CN113095069A - Data-driven neural network architecture system based on emotion analysis - Google Patents
Data-driven neural network architecture system based on emotion analysis Download PDFInfo
- Publication number
- CN113095069A CN113095069A CN202110255563.5A CN202110255563A CN113095069A CN 113095069 A CN113095069 A CN 113095069A CN 202110255563 A CN202110255563 A CN 202110255563A CN 113095069 A CN113095069 A CN 113095069A
- Authority
- CN
- China
- Prior art keywords
- layer module
- data
- neural network
- network architecture
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/247—Thesauruses; Synonyms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/216—Parsing using statistical methods
-
- 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)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the application field of a convolutional neural network in emotion analysis, in particular to a data-driven neural network architecture system based on emotion analysis, which comprises the following modules: the system comprises an embedded layer module, a characteristic convolution layer module, a maximum pooling layer module and a compact layer module, wherein the embedded layer module is input to the whole system module; the feature convolution layer module is used for feature extraction; the maximum pooling layer module reserves local information of word combinations in the text; the compact layer module determines which classification the input belongs to according to a certain weight held in each feature node.
Description
Technical Field
The invention relates to the application field of a convolutional neural network in emotion analysis, in particular to a data-driven neural network architecture system based on emotion analysis.
Background
In the application of the convolutional neural network in emotion analysis, a good effect is achieved under the condition of small calculation amount, but a neural network model usually needs a large amount of data and a large training sample set, and when different types of networks are used, the optimal hyper-parameter setting or design selection is difficult to find.
Problems or disadvantages of the prior art: convolutional neural networks when building network architectures it is difficult to find enough data to provide these networks, optimize their parameters, and make the correct design choices.
Disclosure of Invention
Aiming at the problem that enough data are difficult to find, the invention provides a data-driven neural network architecture system based on emotion analysis.
In order to solve the technical problems, the invention adopts the technical scheme that:
a data driven neural network architecture system based on emotion analysis comprises the following modules: the system comprises an embedded layer module, a characteristic convolution layer module, a maximum pooling layer module and a compact layer module, wherein all the modules are in communication connection in a parallel mode, and the embedded layer module is input to the whole system module; the feature convolution layer module is used for feature extraction; the maximum pooling layer module reserves local information of word combinations in the text; the compact layer module determines which classification the input belongs to according to a certain weight held in each feature node.
In the embedding layer module, word-embedded representations are created or looked up for each text document, and then adjusted during training using neural networks, using pre-training word vectors created from a large corpus of text.
In the feature convolution layer module, where different layers are used in parallel to extract words, 2 grams, 3 grams, and longer features, a fixed number of 80 filters are applied in each layer to generate a feature map.
In the maximum pooling layer module, data is sampled in a sub-sample mode, the most relevant and effective characteristics are selected for the classification stage, and local information of word combinations in the text is reserved.
In the maximum pooling layer module, the length of the generated feature map is as follows by performing experiments on different data sets with different values
The feature maps are concatenated and pushed to a feed-forward layer that works as a classifier, using 0.1L2 regularization criterion and 0.35 dropout rate to mitigate overfitting, and then using binary cross-copy and Adam optimizers to compute loss and optimize training.
In the max pooling layer module, a deeper network structure D >2 is used with a larger training data set, followed by a series of W parallel convolutions with increasing filter size, then the max pooling layer stack of region r is reduced by selecting the most relevant features to classify, repeating the same convolution and max pool stack multiple times to form a network of a total of D stacks, the output of the last max pool layer being the last set of features to be used.
In the compact layer module, the size of the set region is adjusted to generate a feature map with the length of 6 to 18 units, and the optimal classification score is obtained.
Has the advantages that: the optimal choice of the current-stage architecture design is determined through a series of experiments on the neural structures of various architectures, so that a high-performance emotion analysis model is generated. Associated hyper-parameter settings and design choices are also determined. A parallel convolution with a filter length of 3 is usually sufficient to capture relevant text features. Furthermore, the maximum pool area size should be adapted to the length of the text document to produce an optimal feature map. When the aggregate region is resized to generate a feature map of length 6 to 18, the best classification score can be obtained.
Drawings
FIG. 1 is a system flow diagram of the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. 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 invention.
A data-driven neural network architecture system based on emotion analysis, as shown in fig. 1, includes the following modules: the system comprises an embedded layer module, a characteristic convolution layer module, a maximum pooling layer module and a compact layer module, wherein all the modules are in communication connection in a parallel mode, and the embedded layer module is input to the whole system module; the feature convolution layer module is used for feature extraction; the maximum pooling layer module reserves local information of word combinations in the text; the compact layer module determines which classification the input belongs to according to a certain weight held in each feature node.
Further, in the embedding layer module, which is a non-trainable embedding layer, requiring the creation or lookup of word embedding representations for each text document, one possible approach is to initialize the vectors for each word with random values and then adjust them using neural networks during training, which requires a large set of training documents, higher performance can be obtained with pre-trained word vectors created from a large corpus of text where a small training data set is available, the pre-trained vectors for each word are all directly sourced.
Further, in the feature convolution layer module, where W filters are sized so as to be k-1, k-2, k-3 … all of these layers are used in parallel to extract words, 2 grams, 3 grams and longer features, a fixed number of 80 filters are applied in each layer to generate a feature map, and an important parameter of the network is its depth D or the total number of layer stacks.
Further, the max pooling layer module, whose role is to subsample sample the data and select the most relevant and efficient features for the classification phase, utilizes the max pooling layer, which is known to retain local information of word combinations in the text.
Further, in the max pooling level module, the length of the feature map generated by performing experiments on different data sets with different values for the length of the region r is
Where n is the number of words in the training document, the feature maps are concatenated and pushed to the feed forward layer working as a classifier; the latter consists of 80 nodes, overfitting is relieved by adopting 0.1L2 regularization specifications and 0.35 drop-out rate, loss calculation and optimization training are calculated by adopting a binary cross copy and Adam optimizer, and 70/10/20% of data segmentation is used in training, development and testing respectively.
Further, in the max pooling level module, a deeper level of the network structure D >2 is used with a larger training data set, followed by a series of W parallel convolutions with increasing filter size, the convolution operation acts as a word and phrase feature extractor, and the W width can vary depending on the task and the type of data being analyzed; the following is the maximum pooling level stack of the region r, and as with W, the value of r can be adjusted to accommodate the number of words in each document of the training set, and the same convolution and maximum pooling stack are repeated multiple times to form a network of a total of D stacks, the value of D being flexible and should be selected according to the size of the training data set, and the output of the last maximum pooling level is the last set of properties to be used.
Furthermore, in the compact layer module, the length of the document and the size of the feature map have certain regularity, and the size of the set area is adjusted to generate the feature map with the length of 6 to 18 units, so that the optimal classification score can be obtained; applying deep networks to larger datasets yields optimal results in terms of dataset size.
Although only the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and all changes are encompassed in the scope of the present invention.
Claims (7)
1. A data-driven neural network architecture system based on emotion analysis is characterized by comprising the following modules: the system comprises an embedded layer module, a characteristic convolution layer module, a maximum pooling layer module and a compact layer module, wherein all the modules are in communication connection in a parallel mode, and the embedded layer module is input to the whole system module; the feature convolution layer module is used for feature extraction; the maximum pooling layer module reserves local information of word combinations in the text; the compact layer module determines which classification the input belongs to according to a certain weight held in each feature node.
2. The data-driven neural network architecture system based on emotion analysis, as claimed in claim 1, wherein: in the embedding layer module, word-embedded representations are created or looked up for each text document, and then adjusted during training using neural networks, using pre-training word vectors created from a large corpus of text.
3. The system of claim 2, wherein the neural network architecture system is based on sentiment analysis, and comprises: in the feature convolution layer module, where different layers are used in parallel to extract words, 2 grams, 3 grams, and longer features, a fixed number of 80 filters are applied in each layer to generate a feature map.
4. The data-driven neural network architecture system based on emotion analysis according to claim 3, wherein: in the maximum pooling layer module, data is sampled in a sub-sample mode, the most relevant and effective characteristics are selected for the classification stage, and local information of word combinations in the text is reserved.
5. The data-driven neural network architecture system based on emotion analysis according to claim 4, wherein: the max pooling layer module is generated by performing experiments on different data sets with different valuesHas a length of
The feature maps are concatenated and pushed to a feed-forward layer that works as a classifier, using 0.1L2 regularization criterion and 0.35 dropout rate to mitigate overfitting, and then using binary cross-copy and Adam optimizers to compute loss and optimize training.
6. The system of claim 5, wherein the neural network architecture comprises: in the max pooling layer module, a deeper network structure D >2 is used with a larger training data set, followed by a series of W parallel convolutions with increasing filter size, then the max pooling layer stack of region r is reduced by selecting the most relevant features to classify, repeating the same convolution and max pool stack multiple times to form a network of a total of D stacks, the output of the last max pool layer being the last set of features to be used.
7. The data-driven neural network architecture system based on emotion analysis according to claim 6, wherein: in the compact layer module, the size of the set region is adjusted to generate a feature map with the length of 6 to 18 units, and the optimal classification score is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110255563.5A CN113095069A (en) | 2021-03-09 | 2021-03-09 | Data-driven neural network architecture system based on emotion analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110255563.5A CN113095069A (en) | 2021-03-09 | 2021-03-09 | Data-driven neural network architecture system based on emotion analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113095069A true CN113095069A (en) | 2021-07-09 |
Family
ID=76667029
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110255563.5A Pending CN113095069A (en) | 2021-03-09 | 2021-03-09 | Data-driven neural network architecture system based on emotion analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113095069A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE202023102803U1 (en) | 2023-05-22 | 2023-07-17 | Pradeep Bedi | System for emotion detection and mood analysis through machine learning |
-
2021
- 2021-03-09 CN CN202110255563.5A patent/CN113095069A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE202023102803U1 (en) | 2023-05-22 | 2023-07-17 | Pradeep Bedi | System for emotion detection and mood analysis through machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zou et al. | Layer-dependent importance sampling for training deep and large graph convolutional networks | |
| CN111126386B (en) | Sequence domain adaptation method based on countermeasure learning in scene text recognition | |
| CN112699247B (en) | Knowledge representation learning method based on multi-class cross entropy contrast complement coding | |
| CN103324628B (en) | A kind of trade classification method and system for issuing text | |
| CN108764460A (en) | A kind of Time Series Forecasting Methods based on time convolution sum LSTM | |
| CN108319666A (en) | A kind of electric service appraisal procedure based on multi-modal the analysis of public opinion | |
| CN105512311A (en) | Chi square statistic based self-adaption feature selection method | |
| JP6838161B2 (en) | End-to-end modeling methods and systems | |
| CN102567742A (en) | Automatic classification method of support vector machine based on selection of self-adapting kernel function | |
| Xu et al. | Unimelb at semeval-2016 tasks 4a and 4b: An ensemble of neural networks and a word2vec based model for sentiment classification | |
| CN112699899A (en) | Hyperspectral image feature extraction method based on generation countermeasure network | |
| CN102968410A (en) | Text classification method based on RBF (Radial Basis Function) neural network algorithm and semantic feature selection | |
| US20210224650A1 (en) | Method for differentiable architecture search based on a hierarchical grouping mechanism | |
| CN106991127A (en) | A kind of knowledget opic short text hierarchy classification method extended based on topological characteristic | |
| CN107357895B (en) | Text representation processing method based on bag-of-words model | |
| CN113095069A (en) | Data-driven neural network architecture system based on emotion analysis | |
| CN107273922A (en) | A kind of screening sample and weighing computation method learnt towards multi-source instance migration | |
| CN114612761A (en) | Network architecture searching method for image recognition | |
| CN111353534A (en) | Graph data category prediction method based on adaptive fractional order gradient | |
| Sotthisopha et al. | Improving short text classification using fast semantic expansion on multichannel convolutional neural network | |
| Shinghal et al. | A bottom-up and top-down approach to using context in text recognition | |
| CN108647206A (en) | Chinese spam filtering method based on chaotic particle swarm optimization CNN networks | |
| CN111651660B (en) | Method for cross-media retrieval of difficult samples | |
| CN117033961A (en) | Multi-mode image-text classification method for context awareness | |
| CN109919235B (en) | Deep learning image classification model training method based on manual intervention sample set weight |
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 |