CN106227718A - Land based on CNN sky call semantic consistency method of calibration - Google Patents
Land based on CNN sky call semantic consistency method of calibration Download PDFInfo
- Publication number
- CN106227718A CN106227718A CN201610573975.2A CN201610573975A CN106227718A CN 106227718 A CN106227718 A CN 106227718A CN 201610573975 A CN201610573975 A CN 201610573975A CN 106227718 A CN106227718 A CN 106227718A
- Authority
- CN
- China
- Prior art keywords
- statement
- rehearsing
- semantic
- term vector
- word
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- 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
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Machine Translation (AREA)
Abstract
A kind of land based on CNN sky call semantic consistency method of calibration.It include picking out from the sky calling record of real land difference rehearse statement to and constitute the positive sample data of semantic congruence;Concurrently form semantic inconsistent negative sample data, positive sample data and negative sample data constitute corpus;By statement of rehearsing each in corpus to participle;Each word is changed into term vector, is made up of term vector storehouse term vector;Each word is formed the matrix of correspondence: using the matrix of each statement of rehearsing as the input of convolutional neural networks input layer, form two semantic vectors of statement pair of rehearsing;Two semantic vectors of statement pair of rehearsing are calculated cosine similarity: utilize k-nearest neighbor to carry out the steps such as classification.Present invention utilizes the advantage of the parameter sharing of convolutional neural networks, it is possible to reduce the operation time, and statement entirety can be processed the semantic vector of computing statement when.
Description
Technical field
The invention belongs to the semantic consistency calibration technology field of land sky call in AIRLINE & AIRPORT, particularly relate to a kind of base
Land sky call semantic consistency method of calibration in CNN.
Background technology
During aircraft, in order to ensure the safe efficient operation of airborne vehicle, air traffic controller (is called for short
" Air Traffic Administers ") and pilot allow for accurately and timely understanding the intention of both sides, thus ensure that the instruction of navigation can be accurate
Inerrably pass on.Pilot needs to repeat the instruction of Air Traffic Administers, and to guarantee the correctness that instruction is passed on, this process is referred to as multiple
Read aloud.And one of correct important means being to ensure that flight safety of rehearsing.Therefore, the accuracy that raising pilot rehearses can have
Effect ground reduces aviation accident and the generation of event.For the accuracy rehearsed of guarantee, International Civil Aviation Organization (ICAO) sets up and not
Break and improve a series of transmission standard and measure.But the aviation accident occurred owing to rehearsing and time is not because of science and technology
Development or transmission standard improve and reduce.Rehearse mistake reason in addition to having mechanical factor, the most artificial reason.
Owing to work, pilot and Air Traffic Administers can produce the mistake rehearsed because of health or the reason of spirit for a long time, therefore
If can utilize current technology that the concordance of semanteme of rehearsing is identified in this process, and by anti-for the result identified of rehearsing
Feed both sides, can reduce the mistake rehearsed undoubtedly.
Convolutional neural networks (CNN) in degree of depth study has well application at computer vision field, in the last few years
The most constantly it is used in the task of natural language processing, such as machine translation etc..Owing to convolutional neural networks has parameter sharing
Advantage, therefore compared to additive method consume time few, meet the feature of promptness of rehearsing.But do not find volume at present
The correlation technique of the model of long-pending neutral net semantic consistency in terms of the sky call of land.
Summary of the invention
In order to solve the problems referred to above, it is an object of the invention to provide a kind of land based on CNN sky call semantic consistency
Method of calibration.
In order to achieve the above object, land based on the CNN sky call semantic consistency method of calibration that the present invention provides includes
The following step carried out in order:
1) from the sky calling record of real land, quality is picked out good and meet the difference of land sky transmission standard and rehearse statement
Right, each statement of rehearsing forms by two statements of rehearsing, and is then store as text txt form, by these statements of rehearsing to structure
Become the positive sample data of semantic congruence;Simultaneously the most rule of thumb with the form of positive sample, professional be artificially formed semanteme and differ
The negative sample data caused, are constituted corpus by positive sample data and negative sample data;
2) separate with space between the word of statement centering of rehearsing each in above-mentioned corpus, with to statement of rehearsing to entering
Row participle;
3) above-mentioned each word through participle is changed into term vector, the term vector of all words constitute term vector
Storehouse;
4) each word in above-mentioned term vector storehouse is formed corresponding matrix:
Above-mentioned each word is searched in term vector storehouse corresponding term vector, the most again term vector is being answered according to word
Read aloud in statement occur sequentially form the matrix that can represent statement of rehearsing, each statement of rehearsing in such corpus can shape
Become a matrix;
5) using the matrix of above-mentioned each statement of rehearsing as the input of convolutional neural networks input layer, identical mistake is carried out respectively
The training study of journey, forms two semantic vectors of statement pair of rehearsing, represents the instruction language of statement centering Air Traffic Administers of rehearsing respectively
The semanteme of sentence and pilot rehearse the semanteme of statement;
6) two semantic vectors to each statement pair of rehearsing of above-mentioned formation utilize formula calculate cosine similarity:
7) utilize k-nearest neighbor to classify according to the cosine similarity of the above-mentioned each statement pair of rehearsing calculated, be divided into
One class of semantic similitude and the inconsistent class of semanteme.
In step 2) in, described participle uses the ICTCLAS of the Chinese Academy of Sciences to analyze system.
In step 3) in, the described term vector that changed into by each word through participle is to utilize word2vec instrument to enter
OK, each word forms the term vector of 50 dimensions.
In step 6) in, the computing formula of described cosine similarity is:
Wherein, y (ATC) represents the semantic vector of Air Traffic Administers directive statement, and y (Pilot) represents pilot and rehearses statement
Semantic vector.
In step 7) in, the described method utilizing k-nearest neighbor to carry out classifying is: first calculate to be sorted in corpus rehearsing
Statement, to the COS distance to other each statements pair of rehearsing, then takes out the distance statement of rehearsing less than setpoint distance threshold value
Right, these statements of rehearsing, to being looped around statement pair of rehearsing nearest around statement of rehearsing to be sorted according to distance threshold exactly, select
Go out these statement centering ratios of rehearsing maximum rehearse statement to bunch class, then these statements of rehearsing are to being just attributed to this class.
Land based on the CNN sky call semantic consistency method of calibration that the present invention provides make use of the ginseng of convolutional neural networks
The advantage that number is shared, it is possible to reduce the operation time, and statement entirety can be carried out the semantic vector of computing statement when
Process.
Accompanying drawing explanation
Land based on the CNN sky call semantic consistency method of calibration flow chart that Fig. 1 provides for the present invention.
Fig. 2 is convolutional neural networks basic framework schematic diagram.
Fig. 3 is pond layer calculating process schematic diagram, and wherein (a) is the output of convolutional layer, and (b) is the output of pond layer.
Fig. 4 is the convolutional neural networks schematic diagram of a kind of structure that the present invention uses.
Fig. 5 is the convolutional neural networks schematic diagram of the another kind of structure that the present invention uses.
Detailed description of the invention
Land based on the CNN sky call semantic consistency verification with specific embodiment, the present invention provided below in conjunction with the accompanying drawings
Method is described in detail.
As it is shown in figure 1, land based on the CNN sky call semantic consistency method of calibration that the present invention provides includes entering in order
The following step of row:
1) from the sky calling record of real land, quality is picked out good and meet the difference of land sky transmission standard and rehearse statement
Right, each statement of rehearsing forms by two statements of rehearsing, and is then store as text txt form, by these statements of rehearsing to structure
Become the positive sample data of semantic congruence;Simultaneously the most rule of thumb with the form of positive sample, professional be artificially formed semanteme and differ
The negative sample data caused, are constituted corpus by positive sample data and negative sample data;
2) separate with space between the word of statement centering of rehearsing each in above-mentioned corpus, with to statement of rehearsing to entering
Row participle;
The purpose of participle is for the ease of processing.The present invention uses the ICTCLAS of the Chinese Academy of Sciences to analyze system by above-mentioned language material
In storehouse, the word of each statement centering of rehearsing carries out participle, and word segmentation result is as shown in the table:
Table 1 is to the statement result to carrying out participle of rehearsing
3) above-mentioned each word through participle is changed into term vector, the term vector of all words constitute term vector
Storehouse;
The present invention utilizes word2vec instrument that the word through participle carries out the conversion of term vector, and each word forms 50
The term vector of dimension.
4) each word in above-mentioned term vector storehouse is formed corresponding matrix:
Above-mentioned each word is searched in term vector storehouse corresponding term vector, the most again term vector is being answered according to word
Read aloud in statement occur sequentially form the matrix that can represent statement of rehearsing, each statement of rehearsing in such corpus can shape
Become a matrix.The word number of one statement of rehearsing of statement centering of assuming to rehearse is n, and each word can be in term vector storehouse
In find corresponding 50 dimension term vectors, then this statement of rehearsing just can form the matrix of n*50 size.
5) using the matrix of above-mentioned each statement of rehearsing as the input of convolutional neural networks input layer, identical mistake is carried out respectively
The training study of journey, forms two semantic vectors of statement pair of rehearsing, represents the instruction language of statement centering Air Traffic Administers of rehearsing respectively
The semanteme of sentence and pilot rehearse the semanteme of statement.
The basic theories of 5.1 convolutional neural networks
Convolutional neural networks is the one of degree of depth study, and its ordinary construction is as shown in Figure 2.
5.1.1 convolutional layer
The Internet that convolutional layer is made up of convolutional layer neuron.In convolutional layer, input data or last layer obtain
Feature convolution kernel with fixed size in this layer carry out convolution, the output after excitation function of the result after convolution
It is the formation of the output characteristic of this convolutional layer.The output characteristic of general each layer all output with last layer has relation.Convolution
The computing that carried out of layer can use below equation:
Represent the output of the jth neuron of l layer;σ: representing excitation function, used herein is sigmoid letter
Number;Mj: represent the selection of input feature vector;K: represent convolution kernel;B: represent biasing.
The convolution kernel size of each neuron of convolutional layer and input data square neighborhood be connected, and step-length
It is defaulted as 1, so for the data that input is n*m, each feature sizes formed in convolutional layer output is (n-convolution kernel size
+ 1) * (m-convolution kernel size+1).
5.1.2 pond layer
The Internet that pond layer is made up of pond layer neuron.In the structure of general convolutional neural networks, convolution
Layer and pond layer are all to occur in pairs.After obtaining feature by the convolutional layer of last layer, next step is exactly special to these
Levy and carry out similar secondary feature extraction.The convolutional layer output of last layer is as the input of next layer of pond layer, to pond layer
Averaging in the region of the squared magnitude of size or take maximum, this pondization processes and is possible not only to reduce dimension, and
Result (being not easy over-fitting) can be improved.If input being done maximum pond, it is simply that to the number in the square area of pond layer size
According to taking maximum, average exactly if doing average pondization, and two kinds of Chi Huadou are nonoverlapping ponds.So pond layer
The size of each feature formed is (n-convolution kernel size+1)/pond layer size * (m-convolution kernel size+1)/pond layer
Size.Fig. 3 is shown that the process in pond in the case of layer size=3, pond.
The structure that the present invention uses is two-layer convolutional layer and two-layer pond layer, convolutional layer and pond layer are alternately present, by defeated
Enter data input and carry out twice convolution and pond to convolutional neural networks, the semanteme that the vector of formation is referred to as in semantic space to
Amount.Each layer all comprises trainable parameter.The input data of this network are the matrixes of n*m, and the vector finally exported is for rehearsing
The semantic vector of statement, is used for characterizing the semanteme of statement of rehearsing.
5.2 convolutional neural networks model structures
Following two convolutional neural networks structure is mainly used to verify the feasibility of this method:
5.2.1 the first structure
The first structure is 4 layers of convolutional neural networks, i.e. level 2 volume lamination, 2 layers of pond layer, and the output of last output layer is corresponding
Semantic vector, as shown in Figure 4.
Wherein SATCRepresent the statement matrix of rehearsing of Air Traffic Administers instruction, SpRepresent the statement matrix of rehearsing that pilot rehearses, volume
Lamination 1 and convolutional layer 2 represent first and second convolutional layers, average pond layer 1 and average pond layer 2 respectively and represent first respectively
With second average pond layer, the V eventually formedATCRepresent the semantic vector of Air Traffic Administers instruction, VpRepresent the language that pilot rehearses
Justice vector.The most finally rehearse statement to forming corresponding two semantic vectors.
5.2.2 the second structure
In the present invention, owing to using the corpus arrived little, and term vector only has 50 dimensions, therefore have employed another kind
Structure, as shown in Figure 5.
6) two semantic vectors to each statement pair of rehearsing of above-mentioned formation utilize equation below calculating cosine similar
Degree:
Wherein, y (ATC) represents the semantic vector of Air Traffic Administers directive statement, and y (Pilot) represents pilot and rehearses statement
Semantic vector.
7) utilize k-nearest neighbor to classify according to the cosine similarity of the above-mentioned each statement pair of rehearsing calculated, be divided into
One class of semantic similitude and the inconsistent class of semanteme.
The method utilizing k-nearest neighbor to carry out classifying is: first calculate in corpus statement of rehearsing to be sorted to each to other
Rehearse the COS distance of statement pair, then take out the distance statement pair of rehearsing less than setpoint distance threshold value, these statements pair of rehearsing
It is exactly to be looped around statement pair of rehearsing nearest around statement of rehearsing to be sorted according to distance threshold, selects these statement centerings of rehearsing
What ratio was maximum rehearse statement to bunch class, then these statements of rehearsing are to being just attributed to this class.
Experiment and analysis
In order to verify the effect of the inventive method, the present inventor selects 300 to belonging to positive sample number from corpus at random
According to statement and 200 of rehearsing to belonging to negative sample data statement of rehearsing as training data, simultaneously by 200 to belonging to positive sample
The statement and 100 of rehearsing of data rehearses statement as test data to belonging to negative sample data.
1. evaluation criterion:
(1) precision (Precision), can reflect the identification ability to negative sample data, count according to equation below
Calculate:
Precision=TP/ (TP+FP) (3)
(2) recall rate (Recall), characterizes the identification ability aligning sample data, calculates according to formula below:
Recall=TP/ (TP+FN) (4)
(3) F1 value, can reflect the sane degree of structure in convolutional neural networks:
F1=2 Recall Precision/ (Recall+Precision) (5)
Wherein, the sample size of correct classification during TP is positive sample data;FP is the sample of mistake classification in positive sample data
This quantity;FN is the sample size of mistake classification in negative sample data.
2. in the present invention, each experiment is carried out 50 times, and statistical average result is as shown in table 2:
Table 2 experimental result
3. conclusion
From experimental result above it can be seen that the effect of the convolutional neural networks experiment of two kinds of structures is essentially the same, but
It is from precision and F1 value it can be seen that the first structure is more preferable to the identification ability of negative sample data, the knot of convolutional neural networks
Structure is the most sane.From recall rate it can be seen that the second structure is the most better to the identification ability of negative sample data.The first
The testing time of structure is shorter, and the pond layer being primarily due in this structure decreases the dimension of intermediate layer handles data, from
And make amount of calculation reduce, because this time is short.
Claims (5)
1. land based on a CNN sky call semantic consistency method of calibration, it is characterised in that: described land based on CNN is empty
Call semantic consistency method of calibration includes the following step of carrying out in order:
1) from the sky calling record of real land, quality is picked out good and meet the difference of land sky transmission standard and rehearse statement pair, often
Individual statement of rehearsing forms by two statements of rehearsing, and is then store as text txt form, these statement of rehearsing is to constituting semanteme
Consistent positive sample data;Simultaneously the most rule of thumb with the form of positive sample, professional be artificially formed semantic inconsistent negative
Sample data, is constituted corpus by positive sample data and negative sample data;
2) separate with space between the word of statement centering of rehearsing each in above-mentioned corpus, with to statement of rehearsing to carrying out point
Word;
3) above-mentioned each word through participle is changed into term vector, the term vector of all words constitute term vector storehouse;
4) each word in above-mentioned term vector storehouse is formed corresponding matrix:
Above-mentioned each word is searched in term vector storehouse corresponding term vector, the most again by term vector according to word at language of rehearsing
Occur in Ju sequentially forms the matrix that can represent statement of rehearsing, and each statement of rehearsing in such corpus can form one
Individual matrix;
5) using the matrix of above-mentioned each statement of rehearsing as the input of convolutional neural networks input layer, identical process is carried out respectively
Training study, forms two semantic vectors of statement pair of rehearsing, represents the directive statement of statement centering Air Traffic Administers of rehearsing respectively
Semantic and pilot rehearses the semanteme of statement;
6) two semantic vectors to each statement pair of rehearsing of above-mentioned formation utilize formula calculate cosine similarity:
7) utilize k-nearest neighbor to classify according to the cosine similarity of the above-mentioned each statement pair of rehearsing calculated, be divided into semanteme
A similar class and the inconsistent class of semanteme.
Land based on CNN the most according to claim 1 sky call semantic consistency method of calibration, it is characterised in that: in step
Rapid 2), in, described participle uses the ICTCLAS of the Chinese Academy of Sciences to analyze system.
Land based on CNN the most according to claim 1 sky call semantic consistency method of calibration, it is characterised in that: in step
Rapid 3), in, the described term vector that changed into by each word through participle is to utilize word2vec instrument to carry out, each word
Form the term vector of 50 dimensions.
Land based on CNN the most according to claim 1 sky call semantic consistency method of calibration, it is characterised in that: in step
Rapid 6), in, the computing formula of described cosine similarity is:
Wherein, y (ATC) represents the semantic vector of Air Traffic Administers directive statement, and y (Pilot) represents pilot and rehearses the semanteme of statement
Vector.
Land based on CNN the most according to claim 1 sky call semantic consistency method of calibration, it is characterised in that: in step
Rapid 7) in, the described method utilizing k-nearest neighbor to carry out classifying is: first calculate in corpus statement of rehearsing to be sorted to arriving other
The COS distance of each statement pair of rehearsing, then takes out the distance statement pair of rehearsing less than setpoint distance threshold value, these languages of rehearsing
Sentence, to being looped around statement pair of rehearsing nearest around statement of rehearsing to be sorted according to distance threshold exactly, selects these statements of rehearsing
What centering ratio was maximum rehearse statement to bunch class, then these statements of rehearsing are to being just attributed to this class.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610573975.2A CN106227718A (en) | 2016-07-18 | 2016-07-18 | Land based on CNN sky call semantic consistency method of calibration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610573975.2A CN106227718A (en) | 2016-07-18 | 2016-07-18 | Land based on CNN sky call semantic consistency method of calibration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106227718A true CN106227718A (en) | 2016-12-14 |
Family
ID=57531725
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610573975.2A Pending CN106227718A (en) | 2016-07-18 | 2016-07-18 | Land based on CNN sky call semantic consistency method of calibration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106227718A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107704563A (en) * | 2017-09-29 | 2018-02-16 | 广州多益网络股份有限公司 | A kind of question sentence recommends method and system |
CN107730002A (en) * | 2017-10-13 | 2018-02-23 | 国网湖南省电力公司 | A kind of communication network shutdown remote control parameter intelligent fuzzy comparison method |
CN107957993A (en) * | 2017-12-13 | 2018-04-24 | 北京邮电大学 | The computational methods and device of english sentence similarity |
CN108920464A (en) * | 2018-07-03 | 2018-11-30 | 中国民航大学 | A kind of land sky call based on CNN is rehearsed error classification method |
CN109426664A (en) * | 2017-08-30 | 2019-03-05 | 上海诺悦智能科技有限公司 | A kind of sentence similarity calculation method based on convolutional neural networks |
CN109446528A (en) * | 2018-10-30 | 2019-03-08 | 南京中孚信息技术有限公司 | The recognition methods of new fraudulent gimmick and device |
CN109509475A (en) * | 2018-12-28 | 2019-03-22 | 出门问问信息科技有限公司 | Method, apparatus, electronic equipment and the computer readable storage medium of speech recognition |
CN109522555A (en) * | 2018-11-16 | 2019-03-26 | 中国民航大学 | A kind of land sky call based on BiLSTM is rehearsed semantic automatic Verification method |
CN109766554A (en) * | 2019-01-11 | 2019-05-17 | 中国民航大学 | A kind of land sky call based on interactive mode is rehearsed error classification method |
CN112800233A (en) * | 2021-04-13 | 2021-05-14 | 成都数联铭品科技有限公司 | Text position detection method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120209853A1 (en) * | 2006-01-23 | 2012-08-16 | Clearwell Systems, Inc. | Methods and systems to efficiently find similar and near-duplicate emails and files |
CN102955774A (en) * | 2012-05-30 | 2013-03-06 | 华东师范大学 | Control method and device for calculating Chinese word semantic similarity |
CN104866829A (en) * | 2015-05-25 | 2015-08-26 | 苏州大学 | Cross-age face verify method based on characteristic learning |
CN105335496A (en) * | 2015-10-22 | 2016-02-17 | 国网山东省电力公司电力科学研究院 | Customer service repeated call treatment method based on cosine similarity text mining algorithm |
-
2016
- 2016-07-18 CN CN201610573975.2A patent/CN106227718A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120209853A1 (en) * | 2006-01-23 | 2012-08-16 | Clearwell Systems, Inc. | Methods and systems to efficiently find similar and near-duplicate emails and files |
CN102955774A (en) * | 2012-05-30 | 2013-03-06 | 华东师范大学 | Control method and device for calculating Chinese word semantic similarity |
CN104866829A (en) * | 2015-05-25 | 2015-08-26 | 苏州大学 | Cross-age face verify method based on characteristic learning |
CN105335496A (en) * | 2015-10-22 | 2016-02-17 | 国网山东省电力公司电力科学研究院 | Customer service repeated call treatment method based on cosine similarity text mining algorithm |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109426664A (en) * | 2017-08-30 | 2019-03-05 | 上海诺悦智能科技有限公司 | A kind of sentence similarity calculation method based on convolutional neural networks |
CN107704563A (en) * | 2017-09-29 | 2018-02-16 | 广州多益网络股份有限公司 | A kind of question sentence recommends method and system |
CN107704563B (en) * | 2017-09-29 | 2021-05-18 | 广州多益网络股份有限公司 | Question recommendation method and system |
CN107730002B (en) * | 2017-10-13 | 2020-06-02 | 国网湖南省电力公司 | Intelligent fuzzy comparison method for remote control parameters of communication gateway machine |
CN107730002A (en) * | 2017-10-13 | 2018-02-23 | 国网湖南省电力公司 | A kind of communication network shutdown remote control parameter intelligent fuzzy comparison method |
CN107957993A (en) * | 2017-12-13 | 2018-04-24 | 北京邮电大学 | The computational methods and device of english sentence similarity |
CN107957993B (en) * | 2017-12-13 | 2020-09-25 | 北京邮电大学 | English sentence similarity calculation method and device |
CN108920464A (en) * | 2018-07-03 | 2018-11-30 | 中国民航大学 | A kind of land sky call based on CNN is rehearsed error classification method |
CN108920464B (en) * | 2018-07-03 | 2022-03-01 | 中国民航大学 | Land-air communication repeating error classification method based on CNN |
CN109446528A (en) * | 2018-10-30 | 2019-03-08 | 南京中孚信息技术有限公司 | The recognition methods of new fraudulent gimmick and device |
CN109522555A (en) * | 2018-11-16 | 2019-03-26 | 中国民航大学 | A kind of land sky call based on BiLSTM is rehearsed semantic automatic Verification method |
CN109509475A (en) * | 2018-12-28 | 2019-03-22 | 出门问问信息科技有限公司 | Method, apparatus, electronic equipment and the computer readable storage medium of speech recognition |
CN109509475B (en) * | 2018-12-28 | 2021-11-23 | 出门问问信息科技有限公司 | Voice recognition method and device, electronic equipment and computer readable storage medium |
CN109766554A (en) * | 2019-01-11 | 2019-05-17 | 中国民航大学 | A kind of land sky call based on interactive mode is rehearsed error classification method |
CN112800233A (en) * | 2021-04-13 | 2021-05-14 | 成都数联铭品科技有限公司 | Text position detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106227718A (en) | Land based on CNN sky call semantic consistency method of calibration | |
Lista | Statistical Methods For Data Analysis | |
CN101587155B (en) | Oil soaked transformer fault diagnosis method | |
CN107273490B (en) | Combined wrong question recommendation method based on knowledge graph | |
CN104899298B (en) | A kind of microblog emotional analysis method based on large-scale corpus feature learning | |
CN106156848B (en) | A kind of land sky call semantic consistency method of calibration based on LSTM-RNN | |
CN107766324A (en) | A kind of text coherence analysis method based on deep neural network | |
CN107562784A (en) | Short text classification method based on ResLCNN models | |
CN106980858A (en) | The language text detection of a kind of language text detection with alignment system and the application system and localization method | |
CN106886543A (en) | The knowledge mapping of binding entity description represents learning method and system | |
CN104536881A (en) | Public testing error report priority sorting method based on natural language analysis | |
CN108960330A (en) | Remote sensing images semanteme generation method based on fast area convolutional neural networks | |
CN106599933A (en) | Text emotion classification method based on the joint deep learning model | |
CN106557462A (en) | Name entity recognition method and system | |
CN106446526A (en) | Electronic medical record entity relation extraction method and apparatus | |
CN104155574A (en) | Power distribution network fault classification method based on adaptive neuro-fuzzy inference system | |
CN107590127A (en) | A kind of exam pool knowledge point automatic marking method and system | |
CN105334504B (en) | The radar target identification method of nonlinear discriminant projection model based on big border | |
CN110096711A (en) | The natural language semantic matching method of the concern of the sequence overall situation and local dynamic station concern | |
CN103473556B (en) | Hierarchical SVM sorting technique based on rejection subspace | |
CN105606914A (en) | IWO-ELM-based Aviation power converter fault diagnosis method | |
CN111142538B (en) | Unmanned ship operation strategy real-time decision-making method for water environment monitoring | |
CN106681305A (en) | Online fault diagnosing method for Fast RVM (relevance vector machine) sewage treatment | |
CN114842208A (en) | Power grid harmful bird species target detection method based on deep learning | |
CN107239857A (en) | Overhead transmission line methods of risk assessment based on LS_SVM and PCA |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20161214 |
|
WD01 | Invention patent application deemed withdrawn after publication |