CN112651412A - Multi-label classification method and device based on deep learning and storage medium - Google Patents

Multi-label classification method and device based on deep learning and storage medium Download PDF

Info

Publication number
CN112651412A
CN112651412A CN201910956811.1A CN201910956811A CN112651412A CN 112651412 A CN112651412 A CN 112651412A CN 201910956811 A CN201910956811 A CN 201910956811A CN 112651412 A CN112651412 A CN 112651412A
Authority
CN
China
Prior art keywords
classification
prediction result
deep learning
result vector
label
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
Application number
CN201910956811.1A
Other languages
Chinese (zh)
Inventor
蔡鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Telecom Corp Ltd
Original Assignee
China Telecom Corp Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Telecom Corp Ltd filed Critical China Telecom Corp Ltd
Priority to CN201910956811.1A priority Critical patent/CN112651412A/en
Publication of CN112651412A publication Critical patent/CN112651412A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The disclosure provides a deep learning-based multi-label classification method, a deep learning-based multi-label classification device and a storage medium, wherein the method comprises the following steps: obtaining a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable, converting the classification variables into a plurality of classification integer values, and generating a plurality of classification identifier values according to the classification integer values; and predicting the detection sample by using the deep learning network model to obtain a first prediction result vector of the detection sample. The method, the device and the storage medium can be applied to multi-label multi-classification prediction scenes, can learn the correlation among labels, can utilize a deep learning platform to carry out model whole-course training, improve the prediction efficiency and accuracy and can reduce the calculation amount.

Description

Multi-label classification method and device based on deep learning and storage medium
Technical Field
The invention relates to the technical field of deep learning, in particular to a deep learning-based multi-label classification method, a deep learning-based multi-label classification device and a storage medium.
Background
When multi-label multi-classification prediction is carried out, classifier algorithms of some deep learning network models can only output a first-order tensor, cannot meet the output requirement of a second-order tensor of multi-label multi-classification, and cannot directly use classifier algorithm packages (softmax and the like) of a deep learning platform. At present, each label can be classified and modeled independently, if L labels exist, L independent deep learning network models are needed, the calculated amount is greatly increased, each deep learning network model cannot learn possible correlation among label indexes, and the prediction accuracy is reduced.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a method, an apparatus and a storage medium for deep learning based multi-label classification.
According to one aspect of the present disclosure, there is provided a deep learning-based multi-label classification method, including: acquiring a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable; the classification variable is a non-integer numerical value, and the classification determination identifier is used for representing whether the classification variable is selected or not; converting the plurality of classification variables into a plurality of classification integer values, and generating a plurality of classification identification values according to the plurality of classification integer values; wherein each of said class identifier values is used to characterize a determined classification variable of said element label; predicting a detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample; the number of elements of the first prediction result vector is the same as the number of the element labels, and each element of the first prediction result vector is a classification identification value corresponding to one element label.
Optionally, the predicting the detection sample by using the deep learning network model includes: constructing an initial deep learning network model; obtaining training samples based on historical data and obtaining the second prediction result vector corresponding to the training samples; the number of elements of the second prediction result vector is the same as that of the first prediction result vector, and each element of the second prediction result vector is a classification identification numerical value corresponding to one element label; generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set; and the prediction result acquisition module is used for predicting the detection sample by using the trained deep learning network model.
Optionally, the deep learning network model includes: the device comprises an input layer, a hidden layer, a full connection layer and an output layer; and performing regression prediction through the full connection layer to obtain the first prediction result vector and outputting the first prediction result vector through the output layer.
Optionally, the converting the plurality of classification variables into a plurality of classification integer values comprises: obtaining a first rank corresponding to the plurality of categorical variables; converting the plurality of categorical variables into a plurality of categorical integer values based on the first ordering; wherein a second ordering corresponding to the plurality of sorted integer values is the same as the first ordering.
Optionally, the generating a plurality of classification identification values according to the plurality of classification integer values includes: respectively carrying out normalization processing on the plurality of classification integer values to generate a plurality of classification identification values; wherein each of the class identifier values is a decimal between 0 and 1.
Optionally, each element of the first prediction result vector is subjected to inverse normalization processing to obtain the classification integer value corresponding to each element; converting the categorical integer value into the determined categorical variable.
According to another aspect of the present disclosure, there is provided a multi-label classification apparatus based on deep learning, including: a classification variable obtaining module, configured to obtain a plurality of classification variables corresponding to each element tag and a classification determination identifier of each classification variable; the classification variable is a non-integer numerical value, and the classification determination identifier is used for representing whether the classification variable is selected or not; the identification data setting module is used for converting the classification variables into a plurality of classification integer values and generating a plurality of classification identification numerical values according to the classification integer values; wherein each of said class identifier values is used to characterize a determined classification variable of said element label; the prediction result obtaining module is used for predicting a detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample; the number of elements of the first prediction result vector is the same as the number of the element labels, and each element of the first prediction result vector is a classification identification value corresponding to one element label.
Optionally, the prediction model training module is configured to construct an initial deep learning network model; obtaining training samples based on historical data and obtaining the second prediction result vector corresponding to the training samples; the number of elements of the second prediction result vector is the same as that of the first prediction result vector, and each element of the second prediction result vector is a classification identification numerical value corresponding to one element label; generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set; and the prediction result acquisition module is used for predicting the detection sample by using the trained deep learning network model.
Optionally, the deep learning network model includes: the device comprises an input layer, a hidden layer, a full connection layer and an output layer; and performing regression prediction through the full connection layer to obtain the first prediction result vector and outputting the first prediction result vector through the output layer.
Optionally, the identification data setting module is configured to obtain a first rank corresponding to the plurality of classification variables; converting the plurality of categorical variables into a plurality of categorical integer values based on the first ordering; wherein a second ordering corresponding to the plurality of sorted integer values is the same as the first ordering.
Optionally, the identifier data setting module is configured to perform normalization processing on the plurality of classification integer values, respectively, to generate a plurality of classification identifier values; wherein each of the class identifier values is a decimal between 0 and 1.
Optionally, the prediction result obtaining module is configured to perform inverse normalization processing on each element of the first prediction result vector, so as to obtain the classification integer value corresponding to each element; converting the categorical integer value into the determined categorical variable.
According to another aspect of the present disclosure, there is provided a multi-label classification apparatus based on deep learning, including: a memory; and a processor coupled to the memory, the processor configured to perform the method as described above based on instructions stored in the memory.
According to yet another aspect of the present disclosure, a computer-readable storage medium is provided, which stores computer instructions for execution by a processor to perform the method as described above.
The disclosed deep learning-based multi-label classification method, device and storage medium acquire a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable, convert the plurality of classification variables into a plurality of classification integer values, and generate a plurality of classification identifier values according to the plurality of classification integer values; predicting the detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample; converting the output form of the deep learning network model into a first-order tensor output form, and outputting an L-dimensional vector by using a regression method; the method can be applied to multi-label multi-classification prediction scenes, can learn the correlation among labels, can utilize a deep learning platform to carry out the whole-course training of the model, improves the prediction efficiency and accuracy, and can reduce the calculated amount.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without inventive exercise.
FIG. 1 is a schematic flow chart diagram illustrating one embodiment of a deep learning based multi-label classification method according to the present disclosure;
FIG. 2 is a schematic diagram of a deep learning network model training and prediction process in one embodiment of a deep learning based multi-label classification method according to the present disclosure;
FIG. 3A is a diagram of a conventional deep learning network model; FIG. 3B is a schematic diagram of a deep learning network model of the present disclosure;
FIG. 4 is a schematic diagram of a transition of categorical variables;
FIG. 5 is a block diagram of one embodiment of a deep learning based multi-label classification apparatus according to the present disclosure;
fig. 6 is a block diagram of another embodiment of a deep learning based multi-label classification apparatus according to the present disclosure.
Detailed Description
The present disclosure now will be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the disclosure are shown. The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are only a part of the embodiments of the present disclosure, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
The terms "first", "second", and the like are used hereinafter only for descriptive distinction and not for other specific meanings.
Fig. 1 is a schematic flowchart of an embodiment of a deep learning based multi-label classification method according to the present disclosure, as shown in fig. 1:
step 101, obtaining a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable. The classification variable is a non-integer value, and the classification determination identifier is used for representing whether the classification variable is selected or not.
Step 102, converting the plurality of classification variables into a plurality of classification integer values, and generating a plurality of classification identification values according to the plurality of classification integer values.
In one embodiment, each class identification value is used to characterize a determined classification variable of the element label. The conversion of the plurality of categorical variables into the plurality of categorical integer values may employ a variety of methods, such as obtaining a first ordering corresponding to the plurality of categorical variables, converting the plurality of categorical variables into the plurality of categorical integer values based on the first ordering, and a second ordering corresponding to the plurality of categorical integer values being the same as the first ordering.
Various methods can be used to generate the plurality of classifier identifier values from the plurality of classifier integer values, for example, normalization processing is performed on the plurality of classifier integer values to generate the plurality of classifier identifier values, each of which is a decimal between 0 and 1.
And 103, predicting the detection sample by using the deep learning network model to obtain a first prediction result vector of the detection sample.
The deep learning network model can be an existing convolutional neural network model and the like. The number of elements of the first predictor vector is the same as the number of element labels, and each element of the first predictor vector is a classification identification value corresponding to one element label.
The multi-label classification method based on deep learning in the embodiment can be applied to a multi-label multi-classification prediction scene, and solves the problems that the multi-label multi-classification output is not supported by part of deep learning platform classifier algorithm packages, data prediction analysis modeling is difficult to perform, and especially, the deep learning platform is difficult to perform model whole-course training.
Fig. 2 is a schematic diagram of a training and prediction process of a deep learning network model in an embodiment of a deep learning-based multi-label classification method according to the present disclosure, as shown in fig. 2:
step 201, an initial deep learning network model is constructed.
As shown in fig. 3A, a deep learning network model in the prior art performs separate classification modeling on each label, and L models need to be established for L kinds of labels. As shown in fig. 3B, the deep learning network model of the present disclosure includes: an input layer, a hidden layer, a full connection layer, an output layer and the like; only one deep learning network model is needed to be established, regression prediction is carried out through a full connection layer (Dense layer), and a first prediction result vector is obtained and output through an output layer.
Step 202, training samples are obtained based on the historical data, and a second prediction result vector corresponding to the training samples is obtained. The second predictor vector has the same number of elements as the first predictor vector, and each element of the second predictor vector is a class identifier value corresponding to one element label.
And step 203, generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set.
And step 204, predicting the detection sample by using the trained deep learning network model.
And inputting the detection sample into the trained deep learning network model to obtain a first prediction result vector of the detection sample. And performing inverse normalization processing on each element of the first prediction result vector to obtain a classification integer value corresponding to each element, and converting the classification integer value into a determined classification variable.
In one embodiment, as shown in FIG. 4, the IP address detection samples are predicted using a deep learning network model. Table A1 is the original label format table for the prediction results, and the IP address detection samples are listed in Table A1 as "target IP". The predicted result has three element labels, see the "element label column" in table a1, which are "attack path", "attack complexity", and "threat impact", respectively. Each element label is provided with a plurality of classification variables, each classification variable is provided with a classification determination identifier, the classification variables are non-integer numerical values, and the classification determination identifiers are used for representing whether the classification variables are selected or not.
For example, the categorical variable is listed as "categorical (non-integer)" in table a1, the element label "attack path" has two categorical variables, the set of categorical variables is { local, remote }; the element label 'attack complexity' has three classification variables, and the classification variable set is { low, medium and high }; the element label "threat impact" has four categorical variables, the set of categorical variables being { none, low, medium, high }. The classification determination identifies a classification label column in the table a, the classification determination identifier is 0 or 1, if the classification determination identifier is 0, it is determined that the corresponding classification variable is not selected, and if the classification determination identifier is 1, it is determined that the corresponding classification variable is selected.
The plurality of categorical variables is converted into a plurality of categorical integer values. For example, data type conversion is performed on non-integer ordered classification variables to convert the classification variables into integer classification values. The second ordering corresponding to the plurality of categorical integer values is the same as the first ordering of the plurality of categorical variables, and the ordering before and after conversion does not change. As shown in the column of "sort (integer)" in Table A2 in FIG. 4, the sort integer values of { local, remote } are {0, 1}, the sort integer values of { low, medium, high } are {0, 1, 2}, and the sort integer values of { none, low, medium, high } are {0, 1, 2, 3 }.
Carrying out data standardization on classified integer values of three element labels, wherein the standardization method comprises the following steps: dispersion (minimum maximum) normalization, Z-score normalization, normalization processing, etc. For example, a plurality of classification integer values are normalized to generate a plurality of classification identification values, each classification identification value being a decimal between 0 and 1, each classification identification value being used to characterize a determined classification variable of the element label.
The classification identification value set generated according to the classification variable set { local, remote } is {0.000,1.000}, the classification identification value set generated according to the classification variable set { low, medium, high } is {0.000, 0.500, 1.000}, and the classification identification value set generated according to the classification variable set { none, low, medium, high } is {0.000, 0.333, 0.666, 1.000 }. And converting the original second-order tensor output form (, L, C) in the multi-label multi-classification problem into a first-order tensor output form (, L).
And (3) modifying the full-connection layer activation function of the deep learning network model, outputting an L-dimensional vector continuous value by a regression method, and performing regression model training and prediction, wherein the activation function comprises Linear, Relu, Elu and the like. As shown in table a3 in fig. 4, the IP address detection samples are predicted by using the deep learning network model, and a first prediction result vector {1.000,0.000,0.666} of the IP address detection samples is obtained.
The number of elements of the first predictor vector is the same as the number of element labels, i.e. the number of elements of the first predictor vector is three. Each element of the first prediction result vector is a class identifier value corresponding to one element label, i.e., "1.000" is a class identifier value corresponding to "attack path", "0.000" is a class identifier value corresponding to "attack complexity", and "0.666" is a class identifier value corresponding to "threat impact".
And converting the output format of the deep learning network model into a first prediction result vector. And performing inverse normalization on elements (decimal between 0 and 1) in the first prediction result vector, rounding up, performing reverse data type conversion, and restoring a final result of multi-label multi-classification.
For example, each element of the first prediction result vector {1.000,0.000,0.666} is subjected to inverse normalization processing to obtain a classification integer value corresponding to each element, and the classification integer value is converted into a determined classification variable. Performing inverse normalization processing on the '1.000' to obtain a corresponding classification integer value of 1, and converting the classification integer value '1' into a classification variable 'remote', namely, the classification variable selected by the element label 'attack path' is 'remote', as shown in a table A2; performing inverse normalization processing on the '0.000' to obtain a corresponding classification integer value of 0, and converting the classification integer value '0' into a classification variable 'low', namely, the classification variable selected by the element label 'attack complexity' is 'low'; and performing inverse normalization processing on the '0.666' to obtain a corresponding classification integer value of 2, and converting the classification integer value '2' into a classification variable 'middle', namely, the classification variable selected by the element label 'threat influence' is 'middle'.
Acquiring an IP address training sample based on historical data, and acquiring a second prediction result vector corresponding to the IP address training sample in a manual labeling mode; the second predictor vector has the same number of elements as the first predictor vector, and each element of the second predictor vector is a class identifier value corresponding to one element label. And generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set.
The deep learning-based multi-label classification method in the above embodiment converts the second-order tensor Output form (, L, C) (where L is the number of label types and C is the maximum value of the classification number of each label) of the deep learning network model into the first-order tensor Output form (, L) by using data type conversion, data normalization, data format conversion, and the like, and outputs the L-dimensional vector continuous value (the first prediction result vector) of the Output layer (the Output layer) by using a regression method by modifying the Dense layer (the fully-connected layer) activation function.
In one embodiment, as shown in fig. 5, the present disclosure provides a deep learning based multi-label classification apparatus 50, comprising: a classification variable obtaining module 51, an identification data setting module 52, a prediction result obtaining module 53 and a prediction model training module 54. The classification variable obtaining module 51 obtains a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable; the classification variable is a non-integer numerical value, and the classification determination identifier is used for representing whether the classification variable is selected or not.
The identification data setting module 52 converts the plurality of classification variables into a plurality of classification integer values, and generates a plurality of classification identification values according to the plurality of classification integer values; wherein each class identification value is used to characterize a determined classification variable of the element label. The prediction result obtaining module 53 predicts the detection sample by using the deep learning network model to obtain a first prediction result vector of the detection sample; the number of elements of the first prediction result vector is the same as the number of element labels, and each element of the first prediction result vector is a classification identification value corresponding to one element label.
The predictive model training module 54 builds an initial deep learning network model. The predictive model training module 54 obtains training samples based on the historical data and obtains second predictive result vectors corresponding to the training samples; the number of elements of the second prediction result vector is the same as that of the first prediction result vector, and each element of the second prediction result vector is a classification identification value corresponding to one element label. The predictive model training module 54 generates a training sample set based on the training samples and the second prediction result vector, and trains the initial deep learning network model using the training sample set. The prediction result obtaining module 53 predicts the detection sample by using the trained deep learning network model.
The deep learning network model comprises: the device comprises an input layer, a hidden layer, a full connection layer and an output layer; and performing regression prediction through the full connection layer to obtain a first prediction result vector and outputting the first prediction result vector through the output layer. The identification data setting module 52 obtains a first ordering corresponding to the plurality of categorical variables, and the identification data setting module 52 converts the plurality of categorical variables into a plurality of categorical integer values based on the first ordering; wherein a second ordering corresponding to the plurality of sorted integer values is the same as the first ordering.
The identification data setting module 53 performs normalization processing on the plurality of classification integer values, respectively, to generate a plurality of classification identification values; wherein each class identifier value is a decimal between 0 and 1. The prediction result obtaining module 53 performs inverse normalization processing on each element of the first prediction result vector, obtains a classification integer value corresponding to each element, and converts the classification integer value into a determined classification variable.
Fig. 6 is a block diagram of another embodiment of a deep learning based multi-label classification apparatus according to the present disclosure. As shown in fig. 6, the apparatus may include a memory 61, a processor 62, a communication interface 63, and a bus 64. The memory 61 is used for storing instructions, the processor 62 is coupled to the memory 61, and the processor 62 is configured to execute the multi-label classification method based on deep learning based on the instructions stored in the memory 61.
The memory 61 may be a high-speed RAM memory, a non-volatile memory (non-volatile memory), or the like, and the memory 61 may be a memory array. The storage 61 may also be partitioned and the blocks may be combined into virtual volumes according to certain rules. The processor 62 may be a central processing unit CPU, or an application Specific Integrated circuit asic, or one or more Integrated circuits configured to implement the deep learning based multi-tag classification method of the present disclosure.
According to yet another aspect of the present disclosure, a computer-readable storage medium is provided, having stored thereon computer instructions for execution by a processor to perform the method as above.
In the multi-label classification method, apparatus, and storage medium based on deep learning provided in the above embodiments, a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable are obtained, the plurality of classification variables are converted into a plurality of classification integer values, and a plurality of classification identifier values are generated according to the plurality of classification integer values; predicting the detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample; converting a second-order tensor output form of the deep learning network model into a first-order tensor output form, and outputting an L-dimensional vector by using a regression method through modifying an activation function of a full connection layer; the method can be applied to multi-label multi-classification prediction scenes, can learn the correlation among labels, can utilize a deep learning platform to carry out the whole-course training of the model, improves the prediction efficiency and accuracy, and can reduce the calculated amount.
The method and system of the present disclosure may be implemented in a number of ways. For example, the methods and systems of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.
The description of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the disclosure in the form disclosed. Many modifications and variations will be apparent to practitioners skilled in this art. The embodiment was chosen and described in order to best explain the principles of the disclosure and the practical application, and to enable others of ordinary skill in the art to understand the disclosure for various embodiments with various modifications as are suited to the particular use contemplated.

Claims (14)

1. A multi-label classification method based on deep learning comprises the following steps:
acquiring a plurality of classification variables corresponding to each element label and a classification determination identifier of each classification variable; the classification variable is a non-integer numerical value, and the classification determination identifier is used for representing whether the classification variable is selected or not;
converting the plurality of classification variables into a plurality of classification integer values, and generating a plurality of classification identification values according to the plurality of classification integer values; wherein each of said class identifier values is used to characterize a determined classification variable of said element label;
predicting a detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample;
the number of elements of the first prediction result vector is the same as the number of the element labels, and each element of the first prediction result vector is a classification identification value corresponding to one element label.
2. The method of claim 1, the predicting the detection samples using the deep learning network model comprising:
constructing an initial deep learning network model;
obtaining training samples based on historical data and obtaining the second prediction result vector corresponding to the training samples; the number of elements of the second prediction result vector is the same as that of the first prediction result vector, and each element of the second prediction result vector is a classification identification numerical value corresponding to one element label;
generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set;
and predicting the detection sample by using the trained deep learning network model.
3. The method of claim 1, wherein,
the deep learning network model comprises: the device comprises an input layer, a hidden layer, a full connection layer and an output layer; and performing regression prediction through the full connection layer to obtain the first prediction result vector and outputting the first prediction result vector through the output layer.
4. The method of claim 1, the converting the plurality of categorical variables into a plurality of categorical integer values comprising:
obtaining a first rank corresponding to the plurality of categorical variables;
converting the plurality of categorical variables into a plurality of categorical integer values based on the first ordering;
wherein a second ordering corresponding to the plurality of sorted integer values is the same as the first ordering.
5. The method of claim 4, the generating a plurality of classification identification values from the plurality of classification integer values comprising:
respectively carrying out normalization processing on the plurality of classification integer values to generate a plurality of classification identification values; wherein each of the class identifier values is a decimal between 0 and 1.
6. The method of claim 5, further comprising:
performing inverse normalization processing on each element of the first prediction result vector to obtain the classification integer value corresponding to each element;
converting the categorical integer value into the determined categorical variable.
7. A multi-label classification apparatus based on deep learning, comprising:
a classification variable obtaining module, configured to obtain a plurality of classification variables corresponding to each element tag and a classification determination identifier of each classification variable; the classification variable is a non-integer numerical value, and the classification determination identifier is used for representing whether the classification variable is selected or not;
the identification data setting module is used for converting the classification variables into a plurality of classification integer values and generating a plurality of classification identification numerical values according to the classification integer values; wherein each of said class identifier values is used to characterize a determined classification variable of said element label;
the prediction result obtaining module is used for predicting a detection sample by using a deep learning network model to obtain a first prediction result vector of the detection sample;
the number of elements of the first prediction result vector is the same as the number of the element labels, and each element of the first prediction result vector is a classification identification value corresponding to one element label.
8. The apparatus of claim 7, further comprising:
the prediction model training module is used for constructing an initial deep learning network model; obtaining training samples based on historical data and obtaining the second prediction result vector corresponding to the training samples; the number of elements of the second prediction result vector is the same as that of the first prediction result vector, and each element of the second prediction result vector is a classification identification numerical value corresponding to one element label; generating a training sample set based on the training samples and the second prediction result vector, and training the initial deep learning network model by using the training sample set;
and the prediction result acquisition module is used for predicting the detection sample by using the trained deep learning network model.
9. The apparatus of claim 7, wherein,
the deep learning network model comprises: the device comprises an input layer, a hidden layer, a full connection layer and an output layer; and performing regression prediction through the full connection layer to obtain the first prediction result vector and outputting the first prediction result vector through the output layer.
10. The apparatus of claim 7, wherein,
the identification data setting module is used for acquiring a first sequence corresponding to the plurality of classification variables; converting the plurality of categorical variables into a plurality of categorical integer values based on the first ordering; wherein a second ordering corresponding to the plurality of sorted integer values is the same as the first ordering.
11. The apparatus of claim 10, wherein,
the identification data setting module is configured to perform normalization processing on the plurality of classification integer values respectively to generate a plurality of classification identification values; wherein each of the class identifier values is a decimal between 0 and 1.
12. The apparatus of claim 11, wherein,
the prediction result obtaining module is configured to perform inverse normalization processing on each element of the first prediction result vector to obtain the classification integer value corresponding to each element; converting the categorical integer value into the determined categorical variable.
13. A multi-label classification apparatus based on deep learning, comprising:
a memory; and a processor coupled to the memory, the processor configured to perform the method of any of claims 1-6 based on instructions stored in the memory.
14. A computer-readable storage medium having stored thereon computer instructions for execution by a processor of the method of any one of claims 1 to 6.
CN201910956811.1A 2019-10-10 2019-10-10 Multi-label classification method and device based on deep learning and storage medium Pending CN112651412A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910956811.1A CN112651412A (en) 2019-10-10 2019-10-10 Multi-label classification method and device based on deep learning and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910956811.1A CN112651412A (en) 2019-10-10 2019-10-10 Multi-label classification method and device based on deep learning and storage medium

Publications (1)

Publication Number Publication Date
CN112651412A true CN112651412A (en) 2021-04-13

Family

ID=75342408

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910956811.1A Pending CN112651412A (en) 2019-10-10 2019-10-10 Multi-label classification method and device based on deep learning and storage medium

Country Status (1)

Country Link
CN (1) CN112651412A (en)

Similar Documents

Publication Publication Date Title
CN109376615B (en) Method, device and storage medium for improving prediction performance of deep learning network
CN109891897B (en) Method for analyzing media content
CN109816032B (en) Unbiased mapping zero sample classification method and device based on generative countermeasure network
CN106570464B (en) Face recognition method and device for rapidly processing face shielding
CN103679185B (en) Convolutional neural networks classifier system, its training method, sorting technique and purposes
CN110096938B (en) Method and device for processing action behaviors in video
CN101937513A (en) Messaging device, information processing method and program
CN111091175A (en) Neural network model training method, neural network model classification method, neural network model training device and electronic equipment
Zhang et al. Context-aware feature and label fusion for facial action unit intensity estimation with partially labeled data
CN113128478B (en) Model training method, pedestrian analysis method, device, equipment and storage medium
CN108629358B (en) Object class prediction method and device
US20200250544A1 (en) Learning method, storage medium, and learning apparatus
KR20190125029A (en) Methods and apparatuses for generating text to video based on time series adversarial neural network
CN112613349B (en) Time sequence action detection method and device based on deep hybrid convolutional neural network
JP2019075035A (en) Software test device and method
CN112508041A (en) Training method of neural network for spray control based on classification result label
CN116795886B (en) Data analysis engine and method for sales data
CN114170484B (en) Picture attribute prediction method and device, electronic equipment and storage medium
CN112613617A (en) Uncertainty estimation method and device based on regression model
CN114863407A (en) Multi-task cold start target detection method based on visual language depth fusion
JP3568181B2 (en) Neural network analyzer, storage medium
CN114842546A (en) Action counting method, device, equipment and storage medium
CN116092183A (en) Gesture recognition method and device, electronic equipment and storage medium
CN111901594A (en) Visual analysis task-oriented image coding method, electronic device and medium
CN114492601A (en) Resource classification model training method and device, electronic equipment and storage medium

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