CN109582794A - Long article classification method based on deep learning - Google Patents
Long article classification method based on deep learning Download PDFInfo
- Publication number
- CN109582794A CN109582794A CN201811440171.0A CN201811440171A CN109582794A CN 109582794 A CN109582794 A CN 109582794A CN 201811440171 A CN201811440171 A CN 201811440171A CN 109582794 A CN109582794 A CN 109582794A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- text
- long article
- classification
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 15
- 230000006870 function Effects 0.000 claims abstract description 10
- 238000013136 deep learning model Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 3
- 230000006403 short-term memory Effects 0.000 abstract description 3
- 230000015654 memory Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The long article classification method based on deep learning that the invention discloses a kind of, it is extracted at random by pre-process then generation term vector to text, obtained data set is input to convolutional neural networks and lengthens in the model structure of short-term memory network repetition training to frequency of training, loss function is constantly reduced, trained deep learning model is finally obtained.This method, which passes through, to be extracted effective sentence at random and builds the model structure that convolutional neural networks lengthen short-term memory network, can be extracted the feature of long article comprehensively, be guaranteed the accuracy of classification, and can accelerate training speed, improve the efficiency of classification.This method can quickly and correctly obtain classification results, and especially for long text more than 7000 words, and content of text is very close, the article of the careful complexity of classification, and the effect of acquirement is more obvious.
Description
Technical field
The present invention relates to a kind of long article classification methods, more particularly to a kind of long article classification method based on deep learning.
Background technique
With the continuous development of internet and electronic technology, a large amount of paper document is stored in mutually in the form of electronic document
In networking, internet become people propagate information main platform, it is desirable to can according to keyword magnanimity document
The desired information of quick obtaining in data.This requires paper documents specific classification and label.
In traditional classification problem, the form of keyword or key sentence is extracted in people's general choice, is carried out to text
Classification, for the classification problem with clear feature, such as spam filtering, machine automatic question answering etc. has had good
Using.And for technical paper similar in paper document, especially direction, extract key sentence may discrimination it is not high, into
It is easy when row classification so that story label is not clear enough, so to extract more accurate feature, passes through analysis full text, length
Phase memory network (Long-Short Term Memory, abbreviation LSTM) can release subsequent shape according to the state stored before
State.But for the article more than ten thousand words, since data dimension is excessive, only using LSTM to carrying out analysis memory in full, it is easy to
Break the bank makes training become very slowly, analysis result to be caused deviation occur.Convolutional neural networks (Convolutional
Neural Networks, abbreviation CNN) there is the advantages of partially connected, weight is shared, multi-characteristic, it is multiple to greatly reduce calculating
Miscellaneous degree reduces training time and resource, but convolutional neural networks have on learning the dependence before and after long sequence data
There is limitation, for long text training data, convolutional neural networks can accomplish local shape factor, but can not remember longer
Text, the sentence that can not be associated with before and after long article.Cannot comprehensive extraction and analysis feature, there is no assurance that the standard of long article classification
Exactness.Single convolutional neural networks are difficult meet the needs of long text classification.Therefore, in long article analysis field, one kind is found
Can learn full text feature becomes especially to weigh with the training method for guaranteeing accuracy and resource constraint being overcome to improve arithmetic speed
It wants.
Summary of the invention
Goal of the invention: the long article classification method based on deep learning that the technical problem to be solved in the present invention is to provide a kind of,
The problem that local shape factor is not comprehensive enough and memory entire article dimension is too big is overcome in long article analysis field, it can either
Study full text feature can efficiently reduce computational complexity to guarantee the accuracy of classification, reduce the training ground time, improve
The efficiency of classification.
Technical solution: the long article classification method of the present invention based on deep learning, comprising the following steps:
(1) choose and article and obtain text data, and data are pre-processed, for each word generate at random a word to
It measures and is term vector number, the word in text is converted to corresponding term vector, obtains the term vector array file of pure digi-tal；
(2) size for judging every, article words average length, according to every L word for an effective sentence method by long article
Segmentation randomly selects K effectively sentences and forms array X, repeat it is above-mentioned randomly select process n times, obtain data set
(3) above-mentioned data set is input in two-dimensional convolution neural network and carries out feature extraction, it is logical to the data after convolution
It crosses maximum pond method and obtains maximum pond data yi, n times are repeated, the N number of local feature value of long article is obtained
(4) above-mentioned local feature value is input in LSTM neural network, LSTM hide layer number be it is N number of, each layer
Input is local feature value yi, obtain the characteristic parameter of text whole；
(5) features described above parameter is subjected to dropout, abandons partial feature value according to a certain percentage；
(6) data after dropout are obtained into predicted value using softmax algorithmCalculate true tag y and predicted valueBetween loss function H, gradient is reversely updated using Adam bp algorithm, reduces the value of H；
(7) it repeats step (3) to step (6) to be trained, constantly reduces the value of H, protected after reaching the frequency of training of setting
1 deep learning model is deposited, is executed repeatedly, terminates training after whole articles reach traversal number；
(8) trained deep learning model is obtained.
Further, loss function H is to intersect entropy function, expression formula in step (6) are as follows:
Wherein, y is the true tag for inputting document,It is the predicted value of model output.
Text feature is more efficiently extracted using various sizes of convolution nuclear energy, the two-dimensional convolution nerve in step (3)
Network convolutional layer includes the filter of 3 sizes, respectively 3*3,4*4,5*5, and filter depth is 64, step-length 1.
In order to effectively delete garbage in article, the preprocessing process in step (1) includes by acquired textual data
According to deletion document format, punctuation mark, additional character and number.
In order to avoid over-fitting, the ratio in step (5) is 0.5.
In order to also take into account optimization efficiency while complete extraction characteristic value, the traversal number in step (7) is 5
It is secondary.
The utility model has the advantages that this method lengthens short-term memory network by extracting effective sentence at random and building convolutional neural networks
Model structure, the feature of long article can be extracted comprehensively, guarantee classification accuracy, and can with and accelerate training speed, mention
The efficiency of high-class.This method can quickly and correctly obtain classification results, especially for long text more than 7000 words, and
And content of text is very close, the article of the careful complexity of classification, the effect of acquirement is more obvious.In practical applications, this method
Need to save trained model, after inputting a long article document, what system will be fast automatic provides the accurate label of document.
Detailed description of the invention
Fig. 1 is the present embodiment overall flow figure；
Fig. 2 is convolutional neural networks and shot and long term memory network integral frame figure；
Fig. 3 is the accuracy comparison diagram of different models.
Specific embodiment
This method can be not only used for English Text Classification, can be used for the classification of other language texts.Embodiment is with 4 labels
English article for, first from Cornell Univ USA manage Development of Electronic Preprints document databse downloading for training 4 classes not
Same paper data in literature, respectively mathematics (quantum algebra), mathematics (metric geometry), mathematics (algebraic geometry), mathematics are (several
He Xue) totally 38309.
As shown in Figure 1, the method for embodiment includes the next steps:
(1) English papers for the PDF format downloaded are pre-processed.Preprocessing process is to delete not use in text
Information, including the format in document, punctuation mark, additional character, the unconventional English word such as number.
(2) dictionary is established according to the English word in data, as each word generates a random term vector and is word
Vector numbers ultimately generate the array file for the pure digi-tal being made of term vector number.Each array file stores a text
Chapter, the article of same class label are stored under the file of label name.
(3) random extraction is done to full text.It is one according to averagely every 20 words according to the statistical result of data
10,000 English words of one article are divided into 500 valid sentences by words, i.e. L=20.The part of less than 10,000 word of long article is with spot patch
Fill, according to the actual length of article in data set, randomly select 20 valid sentences, i.e. K=20, convolutional layer will capture simultaneously this 20
The feature of a sentence.To sentence repeated sampling 25 times in same piece lengthy document, i.e. N=25 is defeated if Fig. 3 randomly selects process
Enter text and be divided into N group, is expressed as
(4) divide training set data and verifying collection data according to the ratio of 9:1.34479 articles of training set are used for mould
The training optimization of type, 3830 articles for verifying collection are used to verify the actual classification effect of model after optimization.
(5) data of training set are input to deep learning model.As shown in Fig. 2, convolutional neural networks are first to input
25 groups of valid sentences of article do local shape factor, and it is 3*3,4* respectively that the present embodiment, which sets 3 kinds of various sizes of filters,
4,5*5, filter depth is 64, step-length 1.Text feature is more efficiently extracted using various sizes of convolution nuclear energy,
Then pond, y are done to the data after convolution by the method in maximum pondiRepresent the maximum pond number of various sizes of convolution kernel
According to integrating each group pond data, every article will obtain 25 groups of local feature values and be expressed as
(6) these local feature amounts are inputed into LSTM neural network, to remember full text context, obtains higher
Secondary abstract characteristics.The input that each layer of LSTM is that the quantity of CNN local feature value yi, LSTM hidden layer and N value above are protected
Consistent i.e. 25 are held, by LSTM, the characteristic parameter of text whole is obtained, is input to full articulamentum, 0.5 dropout ratio is set
The random partial data of rejecting of example avoids over-fitting, and the method for extracting traversal full text random in this way ensure that and effectively extract article
Full content improves the ability of neural network analysis memory full text.
(7) data after dropout are finally obtained into predicted value using softmax algorithmWith inputting the true classification of text
Label y is compared.It calculates it and intersects entropy functionDefinition H is loss function,
In, y is the input true label classification of document,It is the predicted value of model output, y is number.Four classes have been used in experiment in patent
Document: mathematics (quantum algebra), mathematics (metric geometry), mathematics (algebraic geometry), mathematics (geometry).Experimental mathematics (amount
Subalgebra) label classification be 1, mathematics (metric geometry) label classification be 2, mathematics (algebraic geometry) be 3, mathematics (geometry)
It is 4.It in experiment, directly numbers in order, text is converted to number, calculated so that computer is read.Using Adam bp algorithm
The reversed gradient that updates reduces the value of H, to reduce label y and predicted valueBetween gap, improve accuracy.Training set data is every
A training deep learning model of 100 preservations, including saving multiple weight parameters in convolutional neural networks and LSTM network
W, offset parameter b, and calculate the 100th loss function and accuracy.The present embodiment is tied after selecting whole articles to traverse 5 times
Shu Xunlian, i.e. repeatedly above step 172395 times.The accuracy of all loss functions for having saved network and text classification is observed,
The highest network of accuracy optimal models the most are chosen, trained deep learning model is obtained.
It is finally verified, above-mentioned verifying collection data is input in the network, be verified the prediction classification of collection article,
It compares, obtain classification accuracy rate and saves with story label, the classification capacity of assessment models.Fig. 3 show different parameters with
And the comparison of different category of model accuracy, the accuracy of training set text is illustrated as tendency chart, and with other classical taxonomies
The result of method compares, it can be seen that the accuracy of one side transcript analysis is more more effective than local key word analysis.It is another
Aspect can be seen that under identical frequency of training, and the classification accuracy rate of single CNN and LSTM model is learned than this patent depth
It is much lower to practise model.For the accuracy of this method training set text 99% or so, the accuracy rate of verifying collection text reaches 94%.
Experiments have shown that this method quickly can accurately obtain the tag along sort of long article.
Among practical application, by text input to be sorted into above-mentioned trained deep learning model, lead to
The accurate tag along sort of text can be quickly obtained automatically by crossing this model.
Claims (6)
1. a kind of long article classification method based on deep learning, it is characterised in that the following steps are included:
(1) it chooses article and obtains text data, and data are pre-processed, generate a term vector at random simultaneously for each word
For term vector number, the word in text is converted to corresponding term vector, obtains the term vector array file of pure digi-tal；
(2) size for judging every, article words average length divides long article according to the method that every L word is an effective sentence
Cut, randomly select K effectively sentences and form array X, repeat it is above-mentioned randomly select process n times, obtain data set
(3) above-mentioned data set is input in two-dimensional convolution neural network and carries out feature extraction, to the data after convolution by most
Great Chiization method obtains maximum pond data yi, n times are repeated, the N number of local feature value of long article is obtained
(4) above-mentioned local feature value is input in LSTM neural network, it is N number of, each layer of input that LSTM, which hides layer number,
For local feature value yi, obtain the characteristic parameter of text whole；
(5) features described above parameter is subjected to dropout, abandons partial feature value according to a certain percentage；
(6) data after dropout are obtained into predicted value using softmax algorithmCalculate true tag y and predicted valueBetween
Loss function H, gradient is reversely updated using Adam bp algorithm, reduces the value of H；
(7) it repeats step (3) to step (6) to be trained, constantly reduces the value of H, saved 1 time after reaching the frequency of training of setting
Deep learning model, executes repeatedly, terminates training after whole articles reach traversal number；
(8) trained deep learning model is obtained.
2. the long article classification method according to claim 1 based on deep learning, it is characterised in that: loss in step (6)
Function H is to intersect entropy function, expression formula are as follows:
Wherein, y is the true tag for inputting document,It is the predicted value of model output.
3. the long article classification method according to claim 1 based on deep learning, it is characterised in that: two in step (3)
Dimension convolutional neural networks convolutional layer includes the filter of 3 sizes, and respectively 3*3,4*4,5*5, filter depth are 64, step
A length of 1.
4. the long article classification method according to claim 1 based on deep learning, it is characterised in that: pre- in step (1)
Treatment process includes that acquired text data is deleted document format, punctuation mark, additional character and number.
5. the long article classification method according to claim 1 based on deep learning, it is characterised in that: the ratio in step (5)
Example is 0.5.
6. the long article classification method according to claim 1 based on deep learning, it is characterised in that: time in step (7)
All previous number is 5 times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811440171.0A CN109582794A (en) | 2018-11-29 | 2018-11-29 | Long article classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811440171.0A CN109582794A (en) | 2018-11-29 | 2018-11-29 | Long article classification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109582794A true CN109582794A (en) | 2019-04-05 |
Family
ID=65925069
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811440171.0A Pending CN109582794A (en) | 2018-11-29 | 2018-11-29 | Long article classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109582794A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110457469A (en) * | 2019-07-05 | 2019-11-15 | 中国平安财产保险股份有限公司 | Information classification approach, device based on shot and long term memory network, computer equipment |
| CN110532448A (en) * | 2019-07-04 | 2019-12-03 | 平安科技（深圳）有限公司 | Document Classification Method, device, equipment and storage medium neural network based |
| CN110609898A (en) * | 2019-08-19 | 2019-12-24 | 中国科学院重庆绿色智能技术研究院 | Self-classification method for unbalanced text data |
| CN110633470A (en) * | 2019-09-17 | 2019-12-31 | 北京小米智能科技有限公司 | Named entity recognition method, device and storage medium |
| CN110879934A (en) * | 2019-10-31 | 2020-03-13 | 杭州电子科技大学 | Efficient Wide & Deep learning model |
| CN111538840A (en) * | 2020-06-23 | 2020-08-14 | 基建通(三亚)国际科技有限公司 | Text classification method and device |
| CN112069379A (en) * | 2020-07-03 | 2020-12-11 | 中山大学 | Efficient public opinion monitoring system based on LSTM-CNN |
| CN112133441A (en) * | 2020-08-21 | 2020-12-25 | 广东省人民医院 | Establishment method and terminal of MH post-operation fissure hole state prediction model |
| CN112418354A (en) * | 2020-12-15 | 2021-02-26 | 江苏满运物流信息有限公司 | Goods source information classification method and device, electronic equipment and storage medium |
| CN113239190A (en) * | 2021-04-27 | 2021-08-10 | 天九共享网络科技集团有限公司 | Document classification method and device, storage medium and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169035A (en) * | 2017-04-19 | 2017-09-15 | 华南理工大学 | A kind of file classification method for mixing shot and long term memory network and convolutional neural networks |
| US20170308790A1 (en) * | 2016-04-21 | 2017-10-26 | International Business Machines Corporation | Text classification by ranking with convolutional neural networks |
| CN108875021A (en) * | 2017-11-10 | 2018-11-23 | 云南大学 | A kind of sentiment analysis method based on region CNN-LSTM |
-
2018
- 2018-11-29 CN CN201811440171.0A patent/CN109582794A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170308790A1 (en) * | 2016-04-21 | 2017-10-26 | International Business Machines Corporation | Text classification by ranking with convolutional neural networks |
| CN107169035A (en) * | 2017-04-19 | 2017-09-15 | 华南理工大学 | A kind of file classification method for mixing shot and long term memory network and convolutional neural networks |
| CN108875021A (en) * | 2017-11-10 | 2018-11-23 | 云南大学 | A kind of sentiment analysis method based on region CNN-LSTM |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021000411A1 (en) * | 2019-07-04 | 2021-01-07 | 平安科技（深圳）有限公司 | Neural network-based document classification method and apparatus, and device and storage medium |
| CN110532448A (en) * | 2019-07-04 | 2019-12-03 | 平安科技（深圳）有限公司 | Document Classification Method, device, equipment and storage medium neural network based |
| CN110457469A (en) * | 2019-07-05 | 2019-11-15 | 中国平安财产保险股份有限公司 | Information classification approach, device based on shot and long term memory network, computer equipment |
| CN110609898A (en) * | 2019-08-19 | 2019-12-24 | 中国科学院重庆绿色智能技术研究院 | Self-classification method for unbalanced text data |
| CN110633470A (en) * | 2019-09-17 | 2019-12-31 | 北京小米智能科技有限公司 | Named entity recognition method, device and storage medium |
| CN110879934A (en) * | 2019-10-31 | 2020-03-13 | 杭州电子科技大学 | Efficient Wide & Deep learning model |
| CN110879934B (en) * | 2019-10-31 | 2023-05-23 | 杭州电子科技大学 | Text prediction method based on Wide & Deep learning model |
| CN111538840A (en) * | 2020-06-23 | 2020-08-14 | 基建通(三亚)国际科技有限公司 | Text classification method and device |
| CN112069379A (en) * | 2020-07-03 | 2020-12-11 | 中山大学 | Efficient public opinion monitoring system based on LSTM-CNN |
| CN112133441A (en) * | 2020-08-21 | 2020-12-25 | 广东省人民医院 | Establishment method and terminal of MH post-operation fissure hole state prediction model |
| CN112133441B (en) * | 2020-08-21 | 2024-05-03 | 广东省人民医院 | Method and terminal for establishing MH postoperative crack state prediction model |
| CN112418354A (en) * | 2020-12-15 | 2021-02-26 | 江苏满运物流信息有限公司 | Goods source information classification method and device, electronic equipment and storage medium |
| CN112418354B (en) * | 2020-12-15 | 2022-07-15 | 江苏满运物流信息有限公司 | Goods source information classification method and device, electronic equipment and storage medium |
| CN113239190A (en) * | 2021-04-27 | 2021-08-10 | 天九共享网络科技集团有限公司 | Document classification method and device, storage medium and electronic equipment |
| CN113239190B (en) * | 2021-04-27 | 2024-02-20 | 天九共享网络科技集团有限公司 | Document classification method, device, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109582794A (en) | Long article classification method based on deep learning | |
| CN106502985B (en) | neural network modeling method and device for generating titles | |
| Song et al. | Research on text classification based on convolutional neural network | |
| CN109886020A (en) | Software vulnerability automatic classification method based on deep neural network | |
| CN111767725B (en) | Data processing method and device based on emotion polarity analysis model | |
| CN109376242A (en) | Text classification algorithm based on Recognition with Recurrent Neural Network variant and convolutional neural networks | |
| CN106815369A (en) | A kind of file classification method based on Xgboost sorting algorithms | |
| CN110502753A (en) | A kind of deep learning sentiment analysis model and its analysis method based on semantically enhancement | |
| CN108388554B (en) | Text emotion recognition system based on collaborative filtering attention mechanism | |
| CN112231562A (en) | Network rumor identification method and system | |
| CN107832458A (en) | A kind of file classification method based on depth of nesting network of character level | |
| CN108664512B (en) | Text object classification method and device | |
| CN110069627A (en) | Classification method, device, electronic equipment and the storage medium of short text | |
| CN107688576B (en) | Construction and tendency classification method of CNN-SVM model | |
| CN111046183A (en) | Method and device for constructing neural network model for text classification | |
| CN105787121B (en) | A kind of microblogging event summary extracting method based on more story lines | |
| CN110826298B (en) | Statement coding method used in intelligent auxiliary password-fixing system | |
| CN109918507B (en) | textCNN (text-based network communication network) improved text classification method | |
| CN107357895B (en) | Text representation processing method based on bag-of-words model | |
| CN112070139A (en) | Text classification method based on BERT and improved LSTM | |
| CN113590764A (en) | Training sample construction method and device, electronic equipment and storage medium | |
| CN112148868A (en) | Law recommendation method based on law co-occurrence | |
| CN114996467A (en) | Knowledge graph entity attribute alignment algorithm based on semantic similarity | |
| Huang | A CNN model for SMS spam detection | |
| CN111046177A (en) | Automatic arbitration case prejudging method and 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 | ||
| CB02 | Change of applicant information |
Address after: 210044 No. 219 Ningliu Road, Jiangbei New District, Nanjing City, Jiangsu Province Applicant after: Nanjing University of Information Science and Technology Address before: 211500 Yuting Square, 59 Wangqiao Road, Liuhe District, Nanjing City, Jiangsu Province Applicant before: Nanjing University of Information Science and Technology |
|
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190405 |
|
| RJ01 | Rejection of invention patent application after publication |