CN107992937A - Unstructured data decision method and device based on deep learning - Google Patents
Unstructured data decision method and device based on deep learning Download PDFInfo
- Publication number
- CN107992937A CN107992937A CN201610950494.9A CN201610950494A CN107992937A CN 107992937 A CN107992937 A CN 107992937A CN 201610950494 A CN201610950494 A CN 201610950494A CN 107992937 A CN107992937 A CN 107992937A
- Authority
- CN
- China
- Prior art keywords
- data
- deep learning
- training
- factor
- neural network
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 117
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 108
- 238000003062 neural network model Methods 0.000 claims abstract description 68
- 238000000605 extraction Methods 0.000 claims abstract description 35
- 238000004458 analytical method Methods 0.000 claims abstract description 34
- 230000007935 neutral effect Effects 0.000 claims description 43
- 238000010845 search algorithm Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 description 8
- 238000007405 data analysis Methods 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 210000004218 nerve net Anatomy 0.000 description 3
- 238000013500 data storage Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 241000894007 species Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000010223 real-time analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of unstructured data decision method based on deep learning, specifically include：The deep learning neural network model after training is obtained, wherein the deep learning neural network model after the training is the neural network model of the multiple-factor cascading judgement after the training of multiple-factor training sample data；The hints data of online real time collecting is received, the hints data includes the unstructured data of numerous types of data；Conjoint Analysis carries out the hints data of acquisition by the deep learning neural network model after training, extraction is conducive to the characteristic curve of judgement information without hesitation；The hints data is made decisions according to the feature hint information, generates court verdict；Feed back the court verdict.The above method can carry out unstructured data more efficient, Analysis of Policy Making much sooner, realize on line real time.
Description
Technical field
The present invention relates to field of artificial intelligence, sentences more particularly to a kind of unstructured data based on deep learning
Certainly method and apparatus.
Background technology
Relative to structural data（Data at once, are stored in lane database, can be real come logical expression with bivariate table structure
Existing data）For, it has not been convenient to it is known as unstructured data with database two dimension logical table come the data showed, including it is all
Office documents, text, picture, subset X ML, HTML under standard generalized markup language, all kinds of reports, image and the sound of form
Frequently/video information etc..Unstructured data have data volume is big, change is fast, species is more, abundant in content complexity and structure not
The characteristics of unified.
The features such as change in conventional art due to unstructured data is fast, structure disunity, to unstructured data into
It is time-consuming longer to there is analysis during row data analysis, off-line analysis typically is carried out to historical data, analysis decision efficiency is low.
The content of the invention
Based on this, it is necessary to for it is above-mentioned the problem of, there is provided one kind can to unstructured data carry out more efficiently, more
Add the unstructured data decision method and device based on deep learning of timely on-line decision analysis.
A kind of unstructured data decision method based on deep learning, the described method includes：
The deep learning neural network model after training is obtained, wherein, the deep learning neutral net mould after the training
Type is the multiple-factor cascading judgement neural network model trained through multiple-factor training sample；
The hints data of online real time collecting is received, the hints data is the unstructured number for including numerous types of data
According to；
Conjoint Analysis is carried out to the hints data of reception by the deep learning neural network model after the training,
Extraction is conducive to the feature hint information of judgement；
The hints data is made decisions according to the feature hint information, generates court verdict；
Feed back the court verdict.
In one embodiment, the line of the deep learning neural network model by after the training to reception
The step of rope data carry out Conjoint Analysis, and extraction is conducive to the feature hint information of judgement includes：
Feature extraction is carried out to the hints data of reception by the deep learning neural network model after the training,
Obtain high dimensional feature vector；
The high dimensional feature vector is changed into by the hash code with each dimension incidence relation using hash search algorithm, and
According to the hash code be conducive to the extraction of the feature hint information of judgement.
In one embodiment, in the deep learning neural network model obtained after training, wherein, after the training
Deep learning neural network model be through multiple-factor training sample training multiple-factor cascading judgement neural network model step
Before rapid, further include：
The training sample that receiving terminal uploads, wherein, the training sample data are multiple-factor training sample；
Under off-line state, using deep learning algorithm to each corresponding training of the factor in the multiple-factor training sample
Sample builds deep learning neutral net submodel；
Obtain the incidence relation between the deep learning neutral net submodel of structure；
The deep learning neutral net submodel is merged according to the incidence relation, generation, which can carry out multiple-factor, combines and determine
The deep learning neural network model of plan.
In one embodiment, the multiple-factor training sample includes video data, view data, voice data, text
Data and network data.
In one embodiment, when the hints data received in setting time includes a kind of data type, then
The hints data is analyzed using the deep learning neutral net submodel of the corresponding factor of the data type
Judgement, obtains court verdict.
A kind of unstructured data judgment device based on deep learning, described device include：
Neural network model acquisition module, for obtaining the deep learning neural network model after training, wherein, the instruction
Deep learning neural network model after white silk is the multiple-factor cascading judgement neural network model trained through multiple-factor training sample；
Real time data receiving module, for receiving the hints data of online real time collecting, the hints data be include it is more
The unstructured data of kind data type；
Characteristic information extracting module, for by the deep learning neural network model after the training to described in reception
Hints data carries out Conjoint Analysis, and extraction is conducive to the feature hint information of judgement；
Cascading judgement module, for being made decisions according to the feature hint information to the hints data, generation judgement
As a result；
Court verdict feedback module, for feeding back the court verdict.
In one embodiment, the characteristic information extracting module, is additionally operable to by the deep learning god after the training
Feature extraction is carried out to the hints data of reception through network model, obtains high dimensional feature vector；Using hash search algorithm
The high dimensional feature vector is changed into the hash code with each dimension incidence relation, and is conducive to according to the hash code
The extraction of the feature hint information of judgement.
In one embodiment, described device further includes：
Training sample data receiving module, the training sample uploaded for receiving terminal, wherein, the training sample is more
Factor training sample；
Submodel training module, under off-line state, using deep learning algorithm to the multiple-factor training sample
In each factor corresponding training sample structure deep learning neutral net submodel；
Incidence relation analysis module, the association between the deep learning neutral net submodel for obtaining structure are closed
System；
Joint decision model construction module, for merging the deep learning neutral net submodule according to the incidence relation
Type, generation can carry out the deep learning neural network model of multiple-factor joint decision.
In one embodiment, the multiple-factor training sample includes video data, view data, voice data, text
Data and network data.
In one embodiment, described device further includes：Submodel judging module, for working as what is received in setting time
When the hints data includes a kind of data type, then using the deep learning nerve of the corresponding factor of the data type
Network submodel carries out analysis judgement to the hints data, obtains court verdict.
Above-mentioned unstructured data decision method and device based on deep learning, by using multiple-factor training sample pair
Deep learning neural network model is trained, and is obtained carrying out the neural network model of multiple-factor cascading judgement, is used this
Neural network model after training carries out Conjoint Analysis to the unstructured data with numerous types of data in real time, and extraction has
Beneficial to the feature hint information of data decision, to be screened to analysis data, and then more efficiently, timely carry out data
Analysis judgement, obtains court verdict.The analysis of mutual incidence relation is namely carried out to polytype real time data, after
And confirm the arbitration schemes of more quick intelligence, sentenced in real time using the online of data with realizing that data volume is big, change is fast
Certainly.
Brief description of the drawings
Fig. 1 is the applied environment figure of the unstructured data decision method based on deep learning in one embodiment；
Fig. 2 is the flow chart of the unstructured data decision method based on deep learning in one embodiment；
Fig. 3 is the flow chart of training deep learning neural network model in one embodiment；
Fig. 4 is the structure diagram of the unstructured data decision method based on deep learning in one embodiment；
Fig. 5 is the structure diagram involved by training deep learning neural network model in one embodiment.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, it is right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
As shown in Figure 1, in one embodiment, there is provided a kind of unstructured data decision method based on deep learning
With the applied environment figure of device, which includes data acquisition equipment 110 and server 120, wherein, data collection station
110 can be communicated by network with server 120.Data acquisition equipment may include camera shooting terminal, the network terminal, data storage
Medium etc., wherein, camera shooting terminal can be the unstructured datas such as captured video data, view data and voice data；Network
Terminal can obtain the unstructured datas such as network data, text data, view data, video data；In data storage medium
It is non-structural to be previously stored with the text data being collected into by channel under other lines, view data, video data, voice data etc.
Change data.By network attached server 120 can online real-time reception data acquisition equipment 110 obtain include a variety of data class
The unstructured data of type, and the deep learning neural network model that can carry out multiple-factor judgement after training is called to real-time
The unstructured data of numerous types of data quickly adjudicated, court verdict is obtained, due to the deep learning made decisions
Neutral net can analyze the incidence relation between the unstructured data of numerous types of data, can utilize polytype number at the same time
According to making decisions, judgement can be made much sooner, efficiently.
In one embodiment, as shown in Figure 2, there is provided a kind of unstructured data judgement side based on deep learning
Method, this method are illustrated exemplified by applying in server 120 as shown in Figure 1, specifically comprised the following steps：
Step S202：The deep learning neural network model after training is obtained, wherein, the deep learning nerve net after training
Network model is the neural network model of the multiple-factor cascading judgement through the training of multiple-factor training sample.
Specifically, deep learning refers to multilayer neural network, deep learning can be formed more by combining low-level feature
Add abstract high-rise expression attribute classification or feature, to find that the distributed nature of data represents.Deep neural network is one
The process of feature is successively extracted, and is that computer is automatically extracted from data, it is not necessary to its extraction process of human intervention, its
Matter thought is exactly to stack multiple neuronal layers, and each layer extracts certain feature and information, and the output of this layer is as next
The input of layer.In this way, it is possible to realize that carrying out classification to input information expresses.To find that the distribution of data is special
Sign represents.By taking image recognition as an example, first layer extraction boundary information, second layer extraction boundary profile information, then profile can be with
Subdivision is combined into, subdivision is combined into object, so gets off successively to extract feature from level to level, is formed by combining low-level feature
More abstract high-rise expression attribute classification or feature, judges in picture which is by the various combination of feature or attribute
The object of species.
Deep learning neural network model have it is a variety of, such as convolutional neural networks model, denoising self-encoding encoder, limited Bohr hereby
Graceful machine（Restricted Boltzmann Machine,RBM）Network model etc..Deep learning neural network model is instructed
When practicing, any of the above-described kind of network model may be selected and be trained.
In the present embodiment, carry out above-mentioned deep learning neural network model using multiple-factor training sample and be trained,
Learn the incidence relation between the feature of each factor and each ratio characteristics, and build multiple-factor cascading judgement deep learning nerve net
Network model.Wherein, multiple-factor here can include the video data factor, the view data factor, the voice data factor, textual data
According to any number of in the factor and the network data factor.
Step S204：Receive the hints data that gathers in real time of line, hints data is include numerous types of data non-structural
Change data.
The unstructured data that server is obtained in real time by opening up each data acquisition terminal of multiple interfaces, wherein,
The unstructured data that server receives includes numerous types of data, specifically, can include real time video data, realtime graphic
Data, realaudio data, real-time text data and real-time network data etc..
Step S206：Joint point carries out the hints data of reception by the deep learning neural network model after training
Analysis, extraction are conducive to the feature hint information of judgement.
Step S208：Hints data is made decisions according to feature hint information, generates court verdict.
Specifically, server real-time reception has the unstructured data of numerous types of data, such as real-time reception video counts
According to and network data, the deep learning neural network model after training the video data and network data of real-time reception are carried out special
Sign extraction and analysis, find the feature hint information for being conducive to decision-making, and the feature hint information based on lookup is to real-time architecture
Magnanimity video data and/or network data screened and filtered, with it is more quick, make judgement in time.
Step S210：Feed back court verdict.
The feedback for making decisions structure can be fed back in the form of triggering and alarming or directly push away feedback arrangement
The display terminal specified is sent to show.
For example, the purpose analyzed unstructured data is：The searched targets object from video data, specifically
, the unstructured data decision method based on deep learning for above-mentioned purpose is：
Step 1：The continuous video data that real-time reception camera shooting terminal is sent, and the data of other data types are received,
Wherein, other data types can be voice data, text data etc..
Step 2：Using the deep learning neural network model after training to video data, text data, voice data etc.
The hints data of reception carries out feature extraction and analysis, and extraction is conducive to the feature clue letter of the searched targets from video data
Breath, wherein this feature hint information can be the acoustic informations of the positional information of searched targets, searched targets, and deep learning is neural
Network model using the feature clue information sifting video data for being conducive to make decisions of extraction, believe by the position for such as searching target
Video data near breath, excludes to be unsatisfactory for the video data of locality condition, greatly reduces the data volume for needing judgment analysis,
Data decision is rapider, and then can realize online judgement in real time.Again for example, the sound of searched targets is retrieved in voice data
When message ceases, the incidence relation of video data and voice data in deep learning neural network model is called, navigates to target sound
The corresponding video-data fragment of message breath, and then quick-searching is to the destination object of video.
It should be noted that the deep learning decision method in the present embodiment be not limited only to it is above-mentioned from video data
The decision tasks of searched targets object, can also carry out that other data volumes are big, change is fast using the classification of data, identification, with
The real-time judgement of track etc..
In the present embodiment, deep learning neural network model is trained using multiple-factor training sample, obtaining can
The neural network model of multiple-factor cascading judgement is carried out, using the neural network model after the training to having a variety of numbers in real time
Conjoint Analysis is carried out according to the unstructured data of type, extraction is conducive to the feature hint information of data decision, so as to analysis
Data are screened, and then more efficient, timely progress data analysis judgement, obtain court verdict.Namely to multiple types
The real time data of type carries out the analysis of mutual incidence relation, more quick, intelligence arbitration schemes is then confirmed, to realize
The online real-time judgement using data that data volume is big, change is fast.
In one embodiment, step S206：Clue by the deep learning neural network model after training to reception
Data carry out Conjoint Analysis, and the feature clue information that extraction is conducive to judgement includes：
Feature extraction is carried out to the hints data of reception by the deep learning neural network model after training, obtains higher-dimension
Feature vector.
High dimensional feature vector will be changed into by the hash code with each dimension incidence relation, and root using hash search algorithm
According to hash code be conducive to the extraction of the feature hint information of judgement.
Specifically, high dimensional feature vector refers to more dimensions of the deep learning neural network model after training to real-time reception
According to（Each data type is as a data dimension）Carry out data analysis, generate higher-dimension, reception can be characterized it is non-structural
Change the feature vector of data.
In the present embodiment, in the former space for ensureing each dimension close on relation on the basis of, using hash search algorithm
By the high dimensional feature DUAL PROBLEMS OF VECTOR MAPPING of generation into binary string（binary code）Hash code, the storage of data can be substantially reduced
And communication overhead, so as to effectively improve the efficiency and speed of calculating, realize the real-time analysis of a large amount of unstructured datas with sentencing
Certainly.
In one embodiment, as shown in figure 3, in step S202：The deep learning neural network model after training is obtained,
Wherein, the deep learning neural network model after training is the multiple-factor cascading judgement nerve net trained through multiple-factor training sample
Before network model, the training step to deep learning deep neural network model is further included, is specifically comprised the following steps：
Step S302：The training sample that receiving terminal uploads, wherein training sample is multiple-factor training sample.
Specifically, multiple-factor sample is the sample data for including numerous types of data.In one embodiment, training sample
Data are to include video sample data, image sample data, audio sample data, text sample data and network sample data
Deng the multiple-factor training sample data of multiple types of data.
Step S304：, should to each factor pair in multiple-factor training sample using deep learning algorithm under off-line state
Training sample data structure deep learning neutral net submodel.
Specifically, being pre-processed to the corresponding training sample of each factor, and extracted respectively using deep learning algorithm
The feature of the corresponding training sample of each factor, for example, the moving target feature of extraction video training sample or extraction video
The destination object feature of training sample；Audio frequency characteristics in voice data etc. are extracted, and it is corresponding to establish each factor respectively
The deep learning neutral net submodel of training sample data, as video neutral net submodel, audio neutral net submodel,
Image neutral net submodel, text neutral net submodel, the neutral net submodel of network data.
Step S306：Obtain the incidence relation between the deep learning neutral net submodel of structure.
Step S308：Deep learning neutral net submodel is merged according to incidence relation, generation can carry out multiple-factor joint
The deep learning neural network model of decision-making.
Specifically, excavating the incidence relation between the multiple deep learning neutral net submodels built, and carry out submodule
The fusion of type, and then generate the deep learning neural network model that can carry out multiple-factor joint decision.
Specifically, if the deep learning neural network model for carrying out multiple-factor joint decision of fusion is used to retrieve video
In destination object, then, excavating the essence of the incidence relation between multiple deep learning neutral net submodels is：Establish non-
The incidence relation of video neutral net submodel and video neutral net submodel.Wherein, non-video neutral net submodel is
The submodel of characteristic information structure based on non-video data and extraction, e.g., image neutral net submodel is for extracting mesh
Subjects face feature is marked, text neutral net submodel is used to extract position feature of destination object etc..Establish video nerve
Incidence relation between network submodel and non-video neutral net submodel is so that the joint decision deep learning nerve of structure
Network model can utilize the characteristic information of non-video neutral net submodel extraction, the quick inspection for carrying out destination object in video
The efficiency of Suo Tigao analysis and decisions.
In one embodiment, when the hints data that the terminal received in setting time gathers in real time includes a kind of data
During type, then analysis is carried out to hints data using the deep learning neutral net submodel of the corresponding factor of data type and sentenced
Certainly.
Specifically, the real-time hints data received online when server within a period of time, is only a kind of data type
When, such as only video data when, then directly carry out the analysis of the video data using video depth learning neural network submodel
Judgement.
Whether can be numerous types of data according to the data of reception in the present embodiment, carry out cascading judgement and forms data shape
The switching of formula judgement, when to carry out cascading judgement during numerous types of data, improving the efficiency of judgement,；When for single data type
When, directly single data type is analyzed, data decision will not be stopped because of only single data type.
In one embodiment, as shown in Figure 4, there is provided a kind of unstructured data judgement dress based on deep learning
Put, which includes：
Neural network model acquisition module 402, for obtaining the deep learning neural network model after training, wherein, instruction
Deep learning neural network model after white silk is the multiple-factor cascading judgement neural network model trained through multiple-factor training sample.
Real time data receiving module 404, for receiving the hints data of online real time collecting, hints data be include it is a variety of
The unstructured data of data type.
Characteristic information extracting module 406, for the clue by the deep learning neural network model after training to reception
Data carry out Conjoint Analysis, and extraction is conducive to the feature hint information of judgement.
Cascading judgement module 408, for being made decisions according to feature hint information to hints data, generates court verdict.
Court verdict feedback module 410, for feeding back court verdict.
In one embodiment, characteristic information extracting module 406 is additionally operable to by the deep learning neutral net after training
Model carries out feature extraction to the hints data of reception, obtains high dimensional feature vector；Using hash search algorithm by high dimensional feature
Vector changes into the hash code with each dimension incidence relation, and according to hash code be conducive to the feature hint information of judgement
Extraction.
In one embodiment, as shown in figure 5, deep learning judgment device further includes：
Training sample data receiving module 502, the training sample uploaded for receiving terminal, wherein, training sample is more
Factor training sample.
Submodel training module 504, under off-line state, using deep learning algorithm in multiple-factor training sample
Each factor corresponding training sample structure deep learning neutral net submodel.
Incidence relation analysis module 506, the association between deep learning neutral net submodel for obtaining structure are closed
System.
Joint decision model construction module 508, it is raw for merging deep learning neutral net submodel according to incidence relation
Into the deep learning neural network model that can carry out multiple-factor joint decision.
In one embodiment, multiple-factor training sample includes video data, view data, voice data, text data
And network data.
In one embodiment, deep learning judgment device further includes：Submodel judging module, for when in setting time
When the hints data of interior reception includes a kind of data type, then using the deep learning neutral net of the corresponding factor of data type
Submodel carries out analysis judgement to hints data, obtains court verdict.
One of ordinary skill in the art will appreciate that realize all or part of flow in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, program can be stored in a computer read/write memory medium, such as
In the embodiment of the present invention, which can be stored in the storage medium of computer system, and by the computer system at least
One processor performs, to realize the flow for including the embodiment such as above-mentioned each method.Wherein, storage medium can be magnetic disc, light
Disk, read-only memory（Read-Only Memory, ROM）Or random access memory（Random Access Memory,
RAM）Deng.
Each technical characteristic of embodiment described above can be combined arbitrarily, to make description succinct, not to above-mentioned reality
Apply all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, the scope that this specification is recorded all is considered to be.
Embodiment described above only expresses the several embodiments of the present invention, its description is more specific and detailed, but simultaneously
Cannot therefore it be construed as limiting the scope of the patent.It should be pointed out that come for those of ordinary skill in the art
Say, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the protection of the present invention
Scope.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (10)
1. a kind of unstructured data decision method based on deep learning, the described method includes：
The deep learning neural network model after training is obtained, wherein, the deep learning neural network model after the training is
The multiple-factor cascading judgement neural network model trained through multiple-factor training sample；
The hints data of online real time collecting is received, the hints data is the unstructured data for including numerous types of data；
Conjoint Analysis, extraction are carried out to the hints data of reception by the deep learning neural network model after the training
Be conducive to the feature hint information of judgement；
The hints data is made decisions according to the feature hint information, generates court verdict；
Feed back the court verdict.
2. the according to the method described in claim 1, it is characterized in that, deep learning neutral net by after the training
The step of model carries out Conjoint Analysis to the hints data of reception, and extraction is conducive to the feature hint information of judgement includes：
Feature extraction is carried out to the hints data of reception by the deep learning neural network model after the training, is obtained
High dimensional feature vector；
The high dimensional feature vector is changed into by the hash code with each dimension incidence relation using hash search algorithm, and according to
The hash code be conducive to the extraction of the feature hint information of judgement.
3. according to the method described in claim 1, it is characterized in that, in the deep learning neutral net mould obtained after training
Type, wherein, the deep learning neural network model after the training is combined for the multiple-factor through the training of multiple-factor training sample to be sentenced
Certainly before the step of neural network model, further include：
The training sample that receiving terminal uploads, wherein, the training sample data are multiple-factor training sample；
Under off-line state, using deep learning algorithm to each corresponding training sample of the factor in the multiple-factor training sample
Build deep learning neutral net submodel；
Obtain the incidence relation between the deep learning neutral net submodel of structure；
The deep learning neutral net submodel is merged according to the incidence relation, generation can carry out multiple-factor joint decision
Deep learning neural network model.
4. according to the method described in claim 3, it is characterized in that, the multiple-factor training sample includes video data, image
Data, voice data, text data and network data.
5. according to the method described in claim 2, it is characterized in that, when the hints data received in setting time includes
During a kind of data type, then using the deep learning neutral net submodel of the corresponding factor of the data type to described
Hints data carries out analysis judgement, obtains court verdict.
6. a kind of unstructured data judgment device based on deep learning, it is characterised in that described device includes：
Neural network model acquisition module, for obtaining the deep learning neural network model after training, wherein, after the training
Deep learning neural network model be through multiple-factor training sample train multiple-factor cascading judgement neural network model；
Real time data receiving module, for receiving the hints data of online real time collecting, the hints data is to include a variety of numbers
According to the unstructured data of type；
Characteristic information extracting module, for the clue by the deep learning neural network model after the training to reception
Data carry out Conjoint Analysis, and extraction is conducive to the feature hint information of judgement；
Cascading judgement module, for being made decisions according to the feature hint information to the hints data, generates court verdict；
Court verdict feedback module, for feeding back the court verdict.
7. device according to claim 6, it is characterised in that the characteristic information extracting module, is additionally operable to by described
Deep learning neural network model after training carries out feature extraction to the hints data of reception, obtain high dimensional feature to
Amount；The high dimensional feature vector is changed into by the hash code with each dimension incidence relation using hash search algorithm, and according to
The hash code be conducive to the extraction of the feature hint information of judgement.
8. device according to claim 6, it is characterised in that described device further includes：
Training sample data receiving module, the training sample uploaded for receiving terminal, wherein, the training sample is multiple-factor
Training sample；
Submodel training module, under off-line state, using deep learning algorithm in the multiple-factor training sample
Each corresponding training sample structure deep learning neutral net submodel of the factor；
Incidence relation analysis module, the incidence relation between the deep learning neutral net submodel for obtaining structure；
Joint decision model construction module, for merging the deep learning neutral net submodel according to the incidence relation,
Generation can carry out the deep learning neural network model of multiple-factor joint decision.
9. device according to claim 8, it is characterised in that the multiple-factor training sample includes video data, image
Data, voice data, text data and network data.
10. device according to claim 7, it is characterised in that described device further includes：Submodel judging module, is used for
When the hints data received in setting time includes a kind of data type, then using the corresponding institute of the data type
The deep learning neutral net submodel for stating the factor carries out analysis judgement to the hints data, obtains court verdict.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610950494.9A CN107992937B (en) | 2016-10-26 | 2016-10-26 | Unstructured data judgment method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610950494.9A CN107992937B (en) | 2016-10-26 | 2016-10-26 | Unstructured data judgment method and device based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107992937A true CN107992937A (en) | 2018-05-04 |
| CN107992937B CN107992937B (en) | 2021-12-03 |
Family
ID=62029459
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610950494.9A Expired - Fee Related CN107992937B (en) | 2016-10-26 | 2016-10-26 | Unstructured data judgment method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107992937B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733837A (en) * | 2018-05-28 | 2018-11-02 | 杭州依图医疗技术有限公司 | A kind of the natural language structural method and device of case history text |
| CN109255499A (en) * | 2018-10-25 | 2019-01-22 | 阿里巴巴集团控股有限公司 | Complaint, tip-offs about environmental issues processing method, device and equipment |
| CN110046715A (en) * | 2019-04-24 | 2019-07-23 | 戴文跃 | A kind of clue integrated system basis is constituted |
| CN110412470A (en) * | 2019-04-22 | 2019-11-05 | 上海博强微电子有限公司 | Electric automobile power battery SOC estimation method |
| CN111210218A (en) * | 2018-11-22 | 2020-05-29 | 精工爱普生株式会社 | Information processing apparatus, bill printer, and information processing method |
| CN111400040A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Industrial Internet system based on deep learning and edge calculation and working method |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030033275A1 (en) * | 2001-08-13 | 2003-02-13 | Alpha Shamim A. | Combined database index of unstructured and structured columns |
| US20040148308A1 (en) * | 2003-01-29 | 2004-07-29 | Rajan Rajeev B. | Filestream data storage attribute |
| US20060206656A1 (en) * | 2000-12-22 | 2006-09-14 | Marco Di Benedetto | Apparatus and method for preventing loops in a computer network |
| CN103854483A (en) * | 2014-03-21 | 2014-06-11 | 广东新快易通智能信息发展有限公司 | Vehicle dynamic information system based on GPS (global position system) satellite positioning and ECU (electronic control unit) |
| CN103942240A (en) * | 2013-10-29 | 2014-07-23 | 北京科东电力控制***有限责任公司 | Method for building intelligent substation comprehensive data information application platform |
| WO2014133749A2 (en) * | 2013-03-01 | 2014-09-04 | Inofile Llc | Data capturing and exchange method and system |
| WO2014193757A1 (en) * | 2013-05-28 | 2014-12-04 | Mattersight Corporation | Optimized predictive routing and methods |
| US20150113656A1 (en) * | 2013-10-21 | 2015-04-23 | International Business Machines Corporation | Consistent data masking |
| CN105022835A (en) * | 2015-08-14 | 2015-11-04 | 武汉大学 | Public safety recognition method and system for crowd sensing big data |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105184513A (en) * | 2015-10-16 | 2015-12-23 | 北京恒华伟业科技股份有限公司 | Evaluation result output method and system |
-
2016
- 2016-10-26 CN CN201610950494.9A patent/CN107992937B/en not_active Expired - Fee Related
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060206656A1 (en) * | 2000-12-22 | 2006-09-14 | Marco Di Benedetto | Apparatus and method for preventing loops in a computer network |
| US20030033275A1 (en) * | 2001-08-13 | 2003-02-13 | Alpha Shamim A. | Combined database index of unstructured and structured columns |
| US20040148308A1 (en) * | 2003-01-29 | 2004-07-29 | Rajan Rajeev B. | Filestream data storage attribute |
| WO2014133749A2 (en) * | 2013-03-01 | 2014-09-04 | Inofile Llc | Data capturing and exchange method and system |
| WO2014193757A1 (en) * | 2013-05-28 | 2014-12-04 | Mattersight Corporation | Optimized predictive routing and methods |
| US20150113656A1 (en) * | 2013-10-21 | 2015-04-23 | International Business Machines Corporation | Consistent data masking |
| CN103942240A (en) * | 2013-10-29 | 2014-07-23 | 北京科东电力控制***有限责任公司 | Method for building intelligent substation comprehensive data information application platform |
| CN103854483A (en) * | 2014-03-21 | 2014-06-11 | 广东新快易通智能信息发展有限公司 | Vehicle dynamic information system based on GPS (global position system) satellite positioning and ECU (electronic control unit) |
| CN105022835A (en) * | 2015-08-14 | 2015-11-04 | 武汉大学 | Public safety recognition method and system for crowd sensing big data |
| CN105068661A (en) * | 2015-09-07 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Man-machine interaction method and system based on artificial intelligence |
| CN105184513A (en) * | 2015-10-16 | 2015-12-23 | 北京恒华伟业科技股份有限公司 | Evaluation result output method and system |
Non-Patent Citations (4)
| Title |
|---|
| KEVIN LIN 等: "Deep Learning of Binary Hash Codes for Fast Image Retrieval", 《2015 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS》 * |
| LEI ZHAO 等: "Heterogeneous Features Integration via Semi-supervised Multi-model Deep Networks", 《INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING》 * |
| 张南平 等: "基于模式识别视频搜索技术的研究", 《福建电脑》 * |
| 张清 等: "基于多线索融合的目标跟踪算法及其在DM3730上的实现", 《数据采集与处理》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733837A (en) * | 2018-05-28 | 2018-11-02 | 杭州依图医疗技术有限公司 | A kind of the natural language structural method and device of case history text |
| CN109255499A (en) * | 2018-10-25 | 2019-01-22 | 阿里巴巴集团控股有限公司 | Complaint, tip-offs about environmental issues processing method, device and equipment |
| CN109255499B (en) * | 2018-10-25 | 2021-12-07 | 创新先进技术有限公司 | Complaint and complaint case processing method, device and equipment |
| CN111210218A (en) * | 2018-11-22 | 2020-05-29 | 精工爱普生株式会社 | Information processing apparatus, bill printer, and information processing method |
| CN110412470A (en) * | 2019-04-22 | 2019-11-05 | 上海博强微电子有限公司 | Electric automobile power battery SOC estimation method |
| CN110412470B (en) * | 2019-04-22 | 2021-09-21 | 上海博强微电子有限公司 | SOC estimation method for power battery of electric vehicle |
| CN110046715A (en) * | 2019-04-24 | 2019-07-23 | 戴文跃 | A kind of clue integrated system basis is constituted |
| WO2020216193A1 (en) * | 2019-04-24 | 2020-10-29 | 戴文跃 | Basic configuration of cue integration system |
| CN111400040A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Industrial Internet system based on deep learning and edge calculation and working method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107992937B (en) | 2021-12-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107992937A (en) | Unstructured data decision method and device based on deep learning | |
| Jeong et al. | Image preprocessing for efficient training of YOLO deep learning networks | |
| Tao et al. | An object detection system based on YOLO in traffic scene | |
| CN107330362B (en) | Video classification method based on space-time attention | |
| CN109271539B (en) | Image automatic labeling method and device based on deep learning | |
| CN105678216A (en) | Spatio-temporal data stream video behavior recognition method based on deep learning | |
| WO2021082589A1 (en) | Content check model training method and apparatus, video content check method and apparatus, computer device, and storage medium | |
| CN111738218B (en) | Human body abnormal behavior recognition system and method | |
| CN111310041B (en) | Image-text publishing method, model training method and device and storage medium | |
| CN104680193B (en) | Online objective classification method and system based on quick similitude network integration algorithm | |
| CN106529477A (en) | Video human behavior recognition method based on significant trajectory and time-space evolution information | |
| CN105590099A (en) | Multi-user behavior identification method based on improved convolutional neural network | |
| CN108416288A (en) | The first visual angle interactive action recognition methods based on overall situation and partial situation's network integration | |
| CN108596038A (en) | Erythrocyte Recognition method in the excrement with neural network is cut in a kind of combining form credit | |
| Wang et al. | Multiscale deep alternative neural network for large-scale video classification | |
| CN110705412A (en) | Video target detection method based on motion history image | |
| CN109753884A (en) | A kind of video behavior recognition methods based on key-frame extraction | |
| CN109753853A (en) | One kind being completed at the same time pedestrian detection and pedestrian knows method for distinguishing again | |
| CN107145546A (en) | Monitor video personnel's fuzzy retrieval method based on deep learning | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| CN112417990A (en) | Examination student violation behavior identification method and system | |
| CN109934106A (en) | A kind of user behavior analysis method based on video image deep learning | |
| CN107609509A (en) | A kind of action identification method based on motion salient region detection | |
| CN110008841A (en) | A kind of Expression Recognition model building method and system | |
| CN115880529A (en) | Method and system for classifying fine granularity of birds based on attention and decoupling knowledge distillation |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211203 |