CN108959551A - Method for digging, device, storage medium and the terminal device of neighbour's semanteme - Google Patents
Method for digging, device, storage medium and the terminal device of neighbour's semanteme Download PDFInfo
- Publication number
- CN108959551A CN108959551A CN201810712820.1A CN201810712820A CN108959551A CN 108959551 A CN108959551 A CN 108959551A CN 201810712820 A CN201810712820 A CN 201810712820A CN 108959551 A CN108959551 A CN 108959551A
- Authority
- CN
- China
- Prior art keywords
- neighbour
- semanteme
- sentence
- word
- natural sentence
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The present invention proposes method for digging, device, storage medium and the terminal device of a kind of neighbour's semanteme, wherein the described method includes: obtaining the natural sentence of neighbour's semanteme to be excavated；According to variation from the encoder of encoding model, the natural sentence is encoded, obtains the hidden vector of the natural sentence；Wherein, the hidden vector is located at the average point of encoder output distribution；Neighbour's semantic search is carried out to the hidden vector, obtains the hidden vector of neighbour that there is neighbour's semanteme with the hidden vector；And the hidden vector of neighbour is decoded from coding solution to model code device according to variation, obtain the sentence that there is neighbour's semanteme with the natural sentence.Using the present invention, the approximate sentence of a large amount of semantic syntaxes can be excavated.
Description
Technical field
The present invention relates to field of computer technology more particularly to method for digging, device, the storage mediums of a kind of neighbour semanteme
And terminal device.
Background technique
In NLP (Neuro-Linguistic Programming, neural LISP program LISP) field, the place of natural language
Reason generally can carry out vectorization expression to natural language, then carry out subsequent processing again.Vectorization expression to natural language
There is mature solution, but the semanteme that natural language how is expressed in vectorization expression does not have an effective solution yet
Certainly scheme.Generally, the technical solution of neighbour's justice excavation of natural language processing includes:
1, term vector is carried out using word2vec (word to vector, for generating the correlation model of term vector) to add
With.Unsupervised training is carried out to corpus using word2vec model, obtains the vectorization expression of each word in nature sentence.Right
When natural sentence is modeled, the corresponding vector of each word is summed it up, the vectorization for obtaining whole word indicates.Then, to this to
The point in the space of amount carries out neighbor search, to obtain the sentence of semantic similarity.
2, it is based on LSTM (Long Short-Term Memory, shot and long term memory network) language model, is modeled to obtain
The probability of every a word in corpus.After model training completion, the sentence of sentence collection to be predicted is calculated in training using model
The probability occurred is concentrated, probability is higher, then it is assumed that the sentence is more close with the probability of sentence to be predicted.
But above scheme has following shortcoming:
1, it for scheme 1, is expressed due to the mode for taking term vector to sum it up, obtained expression does not include sentence
The information between word order in structure, sentence.Although the word of two words be it is identical, it is semantic different, also can be by
Word2vec model is mapped to the same point in space.For example, " I likes you " is two different semantic with " you like me "
Sentence, but word2vec model can convert it into identical vector expression.
2, for scheme 2, LSTM model does not carry out vectorization expression to natural sentence, can only utilize LSTM model
Calculate its probability occurred in training corpus, that is to say, that calculate each trained sentence in nature sentence and training corpus
Between similarity, rather than the similarity between any one sentence.It is very limited so to carry out text mining.
Summary of the invention
The embodiment of the present invention provides method for digging, device, storage medium and the terminal device of a kind of neighbour's semanteme, to solve
Or alleviate above one or more technical problems in the prior art.
In a first aspect, the embodiment of the invention provides a kind of method for digging of neighbour's semanteme, comprising:
Obtain the natural sentence of neighbour's semanteme to be excavated；
According to variation from the encoder of encoding model, the natural sentence is encoded, obtains the natural sentence
Hidden vector；Wherein, the hidden vector is located at the average point of encoder output distribution；
Neighbour's semantic search is carried out to the hidden vector, obtain the neighbour that there is neighbour's semanteme with the hidden vector it is hidden to
Amount；And
According to variation from coding solution to model code device, the hidden vector of neighbour is decoded, obtains and has with the natural sentence
There is the sentence of neighbour's semanteme.
With reference to first aspect, it in the first embodiment of first aspect, carries out encoding it to the natural sentence
Before further include:
Word cutting is carried out to the natural sentence, obtains the word of the composition natural sentence；
According to preset dictionary, the word for forming the natural sentence is changed into corresponding numerical value；And
According to term vector model, vector conversion is carried out to the corresponding each integer of each word for forming the natural sentence, is obtained
Obtain the vector expression of the natural sentence.
With reference to first aspect, in second of embodiment of first aspect, the method also includes:
The initial coefficients of the variation from the relative entropy of encoding model are set as zero；Wherein, the relative entropy is for measuring
The distance between the coding distribution of the encoder and prior distribution, the prior distribution is the ideal coding of the encoder
Distribution；And
According to training corpus, iteration is trained from encoding model to the variation；Wherein, with the trained iteration
Number increases, and the coefficient of the relative entropy rises along sine curve until the numerical value of the coefficient is one.
Second of embodiment with reference to first aspect, in the third embodiment of first aspect, the method is also
Include:
According to preset word loss ratio, the word in the training corpus is replaced using noise character.
Second of embodiment with reference to first aspect, in the 4th kind of embodiment of first aspect, the method is also
Include:
During the trained iteration, judge whether the relative entropy is less than threshold limit；And
When the relative entropy is less than the threshold limit, the relative entropy is deleted from the variation from encoding model
It removes.
With reference to first aspect or its any embodiment, in the 5th kind of embodiment of first aspect, the method is also
Include:
According to the encoder, two natural sentences are encoded respectively, it is corresponding to obtain described two natural sentences
Hidden vector；And
According to the corresponding hidden vector of described two nature sentences, the similarity of described two natural sentences is calculated.
Second aspect, the embodiment of the present invention provide a kind of excavating gear of neighbour's semanteme, comprising:
Natural sentence receiving module, for receiving the natural sentence of neighbour's semanteme to be excavated；
Coding obtains module, for the encoder according to variation from encoding model, encodes, obtains to the natural sentence
Obtain the hidden vector of the natural sentence；Wherein, the hidden vector is located at the average point of encoder output distribution；
Neighbour's semantic search module obtains and described hidden to measurer for carrying out neighbour's semantic search to the hidden vector
There is the hidden vector of neighbour of neighbour's semanteme；And
Decoding obtains module, for, from coding solution to model code device, being decoded to the hidden vector of the neighbour according to variation,
Obtain the sentence that there is neighbour's semanteme with the natural sentence.
In conjunction with second aspect, in the first embodiment of second aspect, described device further includes
Sentence word cutting module, for carrying out word cutting to the natural sentence before encoding to the natural sentence,
Obtain the word of the composition natural sentence；
Word number conversion module, for according to preset dictionary, the word for forming the natural sentence to be changed into accordingly
Numerical value；And
Vector conversion module, for according to term vector model, each word of the natural sentence described to composition to be corresponding each whole
Number carries out vector conversion, obtains the vector expression of the natural sentence.
In conjunction with second aspect, in second of embodiment of second aspect, described device further include:
Relative entropy initialization module is set as zero for the initial coefficients by the variation from the relative entropy of encoding model；Its
In, the relative entropy is used to measure the distance between coding distribution and prior distribution of the encoder, and the prior distribution is
The ideal coding of the encoder is distributed；And
Model training iteration module, for being trained iteration from encoding model to the variation according to training corpus；Its
In, as the number of the trained iteration increases, the coefficient of the relative entropy rises the number until the coefficient along sine curve
Value is one.
In conjunction with second of embodiment of second aspect, in the third embodiment of second aspect, described device is also
Include:
Noise character replacement module, for replacing the training corpus using noise character according to preset word loss ratio
In word.
In conjunction with second of embodiment of second aspect, in the 4th kind of embodiment of second aspect, described device is also
Include:
Relative entropy judgment module, for judging whether the relative entropy is less than limitation during the trained iteration
Threshold value；And
Relative entropy removing module, for when the relative entropy is less than the threshold limit, by the relative entropy from described
Variation is deleted from encoding model.
The function of described device can also execute corresponding software realization by hardware realization by hardware.It is described
Hardware or software include one or more modules corresponding with above-mentioned function.
It include processor and memory, the memory in the mining structure of neighbour's semanteme in a possible design
Excavating gear for neighbour's semanteme executes the excavation program of neighbour's semanteme in above-mentioned first aspect, the processor is configured to
For executing the program stored in the memory.The excavating gear of neighbour's semanteme can also include communication interface, be used for
The excavating gear and other equipment or communication of neighbour's semanteme.
The third aspect, the embodiment of the present invention also provide a kind of computer readable storage medium, the excavation for neighbour's semanteme
Computer software instructions used in device, involved by the method for digging including neighbour's semanteme for executing above-mentioned first aspect
And program.
One of technical solution in above-mentioned technical proposal have the following advantages that or the utility model has the advantages that
The embodiment of the present invention carries out vectorization expression to natural sentence using variation self-encoding encoder, is then based on vectorization
The space of hidden vector carries out neighbour's semantic search, can excavate the approximate sentence of a large amount of semantic syntaxes.
One of technical solution in above-mentioned technical proposal have the following advantages that or the utility model has the advantages that
The embodiment of the present invention realizes the coding of nature sentence based on variation self-encoding encoder, a discrete information can be turned
It has been melted into a continuous information.It is when vector expresses nature sentence, it may be considered that suitable to the word in the structure of sentence, sentence
Between sequence.In continuous space, the same or similar sentence of syntactic-semantic is generally mapped in similar space.
Above-mentioned general introduction is merely to illustrate that the purpose of book, it is not intended to be limited in any way.Except foregoing description
Schematical aspect, except embodiment and feature, by reference to attached drawing and the following detailed description, the present invention is further
Aspect, embodiment and feature, which will be, to be readily apparent that.
Detailed description of the invention
In the accompanying drawings, unless specified otherwise herein, otherwise indicate the same or similar through the identical appended drawing reference of multiple attached drawings
Component or element.What these attached drawings were not necessarily to scale.It should be understood that these attached drawings depict only according to the present invention
Disclosed some embodiments, and should not serve to limit the scope of the present invention.
Fig. 1 is the flow diagram of one embodiment of the method for digging of neighbour's semanteme provided by the invention；
Fig. 2 is the flow diagram of one embodiment provided by the invention that pretreated process is carried out to natural sentence；
Fig. 3 is the flow diagram of one embodiment of model training process provided by the invention；
Fig. 4 is the flow diagram of one embodiment of the calculating process of natural statement similarity provided by the invention；
Fig. 5 is the schematic diagram using exemplary one embodiment of the method for digging of neighbour's semanteme provided by the invention；
Fig. 6 is the structural schematic diagram of one embodiment of the excavating gear of neighbour's semanteme provided by the invention；
Fig. 7 is the structural schematic diagram of another embodiment of the excavating gear of neighbour's semanteme provided by the invention；
Fig. 8 is the structural schematic diagram of one embodiment of terminal device provided by the invention.
Specific embodiment
Hereinafter, certain exemplary embodiments are simply just described.As one skilled in the art will recognize that
Like that, without departing from the spirit or scope of the present invention, described embodiment can be modified by various different modes.
Therefore, attached drawing and description are considered essentially illustrative rather than restrictive.
Referring to Fig. 1, can be applied to terminal the embodiment of the invention provides a kind of method for digging of neighbour's semanteme and set
It is standby.Terminal device may include smart phone, plate, computer etc..The present embodiment includes step S100 to step S400, specifically
It is as follows:
S100 obtains the natural sentence of neighbour's semanteme to be excavated.
In the present embodiment, natural sentence refers to one section of sentence with syntax logic, or with a semantic language
Sentence.For example, " I and I junior partner is stupefied ", " life being treasured, far from drugs " etc..
S200 encodes natural sentence according to variation from the encoder of encoding model, obtain nature sentence it is hidden to
Amount；Wherein, hidden vector is located at the average point of encoder output distribution.
In the present embodiment, the hidden vector that variation is weaved into from coding (VariationalAutoEncoder, VAE) model
Space have connection, can be micro- feature, can to complete sentence carry out vectorization expression, accurately express sentence semanteme
And syntax.The selection of variation from the encoder of encoding model does not limit specifically, can choose neural network model to text
It is encoded.Such as: long short-term memory Recognition with Recurrent Neural Network (LSTM Recurrent Neural Network, LSTM-RNN),
It is considered that the clause information of sentence, is capable of providing the expression more richer than the simplicial map of word2vec model.
Since VAE is a kind of frame for generating model, entire VAE frame can still be protected after introducing LSTM-RNN
The mode of unsupervised learning is held, so as to carry out the training of model using a large amount of unsupervised data in internet, to generate
Corresponding model.The encoder of trained VAE model, can complete the coding to natural sentence.The encoder of model is being compiled
Mean value and standard deviation can be exported when code output, the hidden vector of average point at this time be the natural sentence of encoder input it is hidden to
Amount.
S300 carries out neighbour's semantic search to hidden vector, obtains the hidden vector of neighbour for having neighbour's semanteme with hidden vector.
In the present embodiment, natural sentence is usually made of word, and from the point of view of the input information of model, natural sentence is one
A discrete information, the hidden vector obtained after the coding of the encoder of VAE model are the vectors in a real number space, such as
This, a discrete information can be converted to a continuous information by the present embodiment.It, can be from conveniently in continuous space
Ground carries out neighbour's semantic search, such as is scanned for using neural network, achievees the effect that relevant mining.For example, natural language
Sentence A1 " life being treasured, far from gambling " obtains hidden vector (1,1,1) after the coding of VAE model, carries out to hidden vector (1,1,1)
The search of neighbour's semanteme obtains the hidden vector of one of neighbour (1,1,1,0.9).In this way, semantic similarity can be from the two
The distance of hidden vector may determine that out.
S400 is decoded the hidden vector of neighbour according to variation from coding solution to model code device, obtains and has with natural sentence
There is the sentence of neighbour's semanteme.
Example is connected, the hidden vector of neighbour (1,1,1,0.9) can be input to VAE solution to model code device, acquisition is decoded as certainly
Right sentence A2 " life being treasured, far from football lottery ".
In the present embodiment, natural sentence is encoded using VAE model to obtain hidden vector, it is contemplated that natural language
The information such as clause, the semanteme of sentence.Clause, semantic same or similar sentence are mapped to the space apart from hidden vector more neighbour
In, neighbour's semantic search can be carried out to the vector space of hidden vector, obtain the sentence of a large amount of clause semantic similarities.
In one possible implementation, as shown in Fig. 2, before encoding to natural sentence, method further includes
Pretreated process is carried out to natural sentence, may include step S510 to step S530, can be such that
S510 carries out word cutting to natural sentence, obtains the word of composition nature sentence.
By taking natural sentence A1 " treasure life, far from gambling " as an example, the word after word cutting include: treasure, life, it is separate and
Gambling.During word cutting, noise word, ambiguity word, unrecognized word or character etc. can also be filtered out.
The word for forming nature sentence is changed into corresponding numerical value according to preset dictionary by S520.
It in the present embodiment, can be according to common dictionary, dedicated domain dictionary or custom dictionaries come for the word in dictionary
The setting numerical value of language, with formed current embodiment require that dictionary.The word of natural sentence can be changed into numerical value such as integer, real numbers
To indicate.
S530 carries out vector conversion to the corresponding each integer of each word of composition nature sentence, obtains according to term vector model
Vector expression derived from right sentence.
In the present embodiment, for natural sentence A1 " life being treasured, far from gambling ", it is assumed that it is converted by vector,
Obtain the vector a1 (1,2,3,4) of nature sentence A1.This vector value is only to illustrate.
In the present embodiment, VAE is a set of unsupervised production machine learning frame, and basic theories is shown below:
First item in formula after equal signIt is the coding point of the natural sentence of input for relative entropy
The distance between cloth and prior distribution, also referred to as KL distance, the output distribution that can have measured encoder are uniformly distributed with center
Tightness.Section 2 in formula after equal signFor reconstructed error, nature is described
The lower information Loss Rate of the sentence after encoding further decoding, numerical value the better.The low presentation code quality of numerical value is better.
It is the posterior probability (coding distribution) of the natural sentence of input.Prior distributionIt is distributed for the ideal coding of encoder.It is output distribution probability of the nature sentence after encoding further decoding, p (x) is the output distribution of target nature sentence
Probability.X is the natural sentence of input,It is the hidden vector of nature sentence.
In the present embodiment, the system of relative entropy is properly termed as tightness coefficient.Adjusting to the coefficient of relative entropy, can be with
The tightness for controlling the latent space of encoder output, is reasonably encoded.Such as can take tightness coefficient is 0.8.
In the present embodiment, VAE model is initially to be applied to image to generate field, after moving to text generation field,
VAE model there are problems that being easy to happen posteriority collapsing, for example, KL causes coding result to be all zero apart from too high.It was training
It collapses in journey once posteriority occurs, VAE frame will be degenerated to generic language model, and can not carry out hidden vector coding.In order to solve
Such case, the present embodiment propose the mode of model training as shown in Figure 3 come the appearance for avoiding posteriority from collapsing.The present embodiment pair
The training process of VAE model may include step S610 and step S620, as follows:
The initial coefficients of variation from the relative entropy of encoding model are set as zero by S610.
S620 is trained iteration from encoding model to variation according to training corpus；Wherein, with time of training iteration
Number increases, and the coefficient of relative entropy rises along sine curve until the numerical value of coefficient is one.
In the present embodiment, at the training initial stage of VAE model, KL distance can not be considered temporarily, training pattern is degenerated
For standard automatic coding machine.After model carries out the training of several wheels, the coefficient of KL distance can be raised with sinusoidal fashion,
It is finally trained to the VAE model of standard, avoids VAE model degradation at generic language model.Wherein, the maximum of the coefficient of KL distance
Value is one.
It in one possible implementation, can be to training as shown in figure 3, the present embodiment is in the training process of model
Being adjusted in corpus, model training process provided by the embodiment further includes step S630, as follows:
S630, according to preset word loss ratio, using the word in noise character replacement training corpus.
It in the present embodiment, can be in order to encourage VAE solution to model code device can be more using the information in hidden vector
Randomly in the training process of decoder, the part word inputted in training corpus is changed into spcial character, such as: .UNK format
The character that text provides, can prevent decoder excessively to rely on the corpus of input.
In one possible implementation, in the training process of above-mentioned model, the present embodiment can also include: to instruct
During practicing iteration, judge whether relative entropy is less than threshold limit；And when relative entropy is less than threshold limit, by relative entropy
It is deleted from variation from encoding model.
In the present embodiment, when posteriority, which collapses, to be occurred, the numerical value of KL distance is close to 0.Therefore, the present embodiment, can be with
Minimum KL distance is set.For example, when KL distance is lower than KL threshold value, it can be no longer to KL during subsequent mould model training
Distance optimizes.
In one possible implementation, the present embodiment can be calculated similar between two natural sentences by hidden vector
Degree, calculates the similarity between quick sentence.As shown in figure 4, the process for calculating similarity may include:
S710 encodes two natural sentences respectively according to encoder, obtain two natural sentences it is corresponding it is hidden to
Amount.
S720 calculates the similarity of two natural sentences according to the corresponding hidden vector of two nature sentences.
It is exemplary, with natural sentence A1 " life being treasured, far from gambling " and natural sentence A2 " life being treasured, far from football lottery "
For, it is assumed that pass through the hidden of the hidden vector " (1,1,1) " of the available natural sentence A1 of the coding of encoder and nature sentence A2
Vector " (1,1,1,0.9) " calculates the distance between hidden vector " (1,1,1) " and hidden vector " (1,1,1,0.9) ", can obtain
Similarity between natural sentence A1 and natural sentence A2.
In embodiments of the present invention, can also arbitrarily be taken a little in the hidden vector space of encoder output, obtain one it is hidden to
Amount.Then, it is decoded by VAE solution to model code device, the corresponding natural sentence of the hidden vector counter can be released, thus this reality
Sentence-making can also be realized by applying example.
As shown in figure 5, it is the method for digging that the embodiment of the present invention provides neighbour's semanteme using exemplary signal
Figure.
Variation is a set of unsupervised production machine learning frame from coding VAE, and basic theories is shown below:
Wherein, in the theoretical model of VAE, each input data x corresponds to point of a vector z in latent space
Cloth.If applying VAE model in text generation field, input data x can be a natural language sentences, and z is still real
Hidden vector in number space.In this way, a discrete information can be converted into a continuous information.In continuous space
Interior, it is approximate close that the present embodiment can easily carry out ANN (Artificial Neural Network, artificial neural network)
Adjacent semantic search can achieve the purpose of relevant mining.If, it is assumed that " life being treasured, far from gambling " corresponding hidden vector
(1,1,1), the hidden vector come out with sentence similar in its clause " life being treasured, far from football lottery " coding be (1,1,1,
0.9) similarity between two sentences, can be judged from the distance between the two vectors.
First item in formula is the priori probability density of input data x and the KL distance of posterior probability density, has measured volume
Code device output distribution and center be uniformly distributed between tightness.In the present embodiment, in this add a coefficient come into
Row is adjusted, and controls the tightness of latent space, available more reasonable coding.For example, taking the numerical value of tightness coefficient is 0.8.
As shown in figure 5, it is the VAE neural network structure of the present embodiment, parametrization skill again can be used and come to hidden
The distribution of vector is modeled.The computer for executing certainty code can realize random point of expression by parameterizing skill again
Cloth can not be by VAE model conversation at computer code if the not skill formula.Specifically, computer generates one
Then this random number is added mean value multiplied by the standard deviation of modeling output, obtains one and obey any height by even distribution random numbers
The random quantity of this distribution.Encoder can be using mature LSTM-RNN network structure, can be to complicated long Series Modeling.
In the present embodiment, VAE is initially applied to image and generates field, and after moving to text generation field, VAE exists
Be easy to happen posteriority collapse the problem of.It collapses in the training process once posteriority occurs, VAE frame will be degenerated to ordinary language
Model can not carry out the generation task of model, can not also carry out the coding of hidden vector.In order to solve this problem, the present invention is real
It applies example and solves the problems, such as that posteriority collapses using following three method.
1, KL distance is degenerated.KL distance is not considered temporarily at the training initial stage of model, and training frame is degenerated for standard certainly
Dynamic code machine.Training pattern is degenerated for standard automatic coding machine.It, can be with sinusoidal bent after model carries out the training of several wheels
Line mode raises the coefficient of KL distance, is finally trained to the VAE model of standard, avoids VAE model degradation at generic language model.
Wherein, the maximum value of the coefficient of KL distance is one.
2, word loss ratio is set.It, can in order to encourage VAE solution to model code device can be more using the information in hidden vector
Randomly in the training process of decoder, to change the part word inputted in training corpus into characteristic character, such as: .UNK lattice
The character that formula text provides, can prevent decoder excessively to rely on the corpus of input.
3, minimum KL distance is limited.When posteriority, which collapses, to be occurred, the numerical value of KL distance is close to 0.Therefore, it can be set
Minimum KL distance.For example, when KL distance is lower than KL threshold value, it can be no longer to KL distance during subsequent mould model training
It optimizes.
As shown in figure 5, trained VAE model can take its encoder section, that is, complete the coding of nature sentence.Model
When coding, it is believed that the corresponding hidden vector of sentence is located at the average point of its encoder output distribution for example, to " I and I
Junior partner be all stupefied " input coding device, encoder can export mean value and standard deviation, at mean value or mean point
Vector is the hidden vector of " I and I junior partner is stupefied ".So as to obtain more reliable encoding efficiency.Then, it is based on
To the hidden vector after coding, ANN approximation neighbour's semantic search is carried out.It can achieve the approximate semantic purpose excavated.
The present embodiment, which can use variation, has continuous, can be micro- characteristic, vectorization expression from the latent space of encoding model
Natural sentence, and in the vector space carry out ANN approximate neighbour's semantic search, can achieve approximation recall, text mining
Effect.There is no limit can take long short-term memory Recognition with Recurrent Neural Network model pair for selection due to VAE model to encoder
Text is encoded, it may be considered that the clause information of sentence.To compare word2vec model, VAE model can provide more
Text representation abundant.Meanwhile variation is a kind of frame for generating model from coding, entire VAE frame is introducing LSTM-
Still the mode of unsupervised learning can be kept to train after RNN, can use generated into internet it is a large amount of unsupervised
Data.
As shown in fig. 6, the embodiment of the present invention provides a kind of excavating gear of neighbour's semanteme, comprising:
Natural sentence receiving module 100, for receiving the natural sentence of neighbour's semanteme to be excavated；
Coding obtains module 200, for the encoder according to variation from encoding model, compiles to the natural sentence
Code obtains the hidden vector of the natural sentence；Wherein, the hidden vector is located at the average point of encoder output distribution；
Neighbour's semantic search module 300 obtains and the hidden vector for carrying out neighbour's semantic search to the hidden vector
The hidden vector of neighbour with neighbour's semanteme；And
Decoding obtains module 400, for, from coding solution to model code device, being solved to the hidden vector of the neighbour according to variation
Code obtains the sentence for having neighbour's semanteme with the natural sentence.
In conjunction with second aspect, in the first embodiment of second aspect, as shown in fig. 7, described device further include:
Sentence word cutting module 500, for being cut to the natural sentence before being encoded to the natural sentence
Word obtains the word of the composition natural sentence；
Word number conversion module 600, for according to preset dictionary, the word for forming the natural sentence to be changed into accordingly
Numerical value；And
Vector conversion module 700, for according to term vector model, each word of the natural sentence described to composition to be corresponding each
Integer carries out vector conversion, obtains the vector expression of the natural sentence.
In conjunction with second aspect, in second of embodiment of second aspect, described device further include:
Relative entropy initialization module is set as zero for the initial coefficients by the variation from the relative entropy of encoding model；Its
In, the relative entropy is used to measure the distance between coding distribution and prior distribution of the encoder, and the prior distribution is
The ideal coding of the encoder is distributed；And
Model training iteration module, for being trained iteration from encoding model to the variation according to training corpus；Its
In, as the number of the trained iteration increases, the coefficient of the relative entropy rises the number until the coefficient along sine curve
Value is one.
In conjunction with second of embodiment of second aspect, in the third embodiment of second aspect, described device is also
Include:
Noise character replacement module, for replacing the training corpus using noise character according to preset word loss ratio
In word.
In conjunction with second of embodiment of second aspect, in the 4th kind of embodiment of second aspect, described device is also
Include:
Relative entropy judgment module, for judging whether the relative entropy is less than limitation during the trained iteration
Threshold value；And
Relative entropy removing module, for when the relative entropy is less than the threshold limit, by the relative entropy from described
Variation is deleted from encoding model.
The function of described device can also execute corresponding software realization by hardware realization by hardware.It is described
Hardware or software include one or more modules corresponding with above-mentioned function.
It include processor and memory, the memory in the mining structure of neighbour's semanteme in a possible design
Excavating gear for neighbour's semanteme executes the excavation program of neighbour's semanteme in above-mentioned first aspect, the processor is configured to
For executing the program stored in the memory.The excavating gear of neighbour's semanteme can also include communication interface, be used for
The excavating gear and other equipment or communication of neighbour's semanteme.
The embodiment of the present invention also provides a kind of terminal device, as shown in figure 8, the equipment includes: memory 21 and processor
22, being stored in memory 21 can be in the computer program on processor 22.Processor 22 is realized when executing computer program
State the method for digging of neighbour's semanteme in embodiment.The quantity of memory 21 and processor 22 can be one or more.
The equipment further include:
Communication interface 23, for the communication between processor 22 and external equipment.
Memory 21 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.
If memory 21, processor 22 and the independent realization of communication interface 23, memory 21, processor 22 and communication are connect
Mouth 23 can be connected with each other by bus and complete mutual communication.Bus can be industry standard architecture (ISA,
Industry Standard Architecture) bus, external equipment interconnection (PCI, Peripheral Component) be total
Line or extended industry-standard architecture (EISA, Extended Industry Standard Component) bus etc..Always
Line can be divided into address bus, data/address bus, control bus etc..Only to be indicated with a thick line in Fig. 8, but simultaneously convenient for indicating
Only a bus or a type of bus are not indicated.
Optionally, in specific implementation, if memory 21, processor 22 and communication interface 23 are integrated in chip piece
On, then memory 21, processor 22 and communication interface 23 can complete mutual communication by internal interface.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.Moreover, particular features, structures, materials, or characteristics described
It may be combined in any suitable manner in any one or more of the embodiments or examples.In addition, without conflicting with each other, this
The technical staff in field can be by the spy of different embodiments or examples described in this specification and different embodiments or examples
Sign is combined.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic." first " is defined as a result, the feature of " second " can be expressed or hidden
It include at least one this feature containing ground.In the description of the present invention, the meaning of " plurality " is two or more, unless otherwise
Clear specific restriction.
Any process described otherwise above or method description are construed as in flow chart or herein, and expression includes
It is one or more for realizing specific logical function or process the step of executable instruction code module, segment or portion
Point, and the range of the preferred embodiment of the present invention includes other realization, wherein can not press shown or discussed suitable
Sequence, including according to related function by it is basic simultaneously in the way of or in the opposite order, to execute function, this should be of the invention
Embodiment person of ordinary skill in the field understood.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (such as computer based system, including the system of processor or other can be held from instruction
The instruction fetch of row system, device or equipment and the system executed instruction) it uses, or combine these instruction execution systems, device or set
It is standby and use.For the purpose of this specification, " computer-readable medium ", which can be, any may include, stores, communicates, propagates or pass
Defeated program is for instruction execution system, device or equipment or the dress used in conjunction with these instruction execution systems, device or equipment
It sets.
The computer-readable medium of the embodiment of the present invention can be computer-readable signal media or computer-readable deposit
Storage media either the two any combination.The more specific example at least (non-exclusive of computer readable storage medium
List) include the following: there is the electrical connection section (electronic device) of one or more wirings, portable computer diskette box (magnetic dress
Set), random access memory (RAM), read-only memory (ROM), erasable edit read-only storage (deposit by EPROM or flash
Reservoir), fiber device and portable read-only memory (CDROM).In addition, computer readable storage medium can even is that
Can the paper of print routine or other suitable media on it because can for example be swept by carrying out optics to paper or other media
It retouches, is then edited, interprets or handled when necessary with other suitable methods electronically to obtain program, then will
It is stored in computer storage.
In embodiments of the present invention, computer-readable signal media may include in a base band or as carrier wave a part
The data-signal of propagation, wherein carrying computer-readable program code.The data-signal of this propagation can use a variety of
Form, including but not limited to electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media is also
It can be any computer-readable medium other than computer readable storage medium, which can send, pass
It broadcasts or transmits for instruction execution system, input method or device use or program in connection.Computer can
The program code for reading to include on medium can transmit with any suitable medium, including but not limited to: wirelessly, electric wire, optical cable, penetrate
Frequently (Radio Frequency, RF) etc. or above-mentioned any appropriate combination.
It should be appreciated that each section of the invention can be realized with hardware, software, firmware or their combination.Above-mentioned
In embodiment, software that multiple steps or method can be executed in memory and by suitable instruction execution system with storage
Or firmware is realized.It, and in another embodiment, can be under well known in the art for example, if realized with hardware
Any one of column technology or their combination are realized: having a logic gates for realizing logic function to data-signal
Discrete logic, with suitable combinational logic gate circuit specific integrated circuit, programmable gate array (PGA), scene
Programmable gate array (FPGA) etc..
Those skilled in the art are understood that realize all or part of step that above-described embodiment method carries
Suddenly be that relevant hardware can be instructed to complete by program, program can store in a kind of computer readable storage medium
In, which when being executed, includes the steps that one or a combination set of embodiment of the method.
It, can also be in addition, each functional unit in each embodiment of the present invention can integrate in a processing module
It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.If integrated module with
The form of software function module is realized and when sold or used as an independent product, also can store computer-readable at one
In storage medium.Storage medium can be read-only memory, disk or CD etc..
More than, only a specific embodiment of the invention, but scope of protection of the present invention is not limited thereto, and it is any to be familiar with
Those skilled in the art in the technical scope disclosed by the present invention, can readily occur in its various change or replacement, these
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
It is quasi-.
Claims (13)
1. a kind of method for digging of neighbour's semanteme characterized by comprising
Obtain the natural sentence of neighbour's semanteme to be excavated；
According to variation from the encoder of encoding model, the natural sentence is encoded, obtain the natural sentence it is hidden to
Amount；Wherein, the hidden vector is located at the average point of encoder output distribution；
Neighbour's semantic search is carried out to the hidden vector, obtains the hidden vector of neighbour that there is neighbour's semanteme with the hidden vector；With
And
According to variation from coding solution to model code device, the hidden vector of neighbour is decoded, acquisition has close with the natural sentence
Adjacent semantic sentence.
2. the method for digging of neighbour's semanteme as described in claim 1, which is characterized in that encoded to the natural sentence
Before further include:
Word cutting is carried out to the natural sentence, obtains the word of the composition natural sentence；
According to preset dictionary, the word for forming the natural sentence is changed into corresponding numerical value；And
According to term vector model, vector conversion is carried out to the corresponding each integer of each word for forming the natural sentence, obtains institute
State the vector expression of nature sentence.
3. the method for digging of neighbour's semanteme as described in claim 1, which is characterized in that the method also includes:
The initial coefficients of the variation from the relative entropy of encoding model are set as zero；Wherein, the relative entropy is described for measuring
The distance between the coding distribution of encoder and prior distribution, the prior distribution are the ideal coding point of the encoder
Cloth；And
According to training corpus, iteration is trained from encoding model to the variation；Wherein, with the number of the trained iteration
Increase, the coefficient of the relative entropy rises along sine curve until the numerical value of the coefficient is one.
4. the method for digging of neighbour's semanteme as claimed in claim 3, which is characterized in that the method also includes:
According to preset word loss ratio, the word in the training corpus is replaced using noise character.
5. the method for digging of neighbour's semanteme as claimed in claim 3, which is characterized in that the method also includes:
During the trained iteration, judge whether the relative entropy is less than threshold limit；And
When the relative entropy is less than the threshold limit, the relative entropy is deleted from the variation from encoding model.
6. such as the method for digging of neighbour's semanteme described in any one of claim 1 to 5, which is characterized in that the method also includes:
According to the encoder, two natural sentences are encoded respectively, obtain described two natural sentences it is corresponding it is hidden to
Amount；And
According to the corresponding hidden vector of described two nature sentences, the similarity of described two natural sentences is calculated.
7. a kind of excavating gear of neighbour's semanteme characterized by comprising
Natural sentence receiving module, for receiving the natural sentence of neighbour's semanteme to be excavated；
Coding obtains module, for the encoder according to variation from encoding model, encodes to the natural sentence, obtains institute
State the hidden vector of nature sentence；Wherein, the hidden vector is located at the average point of encoder output distribution；
Neighbour's semantic search module is obtained with the hidden vector for carrying out neighbour's semantic search to the hidden vector with close
The adjacent semantic hidden vector of neighbour；And
Decoding obtains module, for, from coding solution to model code device, being decoded, obtaining to the hidden vector of the neighbour according to variation
There is the sentence of neighbour's semanteme with the natural sentence.
8. the excavating gear of neighbour's semanteme as claimed in claim 7, which is characterized in that described device further includes
Sentence word cutting module, for carrying out word cutting to the natural sentence, obtaining before being encoded to the natural sentence
Form the word of the natural sentence；
Word number conversion module, for according to preset dictionary, the word for forming the natural sentence to be changed into corresponding numerical value；
And
Vector conversion module, for according to term vector model, to form the corresponding each integer of each word of the natural sentence into
Row vector conversion obtains the vector expression of the natural sentence.
9. the excavating gear of neighbour's semanteme as claimed in claim 7, which is characterized in that described device further include:
Relative entropy initialization module is set as zero for the initial coefficients by the variation from the relative entropy of encoding model；Wherein, institute
It states relative entropy and is distributed the distance between prior distribution for measuring the coding of the encoder, the prior distribution is the volume
The ideal coding distribution of code device；And
Model training iteration module, for being trained iteration from encoding model to the variation according to training corpus；Wherein,
As the number of the trained iteration increases, the coefficient of the relative entropy rises along sine curve until the numerical value of the coefficient is
One.
10. the excavating gear of neighbour's semanteme as claimed in claim 9, which is characterized in that described device further include:
Noise character replacement module, for being replaced in the training corpus using noise character according to preset word loss ratio
Word.
11. the excavating gear of neighbour's semanteme as claimed in claim 9, which is characterized in that described device further include:
Relative entropy judgment module, for judging whether the relative entropy is less than threshold limit during the trained iteration；
And
Relative entropy removing module, for when the relative entropy is less than the threshold limit, by the relative entropy from the variation
It is deleted from encoding model.
12. a kind of excavation terminal device for realizing neighbour's semanteme, which is characterized in that the terminal device includes:
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processors
Realize the method for digging such as neighbour's semanteme as claimed in any one of claims 1 to 6.
13. a kind of computer readable storage medium, is stored with computer program, which is characterized in that the program is held by processor
The method for digging such as neighbour's semanteme as claimed in any one of claims 1 to 6 is realized when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810712820.1A CN108959551B (en) | 2018-06-29 | 2018-06-29 | Neighbor semantic mining method and device, storage medium and terminal equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810712820.1A CN108959551B (en) | 2018-06-29 | 2018-06-29 | Neighbor semantic mining method and device, storage medium and terminal equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108959551A true CN108959551A (en) | 2018-12-07 |
| CN108959551B CN108959551B (en) | 2021-07-13 |
Family
ID=64485038
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810712820.1A Active CN108959551B (en) | 2018-06-29 | 2018-06-29 | Neighbor semantic mining method and device, storage medium and terminal equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108959551B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978379A (en) * | 2019-03-28 | 2019-07-05 | 北京百度网讯科技有限公司 | Time series data method for detecting abnormality, device, computer equipment and storage medium |
| CN110196894A (en) * | 2019-05-30 | 2019-09-03 | 北京百度网讯科技有限公司 | The training method and prediction technique of language model |
| CN110275936A (en) * | 2019-05-09 | 2019-09-24 | 浙江工业大学 | A kind of similar law case retrieving method based on from coding neural network |
| CN110288965A (en) * | 2019-05-21 | 2019-09-27 | 北京达佳互联信息技术有限公司 | A kind of music synthesis method, device, electronic equipment and storage medium |
| CN111966998A (en) * | 2020-07-23 | 2020-11-20 | 华南理工大学 | Password generation method, system, medium, and apparatus based on variational automatic encoder |
| WO2021008068A1 (en) * | 2019-07-16 | 2021-01-21 | 北京市商汤科技开发有限公司 | Image processing method and apparatus |
| CN112287641A (en) * | 2020-12-25 | 2021-01-29 | 上海旻浦科技有限公司 | Synonym sentence generating method, system, terminal and storage medium |
| CN112507091A (en) * | 2020-12-01 | 2021-03-16 | 百度健康(北京)科技有限公司 | Method, device, equipment and storage medium for retrieving information |
| CN112528015A (en) * | 2020-10-26 | 2021-03-19 | 复旦大学 | Method and device for judging rumor in message interactive transmission |
| WO2021093308A1 (en) * | 2019-11-13 | 2021-05-20 | 百度在线网络技术（北京）有限公司 | Method and apparatus for extracting poi name, device, and computer storage medium |
| TWI770477B (en) * | 2019-09-03 | 2022-07-11 | 日商三菱電機股份有限公司 | Information processing device, storage medium, program product and information processing method |
| CN117077656A (en) * | 2023-09-22 | 2023-11-17 | 腾讯科技（深圳）有限公司 | Demonstration relation mining method and device, medium and electronic equipment |
| CN117113281A (en) * | 2023-10-20 | 2023-11-24 | 光轮智能(北京)科技有限公司 | Multi-mode data processing method, device, agent and medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657350A (en) * | 2015-03-04 | 2015-05-27 | 中国科学院自动化研究所 | Hash learning method for short text integrated with implicit semantic features |
| CN106919556A (en) * | 2017-02-21 | 2017-07-04 | 重庆邮电大学 | A kind of natural language semanteme deep analysis algorithm of use sparse coding |
| CN107194868A (en) * | 2017-05-19 | 2017-09-22 | 成都通甲优博科技有限责任公司 | A kind of Face image synthesis method and device |
| CN107392781A (en) * | 2017-06-20 | 2017-11-24 | 挖财网络技术有限公司 | The recognition methods of customer relationship, the recognition methods of object relationship and device |
| CN107507126A (en) * | 2017-07-27 | 2017-12-22 | 大连和创懒人科技有限公司 | A kind of method that 3D scenes are reduced using RGB image |
| CN107526809A (en) * | 2017-08-23 | 2017-12-29 | 北京百度网讯科技有限公司 | Method and apparatus based on artificial intelligence push music |
| CN107610193A (en) * | 2016-06-23 | 2018-01-19 | 西门子保健有限责任公司 | Use the image rectification of depth production machine learning model |
| US20180101784A1 (en) * | 2016-10-05 | 2018-04-12 | D-Wave Systems Inc. | Discrete variational auto-encoder systems and methods for machine learning using adiabatic quantum computers |
| CN108171324A (en) * | 2017-12-26 | 2018-06-15 | 天津科技大学 | A kind of variation own coding mixed model |
-
2018
- 2018-06-29 CN CN201810712820.1A patent/CN108959551B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657350A (en) * | 2015-03-04 | 2015-05-27 | 中国科学院自动化研究所 | Hash learning method for short text integrated with implicit semantic features |
| CN107610193A (en) * | 2016-06-23 | 2018-01-19 | 西门子保健有限责任公司 | Use the image rectification of depth production machine learning model |
| US20180101784A1 (en) * | 2016-10-05 | 2018-04-12 | D-Wave Systems Inc. | Discrete variational auto-encoder systems and methods for machine learning using adiabatic quantum computers |
| CN106919556A (en) * | 2017-02-21 | 2017-07-04 | 重庆邮电大学 | A kind of natural language semanteme deep analysis algorithm of use sparse coding |
| CN107194868A (en) * | 2017-05-19 | 2017-09-22 | 成都通甲优博科技有限责任公司 | A kind of Face image synthesis method and device |
| CN107392781A (en) * | 2017-06-20 | 2017-11-24 | 挖财网络技术有限公司 | The recognition methods of customer relationship, the recognition methods of object relationship and device |
| CN107507126A (en) * | 2017-07-27 | 2017-12-22 | 大连和创懒人科技有限公司 | A kind of method that 3D scenes are reduced using RGB image |
| CN107526809A (en) * | 2017-08-23 | 2017-12-29 | 北京百度网讯科技有限公司 | Method and apparatus based on artificial intelligence push music |
| CN108171324A (en) * | 2017-12-26 | 2018-06-15 | 天津科技大学 | A kind of variation own coding mixed model |
Non-Patent Citations (1)
| Title |
|---|
| 绿蚁新醅酒，红泥小火炉: ""GAN与VAE（我的理解）"", 《知乎》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978379A (en) * | 2019-03-28 | 2019-07-05 | 北京百度网讯科技有限公司 | Time series data method for detecting abnormality, device, computer equipment and storage medium |
| CN110275936A (en) * | 2019-05-09 | 2019-09-24 | 浙江工业大学 | A kind of similar law case retrieving method based on from coding neural network |
| CN110288965A (en) * | 2019-05-21 | 2019-09-27 | 北京达佳互联信息技术有限公司 | A kind of music synthesis method, device, electronic equipment and storage medium |
| CN110288965B (en) * | 2019-05-21 | 2021-06-18 | 北京达佳互联信息技术有限公司 | Music synthesis method and device, electronic equipment and storage medium |
| CN110196894A (en) * | 2019-05-30 | 2019-09-03 | 北京百度网讯科技有限公司 | The training method and prediction technique of language model |
| CN110196894B (en) * | 2019-05-30 | 2021-06-08 | 北京百度网讯科技有限公司 | Language model training method and language model prediction method |
| WO2021008068A1 (en) * | 2019-07-16 | 2021-01-21 | 北京市商汤科技开发有限公司 | Image processing method and apparatus |
| TWI770477B (en) * | 2019-09-03 | 2022-07-11 | 日商三菱電機股份有限公司 | Information processing device, storage medium, program product and information processing method |
| WO2021093308A1 (en) * | 2019-11-13 | 2021-05-20 | 百度在线网络技术（北京）有限公司 | Method and apparatus for extracting poi name, device, and computer storage medium |
| US11768892B2 (en) | 2019-11-13 | 2023-09-26 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for extracting name of POI, device and computer storage medium |
| CN111966998A (en) * | 2020-07-23 | 2020-11-20 | 华南理工大学 | Password generation method, system, medium, and apparatus based on variational automatic encoder |
| CN112528015A (en) * | 2020-10-26 | 2021-03-19 | 复旦大学 | Method and device for judging rumor in message interactive transmission |
| CN112528015B (en) * | 2020-10-26 | 2022-11-18 | 复旦大学 | Method and device for judging rumor in message interactive transmission |
| CN112507091A (en) * | 2020-12-01 | 2021-03-16 | 百度健康(北京)科技有限公司 | Method, device, equipment and storage medium for retrieving information |
| CN112287641B (en) * | 2020-12-25 | 2021-03-09 | 上海旻浦科技有限公司 | Synonym sentence generating method, system, terminal and storage medium |
| CN112287641A (en) * | 2020-12-25 | 2021-01-29 | 上海旻浦科技有限公司 | Synonym sentence generating method, system, terminal and storage medium |
| CN117077656A (en) * | 2023-09-22 | 2023-11-17 | 腾讯科技（深圳）有限公司 | Demonstration relation mining method and device, medium and electronic equipment |
| CN117077656B (en) * | 2023-09-22 | 2024-01-02 | 腾讯科技（深圳）有限公司 | Demonstration relation mining method and device, medium and electronic equipment |
| CN117113281A (en) * | 2023-10-20 | 2023-11-24 | 光轮智能(北京)科技有限公司 | Multi-mode data processing method, device, agent and medium |
| CN117113281B (en) * | 2023-10-20 | 2024-01-26 | 光轮智能(北京)科技有限公司 | Multi-mode data processing method, device, agent and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108959551B (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108959551A (en) | Method for digging, device, storage medium and the terminal device of neighbour's semanteme | |
| CN108197736B (en) | Air quality prediction method based on variational self-encoder and extreme learning machine | |
| CN105279554B (en) | The training method and device of deep neural network based on Hash coding layer | |
| CN112084331A (en) | Text processing method, text processing device, model training method, model training device, computer equipment and storage medium | |
| CN109346045B (en) | Multi-vocal part music generation method and device based on long-short time neural network | |
| CN110287480A (en) | A kind of name entity recognition method, device, storage medium and terminal device | |
| CN108197294A (en) | A kind of text automatic generation method based on deep learning | |
| CN108984904A (en) | A kind of Home Fashion & Design Shanghai method based on deep neural network | |
| CN116960989B (en) | Power load prediction method, device and equipment for power station and storage medium | |
| CN108920644A (en) | Talk with judgment method, device, equipment and the computer-readable medium of continuity | |
| CN106844341A (en) | News in brief extracting method and device based on artificial intelligence | |
| CN103914527B (en) | Graphic image recognition and matching method based on genetic programming algorithms of novel coding modes | |
| CN102576090A (en) | System and method for modeling a geologic volume of interest | |
| CN106778910A (en) | Deep learning system and method based on local training | |
| CN110347790A (en) | Text duplicate checking method, apparatus, equipment and storage medium based on attention mechanism | |
| CN109214241A (en) | Pedestrian detection method based on deep learning algorithm | |
| CN115907436B (en) | Quality coupling prediction-based water resource water environment regulation and control method and system | |
| CN110348563A (en) | The semi-supervised training method of neural network, device, server and storage medium | |
| CN112365359A (en) | Training method, device, equipment and storage medium for planting decision determination model | |
| CN110134960A (en) | A kind of generation method and relevant device of text | |
| CN106951089A (en) | Gesture interaction method and system | |
| CN114334007A (en) | Gene off-target prediction model training method, prediction method, device and electronic equipment | |
| CN106782513A (en) | Speech recognition realization method and system based on confidence level | |
| CN100495535C (en) | Speech recognition device and speech recognition method | |
| CN108898648A (en) | A kind of K line chart building method, system and relevant device |
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 |