CN111353313A - Emotion analysis model construction method based on evolutionary neural network architecture search - Google Patents
Emotion analysis model construction method based on evolutionary neural network architecture search Download PDFInfo
- Publication number
- CN111353313A CN111353313A CN202010115675.6A CN202010115675A CN111353313A CN 111353313 A CN111353313 A CN 111353313A CN 202010115675 A CN202010115675 A CN 202010115675A CN 111353313 A CN111353313 A CN 111353313A
- Authority
- CN
- China
- Prior art keywords
- chromosome
- unit
- layer
- individuals
- neural network
- 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.)
- Withdrawn
Links
Images
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
- 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
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention discloses an emotion analysis model construction method based on evolutionary neural network architecture search, which comprises the following steps of: group initialization; packaging a plurality of convolutional layer units, a plurality of pooling units and a plurality of full-connection units by taking the embedded layer as a first layer, and ending by the full-connection units to randomly generate M chromosomes; the accuracy is adopted as a fitness function to carry out fitness evaluation; selecting a plurality of chromosome individuals by adopting a roulette method to form a first chromosome population; carrying out pairwise crossing on chromosome individuals of the first chromosome population by adopting an unequal length chromosome crossing method to obtain a plurality of chromosome individuals to form a second chromosome population; adding or replacing or deleting a certain convolution layer unit or pooling unit or full-link unit of chromosome individuals of the second chromosome population; and calculating the fitness of the chromosome individuals of the second chromosome population until reaching a preset iteration number, and selecting the chromosome individuals with the optimal neural network structure by adopting the fitness.
Description
Technical Field
The invention relates to the technical field of emotion analysis model construction, in particular to an emotion analysis model construction method based on evolutionary neural network architecture search.
Background
Emotion Analysis (Sentiment Analysis) refers to the process of extracting, analyzing, processing and inducing text with emotion colors. How to effectively acquire emotion information contained in a large amount of texts and analyze and summarize the emotion information on the basis of the emotion information is the main research content of emotion analysis. Because the emotion analysis research relates to a plurality of fields such as information retrieval, machine learning, natural language processing, text mining and the like, the emotion analysis research is widely concerned by various research institutions, and meanwhile, the emotion analysis is also applied to various fields such as market competition, consumer selection, visual analysis, government election and the like.
The nature of sentiment analysis is a classification problem. A piece of text generally comprises subjective text and objective text, wherein the objective text is objectively descriptive of various things and facts and has no emotional tendency. The subjective text mainly comprises the opinions or opinions of the authors on various things and has specific emotional colors related to the dislike and the like of the authors. The subjective text and the objective text form corresponding sentences. In the prior art, the main research methods and technical schemes are divided into the following two types in the overview of the current research work of subjective text emotional tendency analysis:
the first method is a semantic-based emotional dictionary method, which mainly comprises the steps of constructing an emotional dictionary, analyzing emotional tendency words in a text and special structures of corresponding sentences, giving specific weight to each word with emotional tendency in the text sentences, setting an emotional tendency threshold value for the whole sentences, finally carrying out emotional analysis on the text by adopting a corresponding statistical method through a mathematical formula, and dividing the section of text sentences into the emotional tendency if the emotional tendency exceeds the threshold value. The method depends on the weight value given to the emotional words and the threshold value for determining the emotional tendency, and the method for judging the emotional intensity is too simple. If certain errors exist in the weight value given to the emotional words with different emotional intensities and the threshold value for determining the emotional tendency, the final emotional analysis result is not accurate enough.
The second method is based on deep learning, and generally adopts vectors to represent text sentences, a corresponding neural network is constructed, words are embedded into extracted text features and placed into the deep neural network for training, an emotion analysis model is obtained through training, and finally, the trained neural network model is used for carrying out emotion classification on the texts in a test set. The method based on deep learning often achieves better effect in the emotion classification task, but the final emotion analysis effect is different due to different designed neural networks. The neural network architecture design is usually realized by designing and building through artificial experience after years of research by field experts. In addition, the method needs a lot of time for manual parameter adjustment in the early stage, and occupies more computing resources.
Therefore, an emotion analysis method based on evolutionary neural network architecture search with high accuracy and resource saving is urgently needed to be provided. The neural network basic parameter range is given, and the structure of the neural network is automatically searched by using an evolutionary algorithm, so that the emotion analysis accuracy is improved.
Disclosure of Invention
Aiming at the problems, the invention aims to provide an emotion analysis model construction method based on evolutionary neural network architecture search, and the technical scheme adopted by the invention is as follows:
the emotion analysis model construction method based on evolutionary neural network architecture search comprises the following steps:
step S1, group initialization is carried out on the embedding layer, the convolution layer unit, the pooling unit and the full-connection unit;
step S2, using the embedded layer as the first layer, packaging several convolution layer units, several pooling units and several full-connection units, and ending with the full-connection units, randomly generating M chromosomes; the total number of layers of any chromosome is less than the total number of layers of a preset chromosome; m is a natural number greater than 1;
step S3, adopting the accuracy as the fitness function to evaluate the fitness;
step S4, selecting a plurality of chromosome individuals by adopting a roulette method to form a first chromosome population;
step S5, crossing each chromosome individual of the first chromosome population pairwise by adopting a chromosome crossing method with unequal length to obtain a plurality of chromosome individuals to form a second chromosome population;
step S6, adding, replacing or deleting a certain convolution layer unit, pooling unit or full-link unit of chromosome individuals of the second chromosome population;
and step S7, calculating the fitness of the chromosome individuals of the second chromosome population, feeding back to the step S4 until the preset iteration times are reached, and selecting the chromosome individuals with the optimal neural network structure by adopting the fitness.
Preferably, the convolutional layer unit is formed by sequentially packaging a convolutional layer, a ReLU activation function and a BN layer.
Preferably, the full-connection unit is formed by sequentially packaging a full-connection layer and a Dropout layer.
Further, the group initializing the embedding layer, the convolutional layer unit, the pooling unit and the fully-connected unit comprises
Initializing the dictionary length, the input sequence length and the word embedding latitude of the embedding layer;
initializing the size of convolution kernels, the number of the convolution kernels and convolution step length of convolution layers of the convolution layer unit;
initializing the size of a pooling window and the step size of a pooling layer of a pooling unit;
the number of neurons in the fully-connected layer of the fully-connected unit is initialized.
Preferably, the fitness evaluation is performed by using the accuracy as a fitness function, and the expression is as follows:
wherein, P refers to the number of comments with correct emotion analysis tendency predicted by the model, and R refers to the total number of comments needing to be predicted.
Furthermore, a plurality of chromosome individuals are selected by adopting a roulette method to form a first chromosome population, and the specific steps are as follows:
calculating the fitness f (x) of any chromosome individual in the first chromosome populationi) Wherein i ═ (1,2, …, M);
calculating the probability p (x) that any chromosome individual in the first chromosome population is inherited to the next generationi) The expression is as follows:
calculating the cumulative probability q of any chromosome individualiThe expression is as follows:
randomly generating a random number corresponding to any chromosome individual in the interval [0,1], and determining the selected times of the chromosome individual by using the position of the random number in the interval.
Preferably, the method for crossing each two chromosome individuals of the first chromosome population by using chromosome crossing methods with unequal lengths to obtain a plurality of chromosome individuals comprises the following steps:
splitting a convolutional layer unit, a pooling unit and a full-link unit of a chromosome individual, and forming a convolutional layer unit chain, a pooling unit chain and a full-link unit chain which are sequentially connected and are mutually independent according to positions in the chromosome individual;
aligning and crossing the left sides of the convolutional layer unit chains of the two chromosome individuals; the left sides of the pooled unit chains of the two chromosome individuals are aligned and crossed, and the left sides of the fully-linked unit chains of the two chromosome individuals are aligned and crossed.
Further, encapsulating the plurality of convolutional layer units, the plurality of pooling units and the plurality of fully connected units, and ending with the fully connected units; wherein, a plurality of convolution layer units and a plurality of pooling units are combined in an arbitrary crossing way.
Compared with the prior art, the invention has the following beneficial effects:
(1) the invention adopts an embedded layer as a first layer, converts positive integers into vectors with fixed size, aims to project relatively sparse data on each latitude to relatively low latitude, and can perform operation of a real number set on each latitude; the embedded layer is used for reducing the dimension of the data and expressing the data density, so that the subsequent operations such as convolution, pooling and the like are facilitated;
(2) the unsaturated activation function and the BN layer are added after the convolution layer to carry out encapsulation, so that the problem of corresponding gradient disappearance of the model is effectively solved, and the convergence speed of the network can be accelerated. Since there are many unsaturated activation functions ReLU and its variants, one may be randomly selected at the time of addition; meanwhile, a dropout layer is added behind the full connection layer to prevent the occurrence of overfitting;
(3) the method adopts the accuracy (accuracycacy) as a fitness function to evaluate the fitness so as to find the emotion analysis neural network with better effect;
(4) the invention adopts a roulette method to select a plurality of excellent chromosome individuals and eliminates inferior chromosome individuals;
(5) the invention adopts a chromosome crossing method with unequal length, namely, chromosome crossing is converted into each structural unit in the chromosome to carry out crossing; the design has the advantages that not only can the diversity of the whole population network be increased, but also the dominant genes of the parents can be inherited to the offspring;
(6) the invention carries out environment selection through population iteration operation to eliminate network structures with lower fitness; the method is beneficial to accelerating the evolution progress and finding the emotion analysis network structure with higher accuracy;
in conclusion, the method has the advantages of high accuracy, resource saving and the like, and has high practical value and popularization value in the technical field of emotion analysis model construction.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope of protection, and it is obvious for those skilled in the art that other related drawings can be obtained according to these drawings without inventive efforts.
FIG. 1 is a flow chart of chromosome creation according to the present invention.
FIG. 2 is a diagram of the case crossing process of chromosomes according to the present invention.
FIG. 3 is a flow chart of the present invention.
Detailed Description
To further clarify the objects, technical solutions and advantages of the present application, the present invention will be further described with reference to the accompanying drawings and examples, and embodiments of the present invention include, but are not limited to, the following examples. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Examples
As shown in fig. 1 to fig. 3, the present embodiment provides an emotion analysis model construction method based on evolutionary neural network architecture search, which mainly includes the following steps:
the first step, carry on the group initialization to embedding layer, convolution layer unit, pooling unit and all-connected unit;
in the population initialization process, an initial population is randomly generated by adopting a variable length gene coding mode, and each individual in the initialization population represents a corresponding convolutional neural network structure and an initial parameter corresponding to the structure. In the gene coding process, a convolution layer, a pooling layer and a full-connection layer corresponding to the convolutional neural network are respectively initialized. Each layer needs to include relevant initial parameters corresponding to its structural units, and specific initialization coding information is shown in table 2. It should be noted that in the process of processing text by neural network, the first layer is an Embedding layer (Embedding layer), and the Embedding layer converts positive integers into vectors with fixed size, and the purpose is to project data with relatively sparse data at each latitude to data with relatively low latitude, and each latitude can be operated by a real number set. The embedding layer is used for reducing the dimension of the data and expressing the data densely, and the subsequent operations of convolution, pooling and the like are facilitated. The convolution operation adopts 1-dimensional convolution, and the pooling operation adopts 1-dimensional pooling. Meanwhile, the invention sets an evolution range in each structural unit of the convolutional network, adds an unsaturated activation function and a BN layer behind the convolutional layer, helps the model to solve the problem of corresponding gradient disappearance and can also accelerate the convergence speed of the network. Since there are many variations of the unsaturated activation function ReLU and its variants, one may be randomly selected at the time of addition. In addition, in order to prevent the occurrence of overfitting, a dropout layer is also added after the convolutional layer and the fully-connected layer. In this embodiment, the population initialization content is shown in table 2:
TABLE 1 coding information
Secondly, packaging a plurality of convolution layer units, a plurality of pooling units and a plurality of full-connection units by taking the embedded layer as a first layer, and ending by the full-connection units to randomly generate M chromosomes; the total number of layers of any chromosome is less than the total number of layers of a preset chromosome; m is a natural number greater than 1;
when population initialization is performed and each individual chromosome is constructed, the chromosome construction is mainly composed of 4 types of structural units. The 4 types of main structural units are respectively an embedded layer, a convolution unit, a pooling unit and a full-connection unit. The layers of insertions in these 4 types of building blocks are placed foremost and occur only once when building chromosomes. After the embedding layer, there is a convolution unit. In order to accelerate the convergence speed of the neural network and avoid the problem of gradient dispersion of the model, the unsaturated activation function ReLu and a BN layer are added behind the convolution layer. Wherein ReLu is in front of BN layer. Namely, the convolution layer, the ReLU activation function and the BN layer form a convolution unit. The pooling unit is composed of a pooling layer. After the convolution unit, a pooling unit can be added, and the convolution unit can be added continuously. These sequences are randomly generated in order to find the optimal network structure sequence by genetic algorithm evolution. And finally, a full-connection unit which comprises a full-connection layer and a Dropout layer, wherein the Dropout layer is added behind the full-connection layer to prevent the occurrence of overfitting. On each chromosome, the number of convolutional layers, pooling layers, fully-connected layers will be set between a given range of total layer numbers.
Thirdly, adopting the accuracy as a fitness function to evaluate the fitness;
in the training process, the invention takes the accuracy (accuracycacy) commonly used in the training neural network as a fitness function, and the formula is as follows:
wherein P refers to the number of comments with correct emotion analysis tendency of model prediction, and R refers to the total number of comments needing prediction. The accuracy is adopted as a fitness function of the algorithm, so that the higher the accuracy of the model for identifying the emotional tendency is, the better the effect of the found neural network is. The invention herein uses Adam optimizer in Keras to compile the model, and since the emotion analysis task is essentially a binary task, the loss function used is binary cross entropy (binary cross entropy).
Fourthly, selecting a plurality of chromosome individuals by adopting a roulette method to form a first chromosome population, and specifically comprising the following steps:
(1) calculating the fitness f (x) of any chromosome individual in the first chromosome populationi) Wherein i ═ (1,2, …, M);
(2) calculating the probability p (x) that any chromosome individual in the first chromosome population is inherited to the next generationi) The expression is as follows:
(3) calculating the cumulative probability q of any chromosome individualiThe expression is as follows:
(4) randomly generating a random number corresponding to any chromosome individual in the interval [0,1], and determining the selected times of the chromosome individual by using the position of the random number in the interval.
Fifthly, pairwise crossing chromosome individuals of the first chromosome population by adopting an unequal length chromosome crossing method to obtain a plurality of chromosome individuals to form a second chromosome population; a case of crossing of two chromosome individuals is given in fig. 2: because the invention adopts the variable-length gene coding strategy in the population initialization process, the length of the chromosomes generated in each population network can be different, and the chromosomes can not be directly crossed when being crossed. This example uses the unequal-length chromosome crossing method, that is, the chromosome crossing is converted into crossing of each structural unit in the chromosome. In addition, in each structural unit of the neural network, the equal-length chromosome segments can be directly crossed, and the redundant chromosome segments are not crossed.
In fig. 2, two pairs of chromosomes are recombined, that is, the convolution unit, the pooling unit and the full-link unit in the chromosomes are extracted and recombined in pairs, and then the modules corresponding to each chromosome are crossed. In each of the intersecting modules, equal length segments of dye can be directly intersected, while the excess segments of dye do not.
And sixthly, adding, replacing or deleting a certain convolution layer unit, pooling unit or full-link unit of the chromosome individuals of the second chromosome population. It should be noted that, in the implementation process of the three variation methods, the number of network units added is large, so that the number of network layers exceeds the maximum search range set by the original evolved neural network. When the number of network layers reaches a maximum, only replacement or deletion operations can be selected. Similarly, when the number of network layers is less than or equal to the minimum number of network layers, only the addition or the replacement can be performed.
And seventhly, calculating the fitness of the chromosome individuals of the second chromosome population, feeding back to the fourth step until a preset iteration number is reached, and selecting the chromosome individuals with the optimal neural network structure by adopting the fitness.
To verify the advantages of the present invention, the following comparisons were made:
the data set (IMDB) used by the invention is from the review of 50000 network movie databases, wherein 25000 pieces of data are used as training sets, and 25000 pieces of data are used as test sets. Meanwhile, the comments of positive and negative emotional tendency in the training set and the test set respectively account for 50%. For each comment, when the comment is subjected to text vectorization representation, a dictionary containing 10000 words is established, and the maximum length of a figure list is intercepted to be 200, namely the text length of each comment finally obtained is 200, comments less than the length are filled with zero at the end, and comments beyond the length are truncated. In addition, the invention takes 20% of data provided in the training set as a validation set, so as to validate the effect of the evolved neural network. The specific division of the review data set is shown in Table 2
IMDB data set used in Table 2
In the emotion analysis problem, the emotion analysis method based on deep learning is sensitive to emotion tendency cognition, so that the situation that a large amount of time is spent on weight assignment on emotion tendency words in the early stage is avoided, the extracted key features of the text are accurate, and the emotion analysis recognition accuracy is high. The method is based on the convolutional neural network, combines with the genetic algorithm, and takes the accuracy of the neural network to emotion analysis as the fitness of the genetic algorithm under the condition of giving the initialized basic parameter range of the neural network. And continuously searching the neural network structure units in the range, and automatically designing a neural network architecture to obtain a better neural network model. Meanwhile, the dependence of the method on the manual experience is reduced. Meanwhile, the method is compared with several methods which are commonly researched at present, and the specific experimental results are as follows:
TABLE 3 comparison of the results
For the above 3 methods, MLP refers to a multi-layer perceptron, Simple-RNN is an RNN network used for emotion analysis, LSTM is a full name of long-short term memory network, and ECNN is the method proposed by the present invention, and can be obtained by mutual comparison with other methods, and the accuracy of the neural network model obtained according to the evolution method of the present invention will be better than other common methods. Therefore, the emotion analysis is performed by adopting the evolutionary neural network, and the accuracy of the emotion analysis can be effectively improved. Compared with the prior art, the method has the specific and prominent substantive characteristics and remarkable progress, and has very high practical value and popularization value in the technical field of emotion analysis model construction.
The above-mentioned embodiments are only preferred embodiments of the present invention, and do not limit the scope of the present invention, but all the modifications made by the principles of the present invention and the non-inventive efforts based on the above-mentioned embodiments shall fall within the scope of the present invention.
Claims (8)
1. The emotion analysis model construction method based on evolutionary neural network architecture search is characterized by comprising the following steps of:
step S1, group initialization is carried out on the embedding layer, the convolution layer unit, the pooling unit and the full-connection unit;
step S2, using the embedded layer as the first layer, packaging several convolution layer units, several pooling units and several full-connection units, and ending with the full-connection units, randomly generating M chromosomes; the total number of layers of any chromosome is less than the total number of layers of a preset chromosome; m is a natural number greater than 1;
step S3, adopting the accuracy as the fitness function to evaluate the fitness;
step S4, selecting a plurality of chromosome individuals by adopting a roulette method to form a first chromosome population;
step S5, crossing each chromosome individual of the first chromosome population pairwise by adopting a chromosome crossing method with unequal length to obtain a plurality of chromosome individuals to form a second chromosome population;
step S6, adding, replacing or deleting a certain convolution layer unit, pooling unit or full-link unit of chromosome individuals of the second chromosome population;
and step S7, calculating the fitness of the chromosome individuals of the second chromosome population, feeding back to the step S4 until the preset iteration times are reached, and selecting the chromosome individuals with the optimal neural network structure by adopting the fitness.
2. The method for constructing an emotion analysis model based on evolutionary neural network architecture search of claim 1, wherein the convolutional layer unit is formed by sequentially packaging a convolutional layer, a ReLU activation function and a BN layer.
3. The method for constructing an emotion analysis model based on evolutionary neural network architecture search as claimed in claim 1, wherein the fully-connected unit is formed by sequentially encapsulating a fully-connected layer and a Dropout layer.
4. The method of claim 1, wherein the group initialization of the embedding layer, the convolutional layer unit, the pooling unit and the fully-connected unit comprises
Initializing the dictionary length, the input sequence length and the word embedding latitude of the embedding layer;
initializing the size of convolution kernels, the number of the convolution kernels and convolution step length of convolution layers of the convolution layer unit;
initializing the size of a pooling window and the step size of a pooling layer of a pooling unit;
the number of neurons in the fully-connected layer of the fully-connected unit is initialized.
5. The method for constructing an emotion analysis model based on evolutionary neural network architecture search as claimed in claim 1, wherein fitness evaluation is performed by using accuracy as a fitness function, and the expression is as follows:
wherein, P refers to the number of comments with correct emotion analysis tendency predicted by the model, and R refers to the total number of comments needing to be predicted.
6. The method for constructing the emotion analysis model based on the evolutionary neural network architecture search, according to claim 5, is characterized in that a roulette method is adopted to select a plurality of chromosome individuals to form a first chromosome population, and the specific steps are as follows:
calculating the fitness f (x) of any chromosome individual in the first chromosome populationi) Wherein i ═ (1,2, …, M);
calculating the probability p (x) that any chromosome individual in the first chromosome population is inherited to the next generationi) The expression is as follows:
calculating the cumulative probability q of any chromosome individualiThe expression is as follows:
randomly generating a random number corresponding to any chromosome individual in the interval [0,1], and determining the selected times of the chromosome individual by using the position of the random number in the interval.
7. The method for constructing the emotion analysis model based on the evolutionary neural network architecture search, according to claim 1, wherein the chromosome individuals of the first chromosome population are crossed pairwise by adopting an unequal length chromosome crossing method to obtain a plurality of chromosome individuals, comprising the following steps:
splitting a convolutional layer unit, a pooling unit and a full-link unit of a chromosome individual, and forming a convolutional layer unit chain, a pooling unit chain and a full-link unit chain which are sequentially connected and are mutually independent according to positions in the chromosome individual;
aligning and crossing the left sides of the convolutional layer unit chains of the two chromosome individuals; the left sides of the pooled unit chains of the two chromosome individuals are aligned and crossed, and the left sides of the fully-linked unit chains of the two chromosome individuals are aligned and crossed.
8. The method for constructing an emotion analysis model based on evolutionary neural network architecture search of claim 1, wherein the plurality of convolutional layer units, the plurality of pooling units and the plurality of fully connected units are encapsulated and terminated by the fully connected units; wherein, a plurality of convolution layer units and a plurality of pooling units are combined in an arbitrary crossing way.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010115675.6A CN111353313A (en) | 2020-02-25 | 2020-02-25 | Emotion analysis model construction method based on evolutionary neural network architecture search |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010115675.6A CN111353313A (en) | 2020-02-25 | 2020-02-25 | Emotion analysis model construction method based on evolutionary neural network architecture search |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111353313A true CN111353313A (en) | 2020-06-30 |
Family
ID=71194182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010115675.6A Withdrawn CN111353313A (en) | 2020-02-25 | 2020-02-25 | Emotion analysis model construction method based on evolutionary neural network architecture search |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111353313A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085070A (en) * | 2020-08-19 | 2020-12-15 | 北京影谱科技股份有限公司 | Genetic algorithm-based CNN image classification method and system |
| CN112819161A (en) * | 2021-02-02 | 2021-05-18 | 四川大学 | Variable-length gene genetic algorithm-based neural network construction system, method and storage medium |
| CN113642730A (en) * | 2021-08-30 | 2021-11-12 | Oppo广东移动通信有限公司 | Convolutional network pruning method and device and electronic equipment |
| CN113821985A (en) * | 2021-11-22 | 2021-12-21 | 中移(上海)信息通信科技有限公司 | Traffic state prediction method and device and electronic equipment |
| CN114662678A (en) * | 2022-05-17 | 2022-06-24 | 华中科技大学 | Variable activation function convolution neural network and training method thereof |
| CN116522932A (en) * | 2023-06-26 | 2023-08-01 | 云目未来科技(湖南)有限公司 | Text analysis method and system |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102222267A (en) * | 2010-04-14 | 2011-10-19 | 同济大学 | Remote-sensing classification processing method for neural network by genetic algorithm |
| CN104820977A (en) * | 2015-05-22 | 2015-08-05 | 无锡职业技术学院 | BP neural network image restoration algorithm based on self-adaption genetic algorithm |
| CN109165764A (en) * | 2018-06-26 | 2019-01-08 | 昆明理工大学 | A kind of line loss calculation method of genetic algorithm optimization BP neural network |
| CN109284506A (en) * | 2018-11-29 | 2019-01-29 | 重庆邮电大学 | A kind of user comment sentiment analysis system and method based on attention convolutional neural networks |
| CN109299253A (en) * | 2018-09-03 | 2019-02-01 | 华南理工大学 | A kind of social text Emotion identification model construction method of Chinese based on depth integration neural network |
| CN109299142A (en) * | 2018-11-14 | 2019-02-01 | 中山大学 | A kind of convolutional neural networks search structure method and system based on evolution algorithm |
| US10210860B1 (en) * | 2018-07-27 | 2019-02-19 | Deepgram, Inc. | Augmented generalized deep learning with special vocabulary |
| CN109597891A (en) * | 2018-11-26 | 2019-04-09 | 重庆邮电大学 | Text emotion analysis method based on two-way length Memory Neural Networks in short-term |
| CN109902808A (en) * | 2019-02-28 | 2019-06-18 | 华南理工大学 | A method of convolutional neural networks are optimized based on floating-point numerical digit Mutation Genetic Algorithms Based |
| CN110083700A (en) * | 2019-03-19 | 2019-08-02 | 北京中兴通网络科技股份有限公司 | A kind of enterprise's public sentiment sensibility classification method and system based on convolutional neural networks |
| CN110135498A (en) * | 2019-05-17 | 2019-08-16 | 电子科技大学 | A kind of image-recognizing method based on depth Evolutionary Neural Network |
| CN110598552A (en) * | 2019-08-09 | 2019-12-20 | 吉林大学 | Expression recognition method based on improved particle swarm optimization convolutional neural network optimization |
| US20200026951A1 (en) * | 2018-07-19 | 2020-01-23 | Tata Consultancy Services Limited | Systems and methods for end-to-end handwritten text recognition using neural networks |
-
2020
- 2020-02-25 CN CN202010115675.6A patent/CN111353313A/en not_active Withdrawn
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102222267A (en) * | 2010-04-14 | 2011-10-19 | 同济大学 | Remote-sensing classification processing method for neural network by genetic algorithm |
| CN104820977A (en) * | 2015-05-22 | 2015-08-05 | 无锡职业技术学院 | BP neural network image restoration algorithm based on self-adaption genetic algorithm |
| CN109165764A (en) * | 2018-06-26 | 2019-01-08 | 昆明理工大学 | A kind of line loss calculation method of genetic algorithm optimization BP neural network |
| US20200026951A1 (en) * | 2018-07-19 | 2020-01-23 | Tata Consultancy Services Limited | Systems and methods for end-to-end handwritten text recognition using neural networks |
| US10210860B1 (en) * | 2018-07-27 | 2019-02-19 | Deepgram, Inc. | Augmented generalized deep learning with special vocabulary |
| CN109299253A (en) * | 2018-09-03 | 2019-02-01 | 华南理工大学 | A kind of social text Emotion identification model construction method of Chinese based on depth integration neural network |
| CN109299142A (en) * | 2018-11-14 | 2019-02-01 | 中山大学 | A kind of convolutional neural networks search structure method and system based on evolution algorithm |
| CN109597891A (en) * | 2018-11-26 | 2019-04-09 | 重庆邮电大学 | Text emotion analysis method based on two-way length Memory Neural Networks in short-term |
| CN109284506A (en) * | 2018-11-29 | 2019-01-29 | 重庆邮电大学 | A kind of user comment sentiment analysis system and method based on attention convolutional neural networks |
| CN109902808A (en) * | 2019-02-28 | 2019-06-18 | 华南理工大学 | A method of convolutional neural networks are optimized based on floating-point numerical digit Mutation Genetic Algorithms Based |
| CN110083700A (en) * | 2019-03-19 | 2019-08-02 | 北京中兴通网络科技股份有限公司 | A kind of enterprise's public sentiment sensibility classification method and system based on convolutional neural networks |
| CN110135498A (en) * | 2019-05-17 | 2019-08-16 | 电子科技大学 | A kind of image-recognizing method based on depth Evolutionary Neural Network |
| CN110598552A (en) * | 2019-08-09 | 2019-12-20 | 吉林大学 | Expression recognition method based on improved particle swarm optimization convolutional neural network optimization |
Non-Patent Citations (3)
| Title |
|---|
| 杨舰: "进化神经网络的研究及应用" * |
| 林丽莉,冯天瑾,周文晖,郑宏伟: "多种群并行进化神经网络的研究及应用" * |
| 王盛玉;曾碧卿;商齐;韩旭丽;: "基于词注意力卷积神经网络模型的情感分析研究" * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085070A (en) * | 2020-08-19 | 2020-12-15 | 北京影谱科技股份有限公司 | Genetic algorithm-based CNN image classification method and system |
| CN112819161A (en) * | 2021-02-02 | 2021-05-18 | 四川大学 | Variable-length gene genetic algorithm-based neural network construction system, method and storage medium |
| CN112819161B (en) * | 2021-02-02 | 2021-08-31 | 四川大学 | Neural network construction system, method and storage medium for variable-length gene genetic algorithm |
| CN113642730A (en) * | 2021-08-30 | 2021-11-12 | Oppo广东移动通信有限公司 | Convolutional network pruning method and device and electronic equipment |
| CN113821985A (en) * | 2021-11-22 | 2021-12-21 | 中移(上海)信息通信科技有限公司 | Traffic state prediction method and device and electronic equipment |
| CN114662678A (en) * | 2022-05-17 | 2022-06-24 | 华中科技大学 | Variable activation function convolution neural network and training method thereof |
| CN116522932A (en) * | 2023-06-26 | 2023-08-01 | 云目未来科技(湖南)有限公司 | Text analysis method and system |
| CN116522932B (en) * | 2023-06-26 | 2023-09-19 | 云目未来科技(湖南)有限公司 | Text analysis method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111353313A (en) | Emotion analysis model construction method based on evolutionary neural network architecture search | |
| CN111310438B (en) | Chinese sentence semantic intelligent matching method and device based on multi-granularity fusion model | |
| CN110083705B (en) | Multi-hop attention depth model, method, storage medium and terminal for target emotion classification | |
| CN108170736B (en) | Document rapid scanning qualitative method based on cyclic attention mechanism | |
| CN110609897A (en) | Multi-category Chinese text classification method fusing global and local features | |
| CN110717334A (en) | Text emotion analysis method based on BERT model and double-channel attention | |
| CN110263323A (en) | Keyword abstraction method and system based on the long Memory Neural Networks in short-term of fence type | |
| CN112507699B (en) | Remote supervision relation extraction method based on graph convolution network | |
| CN104598611B (en) | The method and system being ranked up to search entry | |
| CN112528676A (en) | Document-level event argument extraction method | |
| CN112667818B (en) | GCN and multi-granularity attention fused user comment sentiment analysis method and system | |
| CN112000772B (en) | Sentence-to-semantic matching method based on semantic feature cube and oriented to intelligent question and answer | |
| Wahid et al. | Cricket sentiment analysis from Bangla text using recurrent neural network with long short term memory model | |
| CN112232087B (en) | Specific aspect emotion analysis method of multi-granularity attention model based on Transformer | |
| CN111400494B (en) | Emotion analysis method based on GCN-Attention | |
| CN113157919B (en) | Sentence text aspect-level emotion classification method and sentence text aspect-level emotion classification system | |
| CN112784532B (en) | Multi-head attention memory system for short text sentiment classification | |
| CN113761893B (en) | Relation extraction method based on mode pre-training | |
| CN108536735B (en) | Multi-mode vocabulary representation method and system based on multi-channel self-encoder | |
| CN112395393A (en) | Remote supervision relation extraction method based on multitask and multiple examples | |
| CN112860904B (en) | External knowledge-integrated biomedical relation extraction method | |
| CN107679225A (en) | A kind of reply generation method based on keyword | |
| CN114357151A (en) | Processing method, device and equipment of text category identification model and storage medium | |
| CN114528835A (en) | Semi-supervised specialized term extraction method, medium and equipment based on interval discrimination | |
| Wu et al. | TADO: Time-varying Attention with Dual-Optimizer Model |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20200630 |