CN109901869A - A kind of computer program classification method based on bag of words - Google Patents
A kind of computer program classification method based on bag of words Download PDFInfo
- Publication number
- CN109901869A CN109901869A CN201910071876.8A CN201910071876A CN109901869A CN 109901869 A CN109901869 A CN 109901869A CN 201910071876 A CN201910071876 A CN 201910071876A CN 109901869 A CN109901869 A CN 109901869A
- Authority
- CN
- China
- Prior art keywords
- api calls
- sequence
- vector
- computer program
- api
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Stored Programmes (AREA)
- Debugging And Monitoring (AREA)
Abstract
Computer program is converted to an API Calls sequence by custom function or basic block for unit by the computer program classification method based on bag of words that the invention discloses a kind of；The API Calls sequence that computer program extracts is the combination of API Calls, there are context dependencies for each API Calls in one API Calls sequence, utilize bag of words of the API Calls sequence training based on context relation, the vector for obtaining each API Calls sequence and API Calls indicates, similarity is calculated by the Euclidean distance of API Calls sequence, determines the classification of unknown computer program.The present invention greatly reduces computation complexity while the good whole semantic feature for learning to arrive code and the information of context computer program code gene using three-layer neural network algorithm training computer program code；The base column vector expression of computer program based on deep learning and clustering method detection have preferable effect.
Description
Technical field
The computer program classification method based on bag of words that the present invention relates to a kind of.
Background technique
With the development of the technologies such as computer program shell adding, encryption, the classification of computer behavior is also more and more difficult.Cause
This, needs a kind of problem that computer program classification method is difficult with settlement procedure classification.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of computer programs based on bag of words point
Class method.
The purpose of the present invention is achieved through the following technical solutions:
Computer program is converted to one by customized letter by a kind of computer program classification method based on bag of words
Several or basic block is the API Calls sequence of unit；
The API Calls sequence that computer program extracts is the combination of API Calls, each API tune in an API Calls sequence
Each is obtained using bag of words of the API Calls sequence training based on context relation with there are context dependencies
The vector of API Calls sequence and API Calls indicates, calculates similarity by the Euclidean distance of API Calls sequence, determines unknown
The classification of computer program.
It is preferred that realizing the foot of write-in IDA Pro software using IDA Python program according to dis-assembling code
This document；Computer program is converted to an API Calls sequence by custom function or basic block for unit by scanner program
Column.
It is preferred that classification method initialization is described as follows:
(1) for indicating that the term vector A length an of API Calls is set as k, all API Calls compositions one in method
The matrix of a M × k dimension, i.e. A ∈ AW×k；
(2) for indicating that a computer API Calls sequence AL vector length is k in method；
(3) the calling API up and down of forward-backward correlation is determined for indicating that sliding window length is set as win_size in method
Quantity；
(4) a computer API Calls vector A of m (m=1 ..., M) is initialized in methodm=(1,0 ..., 0)T∈Ik ×1(k dimensional vector)；
(5) the inner parameter θ of m-th of computer API Calls is initialized in methodm=(1,0 ..., 0)T∈Ik×1(k dimension
Vector)；
(6) n-th (n=1 ..., N) a computer API Calls sequence vector is initialized in method(k dimensional vector), wherein the length of the calling sequence is ln。
It is preferred that classification method model training process description is as follows:
(1) it is directed to the corresponding API Calls sequence of each computer programWherein number n=
1,2 ..., N；
(2) i-th of API Calls of n-th of API Calls sequence are directed toInitialize inner parameter e=0, method
Middle initialization is used for inner parameter θn=(1,0 ..., 0)T∈Ik×1(k dimensional vector).Calculate the contextual information vector of the API CallsSequence indicates
Length is 2winsize+ 1, wherein AL (n) indicates the context relation of n-th of API Calls sequence
(3) it is directed to parameterSuccessively iteration updates calculating parameter WhereinIt indicatesVector turns order, θuIndicate API tune
With the corresponding initiation parameter θ vector of sequence u, sigma function is indicated
L function representation:
Eta-function indicates:
For parameterUpdate Au=Au+ e updates ALu=ALu+ e, wherein AuIndicate u pairs of API Calls
The initialization API Calls vector answered, ALuIndicate the API Calls sequence vector where API Calls u, update terminates, return step
(2), until iteration terminates.
(4) it is updated by step (1)-(3) training and obtains training the corresponding vector AL of each computer program in libraryn, with
And the corresponding vector A of each API Callsm。
It is preferred that method detection process is described as follows:
(1) API Calls sequence AL '=(A ' that computer program to be detected extracts1, A '2..., A 'n′), whereinCalling sequence length n '.Initialize the API Calls sequence vector(k dimensional vector)
(2) a API Calls A ' of i-th (i=1 ..., n ') for the API Calls sequence AL ' that computer program to be detected extractsi
∈AM×k, inner parameter e=0 is initialized, initialization is for inner parameter θ '=(1,0 ..., 0) in methodT.Calculate the API tune
Contextual information vectorSequence indicates
(3) it is directed to parameter u={ A 'i}∪Context(A′i), successively iteration updates calculating parameter WhereinIt indicatesVector turns order, θuIndicate API tune
With the corresponding initiation parameter θ vector of sequence u, sigma function is indicatedFunction representationEta-function indicates
For parameter u=Context (A 'i), update Au=Au+ e updates AL '=AL '+e, wherein AuIndicate API Calls u
Corresponding initialization API Calls vector, AL ' expression API Calls sequence vector to be detected；
(4) the corresponding type label of API Calls sequence to be detectedCalculate with it is to be checked
API Calls sequence vector is surveyed apart from the nearest corresponding label of API Calls sequence.
It is preferred that for i-th of API Calls vector of n-th of API Calls sequenceIn setting
Hereafter relying on window size is 2winsize+ 1, the API CallsContextual information sequence indicateThe mesh of model optimization
Scalar functions are to calculate log-likelihood functionIt is calculated each
API Calls vectorWith each API Calls sequence vector AL (n).
The beneficial effects of the present invention are:
(1) using three-layer neural network algorithm training computer program code, the holophrase of code is arrived in good study
While adopted feature and the information of context computer program code gene, computation complexity is greatly reduced；
(2) the base column vector expression of the computer program based on deep learning and classification method detection have preferable effect
Fruit.
Detailed description of the invention
Fig. 1 is the context relation schematic diagram of this method three-layer neural network bag of words.
Specific embodiment
Technical solution of the present invention is described in further detail with reference to the accompanying drawing, but protection scope of the present invention is not limited to
It is as described below.
Computer program is converted to one by customized letter by a kind of computer program classification method based on bag of words
Several or basic block is the API Calls sequence of unit；
There are certain behaviours for each API Calls, and the API Calls sequence that computer program extracts is API Calls
Combination, i.e. there are context dependencies for each API Calls in an API Calls sequence, utilize API Calls sequence training base
In the bag of words of context relation, the vector for obtaining each API Calls sequence and API Calls is indicated, passes through API Calls
The Euclidean distance of sequence calculates similarity, determines the classification of unknown computer program.
In a preferred embodiment, according to dis-assembling code, realize that write-in IDA Pro is soft using IDA Python program
The script file of part；Scanner program instruction, it is unit that computer program, which is converted to one by custom function or basic block,
API Calls sequence.
In a preferred embodiment, classification method initialization is described as follows:
(1) for indicating that the term vector A length an of API Calls is set as k, all API Calls compositions one in method
The matrix of a M × k dimension, i.e. A ∈ AM×k；
(2) for indicating that a computer API Calls sequence AL vector length is k in method；
(3) the calling API up and down of forward-backward correlation is determined for indicating that sliding window length is set as win_size in method
Quantity；
(4) a computer API Calls vector A of m (m=1 ..., M) is initialized in methodm=(1,0 ..., 0)T∈Ik ×1(k dimensional vector)；
(5) the inner parameter θ of m-th of computer API Calls is initialized in methodm=(1,0 ..., 0)T∈Ik×1(k dimension
Vector)；
(6) n-th (n=1 ..., N) a computer API Calls sequence vector is initialized in method
(k dimensional vector), wherein the length of the calling sequence is ln。
In a preferred embodiment, classification method model training process description is as follows:
(1) it is directed to the corresponding API Calls sequence of each computer programWherein number n=
1,2 ..., N；
(2) i-th of API Calls of n-th of API Calls sequence are directed toInitialize inner parameter e=0, method
Middle initialization is used for inner parameter θn=(1,0 ..., 0)T∈Ik×1(k dimensional vector).Calculate the contextual information vector of the API CallsSequence indicates
Length is 2winsize+ 1, wherein AL (n) indicates the context relation of n-th of API Calls sequence
(3) it is directed to parameterSuccessively iteration updates calculating parameter WhereinIt indicatesVector turns order, θuIndicate API tune
With the corresponding initiation parameter θ vector of sequence u, sigma function is indicated
L function representation:
Eta-function indicates:
For parameterUpdate Au=Au+ e updates ALu=ALu+ e, wherein AuIndicate u pairs of API Calls
The initialization API Calls vector answered, ALuIndicate the API Calls sequence vector where API Calls u, update terminates, return step
(2), until iteration terminates.
(4) being updated by step (1)-(3) training obtains training each computer program (API Calls sequence) in library corresponding
Vector ALnAnd the corresponding vector A of each API Callsm。
In a preferred embodiment, method detection process is described as follows:
(1) API Calls sequence AL '=(A ' that computer program to be detected extracts1, A '2..., A 'n′), whereinCalling sequence length n '.Initialize the API Calls sequence vector(k dimensional vector)
(2) a API Calls A ' of i-th (i=1 ..., n ') for the API Calls sequence AL ' that computer program to be detected extractsi
∈AM×k, inner parameter e=0 is initialized, initialization is for inner parameter θ '=(1,0 ..., 0) in methodT.Calculate the API tune
Contextual information vectorSequence indicates
(3) it is directed to parameter u={ A 'i}∪Context(A′i), successively iteration updates calculating parameter WhereinIt indicatesVector turns order, θuIndicate API tune
With the corresponding initiation parameter θ vector of sequence u, sigma function is indicatedFunction representationEta-function indicates
For parameter u=Context (A 'i), update Au=Au+ e updates AL '=AL '+e, wherein AuIndicate API Calls u
Corresponding initialization API Calls vector, AL ' expression API Calls sequence vector to be detected；
(4) the corresponding type label of API Calls sequence to be detectedCalculate with it is to be checked
API Calls sequence vector is surveyed apart from the nearest corresponding label of API Calls sequence, that is, calculates the distance of API Calls sequence two-by-two,
Determine that the smallest distance, that is, looks for a most similar API Calls sequence, as affiliated class categories.
In a preferred embodiment, a new sample is added in detection part, and sample size+1, i.e. re -training are whole
The vector space of a sample, computation model and process finally obtain the vector of new samples, pass through vector as training process
The class categories of Euclidean distance judgement sample.
In a preferred embodiment, classification method uses the context relation of three-layer neural network bag of words, such as Fig. 1
It is shown, for i-th of API Calls vector of n-th of API Calls sequenceSet Context-dependent window size
For 2winsize+ 1, the API CallsContextual information sequence indicateThe mesh of model optimization
Scalar functions are to calculate log-likelihood functionCalculate the logarithm seemingly
Right function obtains each API Calls vectorWith each API Calls sequence vector AL (n).
In a preferred embodiment, in order to calculate the maximum probability that API Calls occur, likelihood function is calculatedIt is updated and is restrained using gradient rise method iteration, exported
Layer obtains the vector of each API Calls and API Calls sequence.
In a preferred embodiment, as shown in Figure 1, first having to define the dimension size k of the vector of API Calls, and
Context size 2win_size, we are for each of training sample API Calls in this way, and win_size of the front
The API Calls sequence at word and subsequent win_size word and place is as the input of model, and the API Calls are as sample
Output, it is expected that softmax maximum probability.
Overall model is a three-layer neural network model, and as shown in Fig. 1, input layer is n-th of API Calls sequence
I-th of API CallsContext relation API Calls term vector；Middle layer be all term vectors of input layer superposition and；
Output layer has been changed to calculate probability value using Huffman tree from archetype using softmax calculating probability value.
Initialization section, using all program API Calls as leaf node, the frequency conduct of program API Calls appearance
The power of node constructs Huffman tree, as output.From root node, the path for reaching specified leaf node is specified to calculate
The probability of program API Calls.
The effect of Huffman tree construction, passes through parameter θm=(1,0 ..., 0)T∈Ik×1What is embodied is an intermediate knot
Fruit simplifies the complexity that gradient calculates for calling the coding of API.
The basic thought of model of the present invention, for each of each API Calls sequence API Calls, it is by it
The contextual information of API Calls sequence determines where 2win_size upper and lower API Calls and place, i.e., this upper and lower
The probability that the combination of text is formed is most rationally maximum.The present invention is modeled by the way of three-layer neural network, using in gradient
The method of liter iterates to calculate the probability of each API Calls, finally obtains the vector expression of each API Calls.
In order to make symbol meaning of the invention definitely, spy concentrates and makes description below:
(1) unified in method to indicate calling sequence using AL, a computer program extracts to obtain a calling sequence, i.e.,
One computer program corresponds a computer calling sequence, one of to call API or be expressed as A, API fastly substantially
Calling quantity is M, and numbering is m=1,2 ..., M.
(2) the computer program quantity in method for training is N, and numbering is n=1,2 ..., N, n-th of calculating
The corresponding API Calls sequence of machine programWhereinThe wherein length of the calling sequence
For ln。
(3) computer program in method for training is divided into S kind, initializes the corresponding type of each API Calls sequence
LabelWherein LnIndicate the computer program grouping where n-th of computer program, i.e. Ln=1,
2 ..., S }.
The present invention is directed to the API Calls sequence of computer program, fully considers the single API tune of computer using bag of words
The contextual information of semantic feature and API Calls execution sequence, finds the API Calls of potentially possible computer program
Composite sequence relationship effectively improves the classification accuracy of computer program.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, it is noted that all
Made any modifications, equivalent replacements, and improvements etc. within the spirit and principles in the present invention should be included in guarantor of the invention
Within the scope of shield.
Claims (6)
1. a kind of computer program classification method based on bag of words, it is characterised in that: computer program is converted to one
It is the API Calls sequence of unit by custom function or basic block；
The API Calls sequence that computer program extracts is the combination of API Calls, and each API Calls are deposited in an API Calls sequence
Each API tune is obtained using bag of words of the API Calls sequence training based on context relation in context dependency
It is indicated with the vector of sequence and API Calls, similarity is calculated by the Euclidean distance of API Calls sequence, determines unknown calculating
The classification of machine program.
2. a kind of computer program classification method based on bag of words according to claim 1, it is characterised in that: foundation
Dis-assembling code realizes the script file of write-in IDA Pro software using IDA Python program；Scanner program, by computer
Program transformation is one by API Calls sequence that custom function or basic block are unit.
3. a kind of computer program classification method based on bag of words according to claim 1, which is characterized in that classification
Method initialization is described as follows:
(1) for indicating that the term vector A length of one API Calls is set as k in method, all one M of API Calls composition ×
The matrix of k dimension, i.e. A ∈ AM×k；
(2) for indicating that a computer API Calls sequence AL vector length is k in method；
(3) the API number of calling up and down of forward-backward correlation is determined for indicating that sliding window length is set as win_size in method
Amount；
(4) a computer API Calls vector A of m (m=1 ..., M) is initialized in methodm=(1,0 ..., 0)T∈Ik×1(k dimension
Vector)；
(5) the inner parameter θ of m-th of computer API Calls is initialized in methodm=(1,0 ..., 0)T∈Ik×1(k dimensional vector)；
(6) n-th (n=1 ..., N) a computer API Calls sequence vector is initialized in method
(k dimensional vector), wherein the length of the calling sequence is ln。
4. a kind of computer program classification method based on bag of words according to claim 1 or 3, which is characterized in that
Classification method model training process description is as follows:
(1) it is directed to the corresponding API Calls sequence of each computer programWherein number n=1,
2 ..., N；
(2) i-th of API Calls of n-th of API Calls sequence are directed toInner parameter e=0 is initialized, in method just
Beginningization is used for inner parameter θn=(1,0 ..., 0)T∈Ik×1(k dimensional vector) calculates the contextual information vector of the API CallsSequence indicates
Length is 2winsize+ 1, wherein AL (n) indicates the context relation of n-th of API Calls sequence
(3) it is directed to parameterSuccessively iteration updates calculating parameter E=e+g θu,WhereinIt indicatesVector turns order, θuIndicate API Calls
The corresponding initiation parameter θ vector of sequence u, sigma function indicate
L function representation:
Eta-function indicates:
For parameterUpdate Au=Au+ e updates ALu=ALu+ e, wherein AuIndicate that API Calls u is corresponding
Initialize API Calls vector, ALuIndicating the API Calls sequence vector where API Calls u, update terminates, return step (2),
Until iteration terminates；
(4) it is updated by step (1)-(3) training and obtains training the corresponding vector AL of each computer program in libraryn, and it is each
The corresponding vector A of API Callsm。
5. a kind of computer program classification method based on bag of words according to claim 1, which is characterized in that method
Detection process is described as follows:
(1) API Calls sequence AL '=(A ' that computer program to be detected extracts1, A '2..., A 'n′), wherein
Calling sequence length n ' initializes the API Calls sequence vector
(2) a API Calls A ' i ∈ A of i-th (i=1 ..., n ') for the API Calls sequence AL ' that computer program to be detected extractsM ×k, inner parameter e=0 is initialized, initialization is for inner parameter θ '=(1,0 ..., 0) in methodT, calculate the API Calls
Contextual information vectorSequence indicates
(3) it is directed to parameter u={ A 'i}∪Context(A′i), successively iteration updates calculating parameter E=e+g θu,WhereinIt indicatesVector turns order, θuIndicate API Calls sequence
The corresponding initiation parameter θ vector of u is arranged, sigma function indicatesFunction representationEta-function indicates
For parameter u=Context (A 'i), update Au=Au+ e updates AL '=AL '+e, and wherein Au indicates that API Calls u is corresponding
Initialization API Calls vector, AL ' expression API Calls sequence vector to be detected；
(4) the corresponding type label of API Calls sequence to be detectedCalculate with it is to be detected
API Calls sequence vector is apart from the nearest corresponding label of API Calls sequence.
6. a kind of computer program classification method based on bag of words according to claim 1, it is characterised in that: be directed to
I-th of API Calls vector of n-th of API Calls sequenceContext-dependent window size is set as 2winsize
+ 1, the API CallsContextual information sequence indicateThe mesh of model optimization
Scalar functions are to calculate log-likelihood functionIt is calculated each
API Calls vectorWith each API Calls sequence vector AL (n).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910071876.8A CN109901869B (en) | 2019-01-25 | 2019-01-25 | Bag-of-words model-based computer program classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910071876.8A CN109901869B (en) | 2019-01-25 | 2019-01-25 | Bag-of-words model-based computer program classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109901869A true CN109901869A (en) | 2019-06-18 |
| CN109901869B CN109901869B (en) | 2022-03-18 |
Family
ID=66944183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910071876.8A Active CN109901869B (en) | 2019-01-25 | 2019-01-25 | Bag-of-words model-based computer program classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109901869B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111210326A (en) * | 2019-12-27 | 2020-05-29 | 大象慧云信息技术有限公司 | Method and system for constructing user portrait |
| CN116894209A (en) * | 2023-09-05 | 2023-10-17 | 北京开源芯片研究院 | Sampling point classification method, device, electronic equipment and readable storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130259311A1 (en) * | 2010-09-02 | 2013-10-03 | Edge 3 Technologies, Inc | Method and Apparatus for Spawning Specialist Belief Propagation Networks |
| CN105868108A (en) * | 2016-03-28 | 2016-08-17 | 中国科学院信息工程研究所 | Instruction-set-irrelevant binary code similarity detection method based on neural network |
| CN106997367A (en) * | 2016-01-26 | 2017-08-01 | 华为技术有限公司 | Sorting technique, sorter and the categorizing system of program file |
| CN107885999A (en) * | 2017-11-08 | 2018-04-06 | 华中科技大学 | A kind of leak detection method and system based on deep learning |
| CN108647518A (en) * | 2018-03-16 | 2018-10-12 | 广东工业大学 | A kind of Android platform malware detection method based on deep learning |
| CN108763360A (en) * | 2018-05-16 | 2018-11-06 | 北京旋极信息技术股份有限公司 | A kind of sorting technique and device, computer readable storage medium |
| CN108804469A (en) * | 2017-05-04 | 2018-11-13 | 腾讯科技（深圳）有限公司 | A kind of web page identification method and electronic equipment |
| CN108959924A (en) * | 2018-06-12 | 2018-12-07 | 浙江工业大学 | A kind of Android malicious code detecting method of word-based vector sum deep neural network |
-
2019
- 2019-01-25 CN CN201910071876.8A patent/CN109901869B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130259311A1 (en) * | 2010-09-02 | 2013-10-03 | Edge 3 Technologies, Inc | Method and Apparatus for Spawning Specialist Belief Propagation Networks |
| CN106997367A (en) * | 2016-01-26 | 2017-08-01 | 华为技术有限公司 | Sorting technique, sorter and the categorizing system of program file |
| CN105868108A (en) * | 2016-03-28 | 2016-08-17 | 中国科学院信息工程研究所 | Instruction-set-irrelevant binary code similarity detection method based on neural network |
| CN108804469A (en) * | 2017-05-04 | 2018-11-13 | 腾讯科技（深圳）有限公司 | A kind of web page identification method and electronic equipment |
| CN107885999A (en) * | 2017-11-08 | 2018-04-06 | 华中科技大学 | A kind of leak detection method and system based on deep learning |
| CN108647518A (en) * | 2018-03-16 | 2018-10-12 | 广东工业大学 | A kind of Android platform malware detection method based on deep learning |
| CN108763360A (en) * | 2018-05-16 | 2018-11-06 | 北京旋极信息技术股份有限公司 | A kind of sorting technique and device, computer readable storage medium |
| CN108959924A (en) * | 2018-06-12 | 2018-12-07 | 浙江工业大学 | A kind of Android malicious code detecting method of word-based vector sum deep neural network |
Non-Patent Citations (2)
| Title |
|---|
| TRUNG KIEN TRAN 等: ""NLP-based approaches for malware classification from API sequences"", 《2017 21ST ASIA PACIFIC SYMPOSIUM ON INTELLIGENT AND EVOLUTIONARY SYSTEMS (IES)》 * |
| 杨晓敏 等: ""一种基于词袋模型的图像分类方法"", 《太赫兹科学与电子信息学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111210326A (en) * | 2019-12-27 | 2020-05-29 | 大象慧云信息技术有限公司 | Method and system for constructing user portrait |
| CN116894209A (en) * | 2023-09-05 | 2023-10-17 | 北京开源芯片研究院 | Sampling point classification method, device, electronic equipment and readable storage medium |
| CN116894209B (en) * | 2023-09-05 | 2023-12-22 | 北京开源芯片研究院 | Sampling point classification method, device, electronic equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109901869B (en) | 2022-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108228686B (en) | Method and device for realizing image-text matching and electronic equipment | |
| CN109460553B (en) | Machine reading understanding method based on threshold convolutional neural network | |
| CN106126507B (en) | A kind of depth nerve interpretation method and system based on character code | |
| CN109241283A (en) | A kind of file classification method based on multi-angle capsule network | |
| CN109657226B (en) | Multi-linkage attention reading understanding model, system and method | |
| CN106383816B (en) | The recognition methods of Chinese minority area place name based on deep learning | |
| CN108647207A (en) | Natural language modification method, system, equipment and storage medium | |
| JP2020520492A (en) | Document abstract automatic extraction method, device, computer device and storage medium | |
| CN110263227A (en) | Clique based on figure neural network finds method and system | |
| CN106997474A (en) | A kind of node of graph multi-tag sorting technique based on deep learning | |
| CN110033089B (en) | Method and system for optimizing parameters of handwritten digital image recognition deep neural network based on distributed estimation algorithm | |
| US20200134471A1 (en) | Method for Generating Neural Network and Electronic Device | |
| CN110147444B (en) | Text prediction method and device based on neural network language model and storage medium | |
| CN113220876B (en) | Multi-label classification method and system for English text | |
| CN113688634B (en) | Fine-grained emotion analysis method | |
| CN110968299A (en) | Front-end engineering code generation method based on hand-drawn webpage image | |
| CN109300128A (en) | The transfer learning image processing method of structure is implied based on convolutional Neural net | |
| CN109901869A (en) | A kind of computer program classification method based on bag of words | |
| CN111339308B (en) | Training method and device of basic classification model and electronic equipment | |
| CN113128478A (en) | Model training method, pedestrian analysis method, device, equipment and storage medium | |
| CN109710921A (en) | Calculation method, device, computer equipment and the storage medium of Words similarity | |
| CN111400713B (en) | Malicious software population classification method based on operation code adjacency graph characteristics | |
| CN113779225A (en) | Entity link model training method, entity link method and device | |
| CN115114409A (en) | Civil aviation unsafe event combined extraction method based on soft parameter sharing | |
| CN111651985A (en) | Method and device for Chinese word segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |