | 2019-08-12 | 2024-01-23 | 大箴(杭州)科技有限公司 | Short text integrity recognition method, device, storage medium and computer equipment |
| CN110610001A (en) * | 2019-08-12 | 2019-12-24 | 大箴(杭州)科技有限公司 | Short text integrity identification method and device, storage medium and computer equipment |
| CN112417145A (en) * | 2019-08-23 | 2021-02-26 | 武汉Tcl集团工业研究院有限公司 | Text multi-classification model generation method, text processing device and medium |
| CN110532368A (en) * | 2019-09-04 | 2019-12-03 | 深圳前海达闼云端智能科技有限公司 | Question answering method, electronic equipment and computer readable storage medium |
| CN110619047A (en) * | 2019-09-16 | 2019-12-27 | 出门问问信息科技有限公司 | Method and device for constructing natural language model and readable storage medium |
| CN110619047B (en) * | 2019-09-16 | 2022-09-02 | 出门问问信息科技有限公司 | Method and device for constructing natural language model and readable storage medium |
| CN110674253A (en) * | 2019-09-23 | 2020-01-10 | 出门问问信息科技有限公司 | Semantic processing method and equipment |
| CN110807332B (en) * | 2019-10-30 | 2024-02-27 | 腾讯科技（深圳）有限公司 | Training method, semantic processing method, device and storage medium for semantic understanding model |
| US11967312B2 (en) | 2019-10-30 | 2024-04-23 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for training semantic understanding model, electronic device, and storage medium |
| CN110807332A (en) * | 2019-10-30 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Training method of semantic understanding model, semantic processing method, semantic processing device and storage medium |
| CN112749565A (en) * | 2019-10-31 | 2021-05-04 | 华为终端有限公司 | Semantic recognition method and device based on artificial intelligence and semantic recognition equipment |
| WO2021082570A1 (en) * | 2019-10-31 | 2021-05-06 | 华为技术有限公司 | Artificial intelligence-based semantic identification method, device, and semantic identification apparatus |
| CN110852108A (en) * | 2019-11-11 | 2020-02-28 | 中山大学 | Joint training method, apparatus and medium for entity recognition and entity disambiguation |
| CN110956018A (en) * | 2019-11-22 | 2020-04-03 | 腾讯科技（深圳）有限公司 | Training method of text processing model, text processing method, text processing device and storage medium |
| CN110956018B (en) * | 2019-11-22 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Training method of text processing model, text processing method, text processing device and storage medium |
| CN112906370A (en) * | 2019-12-04 | 2021-06-04 | 马上消费金融股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN112906370B (en) * | 2019-12-04 | 2022-12-20 | 马上消费金融股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN113010667A (en) * | 2019-12-20 | 2021-06-22 | 王道维 | Training method for machine learning decision model by using natural language corpus |
| CN111198938A (en) * | 2019-12-26 | 2020-05-26 | 深圳市优必选科技股份有限公司 | Sample data processing method, sample data processing device and electronic equipment |
| CN111198938B (en) * | 2019-12-26 | 2023-12-01 | 深圳市优必选科技股份有限公司 | Sample data processing method, sample data processing device and electronic equipment |
| CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN111209383B (en) * | 2020-01-06 | 2023-04-07 | 广州小鹏汽车科技有限公司 | Method and device for processing multi-turn dialogue, vehicle, and storage medium |
| CN111209383A (en) * | 2020-01-06 | 2020-05-29 | 广州小鹏汽车科技有限公司 | Method and device for processing multi-turn dialogue, vehicle, and storage medium |
| WO2021164244A1 (en) * | 2020-02-18 | 2021-08-26 | 百度在线网络技术（北京）有限公司 | Voice interaction method and apparatus, device and computer storage medium |
| US11978447B2 (en) | 2020-02-18 | 2024-05-07 | Baidu Online Network Technology (Beijing) Co., Ltd. | Speech interaction method, apparatus, device and computer storage medium |
| CN111309915A (en) * | 2020-03-03 | 2020-06-19 | 爱驰汽车有限公司 | Method, system, device and storage medium for training natural language of joint learning |
| CN111667066B (en) * | 2020-04-23 | 2024-06-11 | 北京旷视科技有限公司 | Training method and device of network model, character recognition method and device and electronic equipment |
| CN111667066A (en) * | 2020-04-23 | 2020-09-15 | 北京旷视科技有限公司 | Network model training and character recognition method and device and electronic equipment |
| CN111597804B (en) * | 2020-05-15 | 2023-03-10 | 腾讯科技（深圳）有限公司 | Method and related device for training entity recognition model |
| CN111597804A (en) * | 2020-05-15 | 2020-08-28 | 腾讯科技（深圳）有限公司 | Entity recognition model training method and related device |
| CN111680517A (en) * | 2020-06-10 | 2020-09-18 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for training a model |
| CN111680517B (en) * | 2020-06-10 | 2023-05-16 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for training model |
| CN111538894A (en) * | 2020-06-19 | 2020-08-14 | 腾讯科技（深圳）有限公司 | Query feedback method and device, computer equipment and storage medium |
| CN111737416B (en) * | 2020-06-29 | 2022-08-19 | 重庆紫光华山智安科技有限公司 | Case processing model training method, case text processing method and related device |
| CN111737416A (en) * | 2020-06-29 | 2020-10-02 | 重庆紫光华山智安科技有限公司 | Case processing model training method, case text processing method and related device |
| CN111785350A (en) * | 2020-06-30 | 2020-10-16 | 易联众信息技术股份有限公司 | Information extraction method, application, device and medium |
| CN111949784A (en) * | 2020-08-14 | 2020-11-17 | 中国工商银行股份有限公司 | Outbound method and device based on intention recognition |
| CN112102832A (en) * | 2020-09-18 | 2020-12-18 | 广州小鹏汽车科技有限公司 | Speech recognition method, speech recognition device, server and computer-readable storage medium |
| CN112183631A (en) * | 2020-09-28 | 2021-01-05 | 云知声智能科技股份有限公司 | Method and terminal for establishing intention classification model |
| CN112183631B (en) * | 2020-09-28 | 2024-01-12 | 云知声智能科技股份有限公司 | Method and terminal for establishing intention classification model |
| CN112256826A (en) * | 2020-10-19 | 2021-01-22 | 网易（杭州）网络有限公司 | Emotion analysis method, evaluation method and emotion analysis model training method and device |
| CN112417116A (en) * | 2020-11-18 | 2021-02-26 | 四川长虹电器股份有限公司 | Question understanding model training method and system based on few-sample corpus |
| CN112417116B (en) * | 2020-11-18 | 2022-03-15 | 四川长虹电器股份有限公司 | Question understanding model training method and system based on few-sample corpus |
| CN112417886A (en) * | 2020-11-20 | 2021-02-26 | 平安普惠企业管理有限公司 | Intention entity information extraction method and device, computer equipment and storage medium |
| CN112417886B (en) * | 2020-11-20 | 2024-08-27 | 西藏纳柯电子科技有限公司 | Method, device, computer equipment and storage medium for extracting intention entity information |
| CN112528679B (en) * | 2020-12-17 | 2024-02-13 | 科大讯飞股份有限公司 | Method and device for training intention understanding model, and method and device for intention understanding |
| CN112528679A (en) * | 2020-12-17 | 2021-03-19 | 科大讯飞股份有限公司 | Intention understanding model training method and device and intention understanding method and device |
| CN112765356A (en) * | 2021-01-29 | 2021-05-07 | 苏州思必驰信息科技有限公司 | Training method and system of multi-intention recognition model |
| CN112905893A (en) * | 2021-03-22 | 2021-06-04 | 北京百度网讯科技有限公司 | Training method of search intention recognition model, search intention recognition method and device |
| CN112905893B (en) * | 2021-03-22 | 2024-01-12 | 北京百度网讯科技有限公司 | Training method of search intention recognition model, search intention recognition method and device |
| CN113094475A (en) * | 2021-06-08 | 2021-07-09 | 成都晓多科技有限公司 | Dialog intention recognition system and method based on context attention flow |
| CN113254617A (en) * | 2021-06-11 | 2021-08-13 | 成都晓多科技有限公司 | Message intention identification method and system based on pre-training language model and encoder |
| CN113486178B (en) * | 2021-07-12 | 2023-12-01 | 恒安嘉新（北京）科技股份公司 | Text recognition model training method, text recognition method, device and medium |
| CN113486178A (en) * | 2021-07-12 | 2021-10-08 | 恒安嘉新（北京）科技股份公司 | Text recognition model training method, text recognition device and medium |
| CN114997154A (en) * | 2022-05-11 | 2022-09-02 | 北京科东电力控制系统有限责任公司 | Automatic construction method and system for speaker-to-speaker robot corpus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108920622B (en) | 2021-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108920622A (en) | A kind of training method of intention assessment, training device and identification device | |
| JP6955580B2 (en) | Document summary automatic extraction method, equipment, computer equipment and storage media | |
| CN108711420A (en) | Multilingual hybrid model foundation, data capture method and device, electronic equipment | |
| CN109840287A (en) | A kind of cross-module state information retrieval method neural network based and device | |
| CN104115221B (en) | Changed based on Text To Speech and semantic audio human interaction proof | |
| CN109523994A (en) | A kind of multitask method of speech classification based on capsule neural network | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| CN108711421A (en) | A kind of voice recognition acoustic model method for building up and device and electronic equipment | |
| CN111159395A (en) | Chart neural network-based rumor standpoint detection method and device and electronic equipment | |
| CN107423363A (en) | Art generation method, device, equipment and storage medium based on artificial intelligence | |
| CN111625658A (en) | Voice interaction method, device and equipment based on knowledge graph and storage medium | |
| CN109213856A (en) | Semantic recognition method and system | |
| CN108073565A (en) | The method and apparatus and machine translation method and equipment of words criterion | |
| CN108491515B (en) | Sentence pair matching degree prediction method for campus psychological consultation | |
| CN110223675A (en) | The screening technique and system of training text data for speech recognition | |
| CN106682387A (en) | Method and device used for outputting information | |
| CN107437417A (en) | Based on speech data Enhancement Method and device in Recognition with Recurrent Neural Network speech recognition | |
| CN107679225A (en) | A kind of reply generation method based on keyword | |
| CN112559749A (en) | Intelligent matching method and device for teachers and students in online education and storage medium | |
| Zhou et al. | ICRC-HIT: A deep learning based comment sequence labeling system for answer selection challenge | |
| CN109543176A (en) | A kind of abundant short text semantic method and device based on figure vector characterization | |
| CN111597341A (en) | Document level relation extraction method, device, equipment and storage medium | |
| CN110532393A (en) | Text handling method, device and its intelligent electronic device | |
| CN109859747A (en) | Voice interactive method, equipment and storage medium | |
| CN113392197A (en) | Question-answer reasoning method and device, storage medium and electronic 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |