CN107451192A - A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization - Google Patents
A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization Download PDFInfo
- Publication number
- CN107451192A CN107451192A CN201710508428.0A CN201710508428A CN107451192A CN 107451192 A CN107451192 A CN 107451192A CN 201710508428 A CN201710508428 A CN 201710508428A CN 107451192 A CN107451192 A CN 107451192A
- Authority
- CN
- China
- Prior art keywords
- classification
- positive
- sample
- negative
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization, belong to the fields such as data mining, machine learning and business intelligence.First original CDR data are carried out with the horizontal division and sampling of different positive and negative class ratios, for certain training sample, the characteristic attribute for randomly selecting special ratios is used to construct fundamental classifier;To any training sample, according to the output result structural classification matrix of fundamental classifier, the classification results in each same ratio are polymerize, and the voting results under ratio of all categories are determined by maximum ballot method.Secondary classifier is constructed using the classification results in each ratio grader as new characteristic of division, determines weight of the base grader for test result of each positive negative ratio.The present invention is applied to the uneven classification under various big data scenes, avoids the fluctuation of different positive and negative class sample proportion drag precision, and classification results have stronger stability and robustness, it is possible to achieve higher classification and detection efficiency.
Description
Technical field
The invention belongs to data mining, the field such as machine learning and business intelligence, it is specifically a kind of based on decompose polymerization
The classification and Detection method of telecommunication fraud phone.
Background technology
Telecommunication fraud case in China's happens occasionally in recent years, the property safety of serious threat to the people and the stabilization of society.
Because call volume is huge, supervision department is difficult to carry out real-time monitoring with intercepting to all phones, therefore how to utilize data mining
In classification, abnormality detection the methods of realize automation doubtful fraudulent call detection, be one huge for supervision department
Big challenge.
The practical problem of fraudulent call classification and Detection, it is that data volume is larger first, only by taking international call end as an example, daily
Call volume is more than 20,000,000 times;Meanwhile in original data, it is intercepted and the fraudulent call sample of mark is in whole calls
Small portion is only accounted in record so that data category has significant uneven feature.Such as the swindle being detected on a small quantity
Phone is noted as positive class sample, and remaining is largely conversed and is noted as negative class sample, in the note of current overseas call
In record, positive negative ratio has reached 40:1.In fact, the phenomenon of this class imbalance is present in substantial amounts of practical application scene
In, such as network invasion monitoring, credit card fraud detection etc. has in the abnormality detection problem of supervision.
For having the extensive and data of imbalanced class distribution feature concurrently, it is difficult to be trained by unified model.
On the one hand because data volume is excessive, needed to consume substantial amounts of time and space with single model;On the other hand due to data in itself
Uneven feature, using single model can align class sample classification produce poor fitting (under fitting) phenomenon.
Because original data volume is larger, even if positive and negative class proportional imbalance, still has for positive class ratio and largely be available for training
Sample.In this case, rational sampling how is carried out from a large amount of initial data turns into actual rare classification
A major issue in detection.In addition, the detection method of most of telecommunication fraud only lays particular emphasis on single index at present, for example,
Only pursue the accuracy rate of detection, but this kind of method lacks universality for different types of fraudulent call, cause recall rate compared with
It is low.
In fact, because the training sample of different positive and negative classes can have an impact to the various precision of detection model simultaneously,
Therefore need a kind of method for automatically determining optimal positive and negative class training sample come all kinds of indexs, such as accuracy rate, recall rate etc. it
Between to make more rational balance.
The content of the invention
The present invention in view of under imbalanced class distribution big data classification difficulty and challenge, while in view of sample size compared with
Greatly, the characteristics of positive class sample is also more, a kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization is constructed.
Comprise the following steps that:
, will be by Step 1: collect the CDR data (Call Detail Record, call to detailed data) in communication network
A small amount of fraudulent call record of detection is labeled as positive class sample, and remaining is labeled as negative class sample.
Step 2: setting positive and negative sample proportion as X%, CDR data are carried out with laterally continuous sampling and is divided, is sampled repeatedly
L times, obtain the sample set that L positive and negative class ratios are X%.
The random sampling pattern put back to using having, the data that positive and negative class sample proportion is X% are extracted from CDR data and are remembered
Record.
Step 3: changing positive and negative class ratio successively in the way of unique step, carry out A times, common property gives birth to A*L training
Collection.
Unique step refers to that the difference between the adjacent positive and negative class sample proportion value of any two is fixed;
Step 4: carrying out longitudinal decomposition according to characteristic attribute to CDR data, the different category attribute subset of F kinds is obtained;
Specifically, CDR data share M feature, randomly select Y% attributive character, altogether M*Y% attributive character
Based on grader characteristic of division;Sampling F times is extracted in the random sampling put back to by having, and obtains the different classification category of F kinds
Temper collection.
Step 5: original CDR data are divided in order to which A*L*F training sample area, each training field have spy simultaneously
Fixed positive and negative class ratio and characteristic attribute;
Step 6: for the data in each training sample area, using Decision-Tree Classifier Model in subcharacter attribute space
One grader of upper construction, is obtained A*L*F fundamental classifier;
Step 7: for certain training sample in original CDR data, the output point respectively in A*L*F fundamental classifier
Class prediction result, it is configured to the classification matrix of the training sample.
Specifically, each element in matrix has been corresponded under a kind of specific positive negative ratio, in specific attributive character
The classification results of concentration.
Step 8: being directed to the classification matrix, the classification results of laterally identical positive negative ratio are polymerize, and carry out feature
Set screening.
Comprise the following steps that:
Step 801, the classification matrix for training sample output, using comentropy to L under same Classified Proportion
Classifier result is screened;
Step 802, judge whether comentropy is higher than specific threshold e, if it is, removing the output knot of the fundamental classifier
Fruit;Otherwise, into step 803;
Comentropy is higher than specific threshold e, illustrates that the discrimination under this group of subcharacter set for classification is poor.
Step 803, determine classification of the same Classified Proportion on same characteristic features attribute set by the way of maximum is voted
As a result, tieed up so as to which classification results are polymerized into A*F.
Step 9: in the A*F dimensional feature set after screening, determining that the classification of each row is defeated by the way of maximum is voted
Go out result, obtain the classification results of the different positive negative ratios of A kinds.
The A dimensional feature that classification results under the different positive negative ratios of A kinds are regarded as the training sample represents.
Step 10: for all training samples of CDR data, using Decision-Tree Classifier Model, with each training sample
A dimensional features result construction secondary classifier;
Step 11: for a new test sample, repeat step seven arrives step 9, obtains existing on the test data
The A dimensional features of various positive and negative sample proportions, further using secondary classifier, draw the final classification knot of the test sample
Fruit.
Advantage of the invention is that:
1), a kind of sorting technique based on the classification and Detection for decomposing the telecommunication fraud phone with polymerizeing, suitable for various big
Uneven classification under data scene, avoid the fluctuation of different positive and negative class sample proportion drag precision.
2) a kind of, sorting technique based on the classification and Detection for decomposing the telecommunication fraud phone with polymerizeing, passes through same ratio
Between classification results verification and structure and the secondary classifier in the feature of different proportion so that classification results have stronger
Stability and robustness.
3) a kind of, sorting technique based on the classification and Detection for decomposing the telecommunication fraud phone with polymerizeing, is easily realized parallel
Change and calculate, it is possible to achieve higher classification and detection efficiency.
Brief description of the drawings
Fig. 1 is a kind of sorting technique flow based on the classification and Detection for decomposing the telecommunication fraud phone with polymerizeing of the present invention
Figure;
Fig. 2 is the method flow diagram that the present invention carries out characteristic set screening.
Embodiment
The specific implementation method of the present invention is described in detail below in conjunction with the accompanying drawings.
The present invention in view of under imbalanced class distribution big data classification difficulty and challenge, it is contemplated that sample size is larger, just
The characteristics of class sample is also more, construct a kind of sorting technique based on " decomposing with polymerizeing ";It is related in large-scale data
Imbalance classification and abnormality detection, original extensive CDR data are carried out with the horizontal division and sampling of different positive and negative class ratios, together
When for specific training sample, randomly select special ratios characteristic attribute be used for fundamental classifier construction.For
Any training sample, according to the output result structural classification matrix of fundamental classifier, the classification results in each same ratio are entered
Row polymerization, the extremely inconsistent classification results of classification results under same alike result feature are screened out, and pass through maximum ballot method and determine
Voting results under each positive and negative classification ratio.Meanwhile using the classification results in each ratio grader as new characteristic of division,
And secondary classifier is constructed, so that it is determined that weight of the base grader of each positive negative ratio for test result.
As shown in figure 1, comprise the following steps that:
Step 1: collecting the CDR data in communication network, a small amount of fraudulent call being detected record is labeled as positive class sample
This, remaining is labeled as negative class sample.
The CDR data in communication network are collected, a small amount of fraudulent call sample of detection is labeled as positive class sample, remaining
Negative class sample is labeled as, constructs raw data set.
Step 2: setting positive and negative sample proportion as X%, CDR data are carried out with laterally continuous sampling and is divided, is sampled repeatedly
L times, obtain the sample set that L positive and negative class ratios are X%.
Specifically, for specific positive and negative class ratio X%, fixed qty N is extracted in the random sampling put back to by having
=10000 data record.Sample L=20 times repeatedly in this manner, obtain sample that 20 positive and negative class ratios are X%
Collection.
Step 3: changing positive and negative class ratio successively in the way of unique step, carry out A times, common property gives birth to A*L training
Collection.
Unique step refers to that the difference between the adjacent positive and negative class sample proportion value of any two is fixed;
Change positive and negative class ratio successively in the way of unique step from 1% to 50%, and set 1% as step-length, thus altogether
Produce the sample of the different positive negative ratios of A=50 kinds;According to the division methods of step 2, common property life 20*50=1000 is different
Training sample subset.
Step 4: carrying out longitudinal decomposition according to characteristic attribute to CDR data, different category attribute feature of F kinds is obtained
Collection;
Specifically, feature extraction is carried out to original CDR data, has specifically included the feature of caller and called number and led to
The feature of words, 30 features are extracted altogether;In 30 features, 2% attributive character, i.e. 6 attributive character conducts are randomly selected
The characteristic of division of each fundamental classifier, by there is the random sampling put back to extract sampling 20 times, 20 kinds of different classes are obtained
Other attribute set.
Step 5: original CDR data are divided in order to which A*L*F training sample area, each training field have spy simultaneously
Fixed positive and negative class ratio and characteristic attribute;
According to step 2 to four, original CDR data acquisition systems have been divided into 20*50*20=20000 has specific positive and negative class
The sub- training sample area of ratio and characteristic attribute spatially.
Step 6: for the data in each training sample area, using Decision-Tree Classifier Model in subcharacter attribute space
One grader of upper construction, is obtained A*L*F fundamental classifier;
In each training sample area, using Decision-Tree Classifier Model to the data in each training sample area in subcharacter
Structural classification device in attribute space, 20000 fundamental classifiers are obtained.
Step 7: for certain training sample in original CDR data, the output point respectively in A*L*F fundamental classifier
Class prediction result, it is configured to the classification matrix of the training sample.
For any training sample, 20000 classification prediction results, structure are exported respectively in 20000 fundamental classifiers
Into classification matrix.Specifically, each element in matrix has been corresponded under a kind of specific positive negative ratio, special in specific attribute
Levy the classification results in subset.
Step 8: being directed to the classification matrix, the classification results of laterally identical positive negative ratio are polymerize, and carry out feature
Set screening.
First against same positive and negative sample proportion X%, 20000 points are exported respectively in 20000 fundamental classifiers
Class prediction result, comentropy is calculated, remove the classification results corresponding to character subset of the comentropy more than specific threshold e, surplus
The classification results for using maximum turnout to select positive and negative sample proportion as X% in remaining result, so determined from L classification results
One classification results, finally aggregates into A*F result.
As shown in Fig. 2 comprise the following steps that:
Step 801, the classification matrix for training sample output, using comentropy under same positive and negative sample proportion
L classifier result is screened;
For the classification matrix of each sample output, first using comentropy to 20 grader knots under same Classified Proportion
Fruit is screened;
Step 802, judge whether comentropy is higher than specific threshold e, if it is, removing the output knot of the fundamental classifier
Fruit;Otherwise, into step 803;
If comentropy is higher than specific threshold e, illustrates that the discrimination under this group of subcharacter set for classification is poor, then go
Fall the output result of the fundamental classifier.
Step 803, determine same positive and negative sample proportion on same characteristic features attribute set by the way of maximum is voted
Classification results, so as to which classification results are polymerized into A*F dimensions.
Classification results are polymerized to 1000 dimensions in the present embodiment.
Step 9: in the A*F dimensional feature set after screening, determining that the classification of each row is defeated by the way of maximum is voted
Go out result, obtain the classification results of the different positive negative ratios of A kinds.
Determined using maximum ballot under different positive negative ratios, the classification results of each subcharacter attribute space output;A kinds
The A dimensional feature that classification results under different positive negative ratios are regarded as the training sample represents;
For any training sample in the present embodiment, after L=20 screening, the common 20*50=of A*F dimensional features is obtained
1000 dimensions, F=20 are the numbers sampled for the attributive character of certain percentage.By way of maximum is voted, from F=20
The classification results of certain percentage are obtained in secondary sampling, the training sample may finally be obtained under the different positive negative ratios of A=50 kinds
Classification results, and each classification results can be regarded as the character representation of one 50 of sample dimension.
Step 10: for all training samples of CDR data, using Decision-Tree Classifier Model, with each training sample
A dimensional features result construction secondary classifier;
Step 11: for a new test sample, repeat step seven arrives step 9, obtains existing on the test data
The A dimensional features description of various positive and negative sample proportions, further using secondary classifier, draws the final classification of the test sample
As a result.
Claims (4)
- A kind of 1. classification and Detection method based on the telecommunication fraud phone for decomposing polymerization, it is characterised in that comprise the following steps that:Step 1: collecting the CDR data in communication network, a small amount of fraudulent call being detected record is labeled as positive class sample, Remaining is labeled as negative class sample;Step 2: setting positive and negative sample proportion as X%, CDR data are carried out with laterally continuous sampling and is divided, is sampled L times repeatedly, Obtain the sample set that L positive and negative class ratios are X%;Step 3: changing positive and negative class ratio successively in the way of unique step, carry out A times, common property gives birth to A*L training subset;Step 4: carrying out longitudinal decomposition according to characteristic attribute to CDR data, the different category attribute subset of F kinds is obtained;Specifically, CDR data share M feature, randomly select Y% attributive character, altogether M*Y% attributive character conduct The characteristic of division of fundamental classifier;Sampling F times is extracted in the random sampling put back to by having, and obtains different category attribute of F kinds Collection;Step 5: original CDR data are divided for A*L*F training sample area, each training field have simultaneously it is specific just Negative class ratio and characteristic attribute;Step 6: for the data in each training sample area, using Decision-Tree Classifier Model in subcharacter attribute space structure A grader is made, A*L*F fundamental classifier is obtained;Step 7: for certain training sample in original CDR data, it is pre- that output category is distinguished in A*L*F fundamental classifier Result is surveyed, is configured to the classification matrix of the training sample;Specifically, each element in matrix has been corresponded under a kind of specific positive negative ratio, in specific attributive character subset Classification results;Step 8: being directed to the classification matrix, the classification results of laterally identical positive negative ratio are polymerize, and carry out characteristic set Screening;Step 9: in the A*F dimensional feature set after screening, determining that the classification of each row exports knot by the way of maximum vote Fruit, obtains the classification results of the different positive negative ratios of A kinds, and an A dimensional feature for being regarded as the training sample represents;Step 10: for all training samples of CDR data, using Decision-Tree Classifier Model, tieed up with the A of each training sample Characteristic results construct secondary classifier;Step 11: for a new test sample, repeat step seven arrives step 9, obtained on the test data various The A dimensional features of positive and negative sample proportion, further using secondary classifier, draw the final classification results of the test sample.
- 2. a kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization according to claim 1, its feature It is, in described step two, using there is the random sampling pattern put back to, positive and negative class sample proportion is extracted from CDR data is X% data record.
- 3. a kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization according to claim 1, its feature It is, in described step three, unique step refers to that the difference between the adjacent positive and negative class sample proportion value of any two is fixed.
- 4. a kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization according to claim 1, its feature It is, described step eight, the specific method for carrying out characteristic set screening is as follows:Step 801, the classification matrix for training sample output, using comentropy to L classification under same Classified Proportion Device result is screened;Step 802, judge whether comentropy is higher than specific threshold e, if it is, removing the output result of the fundamental classifier;It is no Then, into step 803;Step 803, classification results of the same Classified Proportion on same characteristic features attribute set are determined by the way of maximum is voted, So as to which classification results are polymerized into A*F dimensions.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710508428.0A CN107451192A (en) | 2017-06-28 | 2017-06-28 | A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710508428.0A CN107451192A (en) | 2017-06-28 | 2017-06-28 | A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107451192A true CN107451192A (en) | 2017-12-08 |
Family
ID=60488333
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710508428.0A Pending CN107451192A (en) | 2017-06-28 | 2017-06-28 | A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107451192A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107995370A (en) * | 2017-12-21 | 2018-05-04 | 广东欧珀移动通信有限公司 | Call control method, device and storage medium and mobile terminal |
CN108810290A (en) * | 2018-07-17 | 2018-11-13 | 中国联合网络通信集团有限公司 | A kind of method and system of the identification of fraudulent call |
CN109492026A (en) * | 2018-11-02 | 2019-03-19 | 国家计算机网络与信息安全管理中心 | A kind of Telecoms Fraud classification and Detection method based on improved active learning techniques |
CN111131626A (en) * | 2019-12-20 | 2020-05-08 | 珠海高凌信息科技股份有限公司 | Group harmful call detection method and device based on stream data atlas and readable medium |
CN112116180A (en) * | 2019-06-20 | 2020-12-22 | 中科聚信信息技术(北京)有限公司 | Integrated scoring model generation method and device and electronic equipment |
US11025782B2 (en) | 2018-12-11 | 2021-06-01 | EXFO Solutions SAS | End-to-end session-related call detail record |
-
2017
- 2017-06-28 CN CN201710508428.0A patent/CN107451192A/en active Pending
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107995370A (en) * | 2017-12-21 | 2018-05-04 | 广东欧珀移动通信有限公司 | Call control method, device and storage medium and mobile terminal |
CN108810290A (en) * | 2018-07-17 | 2018-11-13 | 中国联合网络通信集团有限公司 | A kind of method and system of the identification of fraudulent call |
CN108810290B (en) * | 2018-07-17 | 2021-04-02 | 中国联合网络通信集团有限公司 | Method and system for recognizing fraud calls |
CN109492026A (en) * | 2018-11-02 | 2019-03-19 | 国家计算机网络与信息安全管理中心 | A kind of Telecoms Fraud classification and Detection method based on improved active learning techniques |
CN109492026B (en) * | 2018-11-02 | 2021-11-09 | 国家计算机网络与信息安全管理中心 | Telecommunication fraud classification detection method based on improved active learning technology |
US11025782B2 (en) | 2018-12-11 | 2021-06-01 | EXFO Solutions SAS | End-to-end session-related call detail record |
CN112116180A (en) * | 2019-06-20 | 2020-12-22 | 中科聚信信息技术(北京)有限公司 | Integrated scoring model generation method and device and electronic equipment |
CN112116180B (en) * | 2019-06-20 | 2024-05-31 | 中科聚信信息技术(北京)有限公司 | Integrated score model generation method and device and electronic equipment |
CN111131626A (en) * | 2019-12-20 | 2020-05-08 | 珠海高凌信息科技股份有限公司 | Group harmful call detection method and device based on stream data atlas and readable medium |
CN111131626B (en) * | 2019-12-20 | 2022-01-14 | 珠海高凌信息科技股份有限公司 | Group harmful call detection method and device based on stream data atlas and readable medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107451192A (en) | A kind of classification and Detection method based on the telecommunication fraud phone for decomposing polymerization | |
CN106255116A (en) | A kind of recognition methods harassing number | |
CN107506776A (en) | A kind of analysis method of fraudulent call number | |
CN107273920A (en) | A kind of non-intrusion type household electrical appliance recognition methods based on random forest | |
CN103927398A (en) | Microblog hype group discovering method based on maximum frequent item set mining | |
CN107315954A (en) | A kind of file type identification method and server | |
CN104809393B (en) | A kind of support attack detecting algorithm based on popularity characteristic of division | |
CN106385693A (en) | Telecommunication fraud method for virtual number segments | |
CN106843941B (en) | Information processing method, device and computer equipment | |
CN112101635A (en) | Method and system for monitoring electricity utilization abnormity | |
CN107132266A (en) | A kind of Classification of water Qualities method and system based on random forest | |
CN106973038A (en) | Network inbreak detection method based on genetic algorithm over-sampling SVMs | |
CN107038449A (en) | A kind of recognition methods of fraudulent user and device | |
CN105389480A (en) | Multiclass unbalanced genomics data iterative integrated feature selection method and system | |
CN113516228B (en) | Network anomaly detection method based on deep neural network | |
CN109766956B (en) | Exception detection method based on express big data | |
CN115222303B (en) | Industry risk data analysis method and system based on big data and storage medium | |
CN109886284A (en) | Fraud detection method and system based on hierarchical clustering | |
CN111274338A (en) | Pre-outbound user identification method based on mobile big data | |
CN110837532A (en) | Method for detecting electricity stealing behavior of charging pile based on big data platform | |
CN109033322A (en) | A kind of test method and device of multidimensional data | |
CN104217088B (en) | The optimization method and system of operator's mobile service resource | |
CN107966678B (en) | Localization method, electronic device and storage medium based on signal data screening | |
CN110334773A (en) | Model based on machine learning enters the screening technique of modular character | |
CN107195181B (en) | A method of fake-licensed car is identified according to fake-licensed car recognition rule library |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171208 |