CN110889282A - Text emotion analysis method based on deep learning - Google Patents
Text emotion analysis method based on deep learning Download PDFInfo
- Publication number
- CN110889282A CN110889282A CN201911189487.1A CN201911189487A CN110889282A CN 110889282 A CN110889282 A CN 110889282A CN 201911189487 A CN201911189487 A CN 201911189487A CN 110889282 A CN110889282 A CN 110889282A
- Authority
- CN
- China
- Prior art keywords
- matrix
- information
- node
- graph
- emo
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a text emotion analysis method based on deep learning. (1) Inputting text data, removing stop words, extracting keywords, and forming a keyword set. (2) Forming a dense subgraph by constructing a keyword co-occurrence graph; obtaining vector representation of sentences in the subgraph and the document, and further distributing the sentences to the subgraph; designing edge connection and edge weight between subgraphs to form topological interactive graph expression of the document; (3) and taking the topological interactive graph as the input of an Emo-GCN model, performing extraction node characteristic transformation, and then fusing local structure information to obtain a node aggregation matrix. And carrying out nonlinear transformation on the aggregated information. The Emo-GCN model adopts a hierarchical structure, and features are extracted layer by layer. The invention adopts a novel topological interactive graph to express text information and further uses a graph convolution neural network to carry out text emotion analysis, and still has strong adaptability. The method is applied to product recommendation, market prediction and decision adjustment, and has extremely high commercial value.
Description
Technical Field
The invention relates to a natural language processing method and an image classification method, in particular to a text emotion analysis method.
Background
Text classification is a classic problem in the field of natural language processing, emotion recognition is a relatively challenging task in text classification, and the current methods for processing emotion analysis problems mainly include the following three types: firstly, constructing an emotion dictionary for emotion analysis, constructing emotion vocabularies into the emotion dictionary which is necessary and insufficient for emotion analysis, wherein the emotion dictionary cannot contain all emotion expression forms no matter how the emotion dictionary content is expanded, the emotion polarity of some vocabularies is not clear, the emotion vocabularies may not be used in some sentences but also express certain emotion, and some emotion vocabularies express just opposite meanings in some contexts, so that the method is limited by the problems; secondly, traditional machine learning methods such as Logistic Regression (Logistic Regression), Naive Bayes (Naive Bayes) and the like, wherein Logistic Regression can only be used for linear classification problems, the conclusion of Naive Bayes is based on the prior probability of features, and the features are assumed to be completely independent, and the requirements are often unsatisfied in practical situations, and the classification effect is sometimes poor and satisfactory, so that the method has great limitation; third, the current popular deep learning methods such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and the like all use data based on the european space, and have the greatest characteristic of having a regular spatial structure, and the features can be expressed by a one-dimensional or two-dimensional matrix, so that the processing is relatively efficient. However, in real life, most data are irregular structures, the data can be expressed by using a topological interactive graph without translational invariance, a fixed convolution kernel is difficult to select to adapt to the irregularity of the whole graph, and the data of the structures can cause the CNN or the RNN to lose effectiveness instantly.
Because the nature of text emotion has complexity, a piece of text is not a simple linear combination of words, when describing a sentence, the sentence is regarded as a whole rather than a word set, different meanings and emotions can be expressed by different combinations, different sequences and different numbers of words, and the difficulty of text emotion analysis is caused, so the text emotion analysis is actually a simulation of human mind. The true emotion judgment is not a simple rule listing, but rather a complex network. Therefore, text information is abstracted into a topological interactive graph expression by adopting an Emotion GraphConvolitional Network (Emo-GCN) of text Emotion analysis, and the Emo-GCN can extract features in the graph, so that nodes in the graph are classified, and graph data are classified. The graph data has two features in space: the method comprises the following steps that firstly, node characteristics are provided, and each node has own characteristics which are reflected on points; and secondly, structural characteristics, namely certain relation exists between nodes, and the characteristics are reflected on edges. In general, the graph data considers both node information and structural information, and the graph convolution neural network can learn not only the node characteristics but also the associated information between nodes, so as to finally achieve the purpose of emotion analysis.
Disclosure of Invention
The invention aims to provide a text emotion analysis method based on deep learning with better accuracy.
The purpose of the invention is realized as follows:
(1) preprocessing text data: removing stop words and extracting keywords, wherein the extracted keywords are keyword sets formed by a TextRank keyword extraction algorithm;
(2) constructing a document topological interaction graph: forming a dense subgraph by constructing a keyword co-occurrence graph; obtaining vector representation of sentences in the subgraph and the document, and further distributing the sentences to the subgraph; designing edge connection and edge weight between subgraphs to form topological interactive graph expression of the document;
(3) executing an Emo-GCN training model: taking the topological interactive graph formed in the step (2) as the input of an Emo-GCN model, and firstly performing extraction node feature transformation in the Emo-GCN training model: each node transmits the self characteristic information to the neighbor node after transforming, and the neighbor node passes through the adjacency momentArray A and feature matrix H(l)Obtaining the summary of the neighbor characteristics of each vertex; secondly, fusing local structure information: each node aggregates the characteristic information of the neighbor nodes to obtain a node aggregation matrix; and performing nonlinear transformation on the aggregated information to enhance the expression capability of the model, wherein the nonlinear transformation adopts a nonlinear activation function.
The present invention may further comprise:
1. in step (2), whether to co-occur is determined by determining whether two words appear in one sentence at the same time.
2. The Emo-GCN training model executed in the step (3) specifically comprises the following steps:
(3-1) initial input as a feature matrix H(0)＝N×F(0)Where N is the number of nodes, F(0)For each node, the propagation rules for convolutional layers are as follows:
wherein the adjacency matrix
Is to carry out normalization operation on the adjacent matrix in order to maintain the characteristic matrix H in the information transmission process(l)The original distribution of the sample;
(3-2) extracting node feature transformation: each node sends the changed characteristic information to the neighbor nodes through the adjacency matrix A and the characteristic matrix H(l)Multiplying to obtain summary information of the neighbor characteristics of each vertex;
(3-3) fusing local structural information: each node aggregates the feature information of the neighbor nodes, namely the feature summarizing matrix obtained in the last step is multiplied by the weight value matrix W(l)Obtaining an aggregation matrix;
(3-4) nonlinear transformation: performing nonlinear transformation on the information aggregated in the step (3-3), wherein sigma in a propagation rule formula is a nonlinear activation function, and the nonlinear activation function is a ReLU function or a Sigmoid function;
and (3-5) adding a pooling layer after the convolution layer.
The invention provides a text emotion analysis method with better accuracy based on deep learning. The deep learning method adopts an Emo-GCN algorithm.
Compared with the prior art, the invention has the following advantages: the Emo-GCN model adopts a hierarchical structure, and features are extracted layer by layer, so that an analysis result is more accurate; an end-to-end training mode is adopted, and the model can learn the characteristic information and the structural information of the fusion node by itself only by giving a mark to the node in the graph; in addition, the method can process Non Euclidean structure data, and breaks through the limitation that the CNN can only process translation invariance data. The invention breaks through the inherent mode of the existing text emotion analysis method, adopts novel topological interactive graph expression to further use the graph convolution neural network to analyze the text emotion, and still has strong adaptability.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a block diagram of the Emo-GCN algorithm;
FIG. 3 is a comparative line graph of the Emo-GCN algorithm and other mainstream methods LSTM, CNN and TextCNN on the accuracy rate of the text emotion analysis result.
Detailed Description
The text emotion analysis method based on deep learning mainly comprises the following steps:
(1) preprocessing text data: and removing stop words, extracting keywords, and adopting a TextRank keyword extraction algorithm when extracting the keywords so as to form a keyword set.
(2) Constructing a document topological interaction graph: forming a dense subgraph by constructing a keyword Graph (Key Graph); obtaining vector representation of sentences in the subgraph and the document, and further distributing the sentences to the subgraph; and designing edge connection and edge weight between the subgraphs to form the topological interactive graph expression of the document.
(3) Executing an Emo-GCN training model: taking the topological interaction graph formed in the step (2) as an input of an Emo-GCN model, wherein the model is a first-order local approximation of spectrogram convolution and is oneThe method comprises the following steps that (1) a plurality of graph convolution neural networks with a plurality of layers are provided, each convolution layer processes first-order neighborhood information, information transmission between layers can be realized by overlapping the plurality of convolution layers, and in the model, extraction node feature transformation is firstly carried out: each node transforms its own characteristic information and sends it to the neighbor nodes, passing through the adjacency matrix A and the characteristic matrix H(l)Obtaining the summary of the neighbor characteristics of each vertex; fusing local structure information: and each node aggregates the characteristic information of the neighbor nodes to obtain a node aggregation matrix. And performing nonlinear transformation on the aggregated information to enhance the expression capability of the model, wherein the nonlinear transformation adopts a nonlinear activation function, such as a ReLU function, a Sigmoid function and the like.
Determining whether two words appear in one sentence at the same time in step (2); when the sentences are distributed to the subgraphs, the vector representation of the sentences and the subgraphs can be obtained, and the sentences can be distributed by calculating cosine similarity; coordinate representations of sentences and subgraphs can also be obtained and sentences can be assigned by calculating euclidean distances.
In the step (3), the Emo-GCN training model is executed, and the specific steps are as follows:
(3-1) initial input as a feature matrix H(0)＝N×F(0)Where N is the number of nodes, F(0)For each node, the propagation rules for convolutional layers are as follows:
wherein the adjacency matrix
In order to keep the characteristic information of the self node when information is propagated between layers,
is to carry out normalization operation on the adjacent matrix in order to maintain the characteristic matrix H in the information transmission process(l)The original distribution of the characteristic distribution of the vertex with higher degree and the vertex with lower degree are prevented from generating greatThe difference in (a).
(3-2) extracting node feature transformation: each node sends the changed characteristic information to the neighbor nodes through the adjacency matrix A and the characteristic matrix H(l)And multiplying to obtain the summary information of the neighbor characteristics of each vertex.
(3-3) fusing local structural information: each node aggregates the feature information of the neighbor nodes, namely the feature summarizing matrix obtained in the last step is multiplied by the weight value matrix W(l)An aggregation matrix is obtained.
(3-4) nonlinear transformation: and (4) carrying out nonlinear transformation on the information aggregated in the step (3-3) to enhance the expression capability of the model, wherein sigma in the propagation rule formula is a nonlinear activation function, such as a ReLU function, a Sigmoid function and the like.
(3-5) adding a pooling layer after the convolutional layer prevents the over-fitting problem from occurring to reduce the loss.
The invention is described in more detail below by way of example.
(1) Document data preprocessing: and removing stop words and illegal characters in the document, extracting keywords, and adopting a TextRank keyword extraction algorithm when extracting the keywords so as to form a keyword set. The extraction formula of the TextRank key words is as follows:
wherein S (V)i) Is a ViThe initial value may be set to 1; d is the damping coefficient, typically set to 0.85; in (V)i) Nodes corresponding to the degree of entry in the graph; out (V)j) The nodes corresponding to out degrees in the graph.
(2) Constructing a keyword Graph (Key Graph): each keyword constitutes a vertex in the graph, and the vertices are connected if some keywords co-occur at least in one sentence. If some of the key vertices are highly related to each other, a dense subgraph is formed. The figure reveals the connections between the various keywords.
(3) Sentence allocation: firstly, obtaining the vector representation of each sentence, and the operation is realized by using TF-IDF algorithm, and we can also obtain TF-IDF vector representation of each subgraph in the step (2). And calculating cosine similarity between the subgraph and the sentences, and selecting the subgraph with the highest similarity for sentence distribution.
(4) Constructing a document topological interaction graph: abstracting the subgraph in the step (3) into new vertexes so as to form vertexes in the document topological interaction graph, and calculating cosine similarity between sentences distributed on every two vertexes for edges between the vertexes and edge weights.
(5) Taking the topological interactive graph formed in the step (4) as the input of an Emo-GCN model, wherein the model is a first-order local approximation of spectrogram convolution and is a graph convolution neural network with multiple layers, each convolution layer processes first-order neighborhood information, and information transmission between the layers can be realized by superposing a plurality of convolution layers, and in the model, the initial input is a feature matrix H(0)＝N×F(0)Where N is the number of nodes, F(0)For each node, the propagation rules for convolutional layers are as follows:
wherein
INIs an identity matrix, aims to keep the characteristic information of the node when information is propagated between layers,
is to carry out normalization operation on the adjacent matrix in order to maintain the characteristic matrix H in the information transmission process(l)The original distribution of the feature distribution prevents the vertexes with higher degrees and the vertexes with lower degrees from generating great difference on the feature distribution. Firstly, extracting node characteristicsAnd (3) sign transformation: each node sends the changed characteristic information to the neighbor nodes through the adjacency matrix A and the characteristic matrix H(l)And multiplying to obtain the summary of the neighbor characteristics of each vertex.
(6) Fusing local structure information: each node aggregates the feature information of the neighbor nodes, namely the feature summarizing matrix obtained in the step (5) is multiplied by the weight value matrix W(l)An aggregation matrix is obtained.
(7) Nonlinear transformation: and (4) carrying out nonlinear transformation on the information aggregated in the step (6) to enhance the expression capability of the model, wherein sigma in the propagation rule formula is a nonlinear activation function, such as a ReLU function. The expression of the ReLU function is:
the function divides the input into two sections for demapping, when the input value is smaller than 0, the original value is mapped to 0, and if the input value is larger than 0, the original value is transferred. The advantage of using this function is that the convergence speed is fast, and the activation value can be obtained by a threshold value of the healing drug, and the calculation complexity is low.
(8) Adding a pooling layer after the convolutional layer prevents the over-fitting problem from occurring to reduce losses.
In order to more clearly describe the effectiveness of the text emotion analysis method based on deep learning provided by the invention, table 1 shows the emotion recognition accuracy of various algorithms in 10 experiments.
Table 110 Emotion analysis accuracy for various algorithms
| Training times \ algorithm model | LSTM | CNN | TextCNN | Emo- |
| 1 | 0.82285 | 0.54799 | 0.75382 | 0.85232 |
| 2 | 0.87190 | 0.87009 | 0.79276 | 0.85763 |
| 3 | 0.87049 | 0.86516 | 0.81141 | 0.85632 |
| 4 | 0.86602 | 0.87237 | 0.82849 | 0.86408 |
| 5 | 0.86414 | 0.86336 | 0.83648 | 0.86688 |
| 6 | 0.85433 | 0.85615 | 0.84510 | 0.86266 |
| 7 | 0.85912 | 0.85834 | 0.84745 | 0.86034 |
| 8 | 0.85795 | 0.85458 | 0.84957 | 0.87657 |
| 9 | 0.85176 | 0.85513 | 0.85066 | 0.87121 |
| 10 | 0.85184 | 0.85192 | 0.85035 | 0.88313 |
As can be seen from Table 1, in comparison of various algorithms used for text emotion analysis in 10 experiments, the Emo-GCN algorithm provided by the invention has high classification and identification accuracy. Therefore, the text emotion analysis method based on deep learning provided by the invention has better emotion recognition accuracy and still has certain adaptability under the condition of sparse data.
The deep learning method adopts an emotion recognition algorithm (Emo-GCN) of a graph convolution neural network. The invention is characterized in that: (1) preprocessing text data: inputting text data, removing stop words, extracting keywords, and adopting a TextRank keyword extraction algorithm when extracting the keywords so as to form a keyword set. (2) Constructing a document topological interaction graph: forming a dense subgraph by constructing a keyword Graph (Key Graph); obtaining vector representation of sentences in the subgraph and the document, and further distributing the sentences to the subgraph; and designing edge connection and edge weight between the subgraphs to form the topological interactive graph expression of the document. (3) Executing an Emo-GCN training model: and taking the topological interactive graph as the input of an Emo-GCN model, wherein in the model, firstly, extraction node characteristic transformation is carried out, and then, local structure information is fused to obtain a node aggregation matrix. And performing nonlinear transformation on the aggregated information, enhancing the expression capability of the model, and adding a pooling layer to prevent an overfitting phenomenon. The Emo-GCN model adopts a hierarchical structure, and features are extracted layer by layer, so that an analysis result is more accurate. The invention breaks through the inherent mode of the existing text emotion analysis method, adopts a novel topological interactive graph to express text information and further uses a graph convolution neural network to carry out text emotion analysis, and still has strong adaptability. The method is applied to the aspects of product recommendation, market prediction, decision adjustment and the like, and has extremely high commercial value.
Claims (4)
1. A text emotion analysis method based on deep learning is characterized by comprising the following steps:
(1) preprocessing text data: removing stop words and extracting keywords, wherein the extracted keywords are keyword sets formed by a TextRank keyword extraction algorithm;
(2) constructing a document topological interaction graph: forming a dense subgraph by constructing a keyword co-occurrence graph; obtaining vector representation of sentences in the subgraph and the document, and further distributing the sentences to the subgraph; designing edge connection and edge weight between subgraphs to form topological interactive graph expression of the document;
(3) executing an Emo-GCN training model: taking the topological interactive graph formed in the step (2) as the input of an Emo-GCN model, and firstly performing extraction node feature transformation in the Emo-GCN training model: each node transforms its own characteristic information and sends it to the neighbor nodes, passing through the adjacency matrix A and the characteristic matrix H(l)Obtaining the summary of the neighbor characteristics of each vertex; secondly, fusing local structure information: each node aggregates the characteristic information of the neighbor nodes to obtain a node aggregation matrix; and performing nonlinear transformation on the aggregated information to enhance the expression capability of the model, wherein the nonlinear transformation adopts a nonlinear activation function.
2. The method for analyzing text emotion based on deep learning of claim 1, wherein: in step (2), whether to co-occur is determined by determining whether two words appear in one sentence at the same time.
3. The method for analyzing text emotion based on deep learning of claim 1 or 2, wherein the Emo-GCN training model executed in step (3) specifically comprises the following steps:
(3-1) initial input as a feature matrix H(0)＝N×F(0)Where N is the number of nodes, F(0)For each node, the propagation rules for convolutional layers are as follows:
wherein the adjacency matrix
Is to carry out normalization operation on the adjacent matrix in order to maintain the characteristic matrix H in the information transmission process(l)The original distribution of the sample;
(3-2) extracting node feature transformation: each node sends the changed characteristic information to the neighbor nodes through the adjacency matrix A and the characteristic matrix H(l)Multiplying to obtain summary information of the neighbor characteristics of each vertex;
(3-3) fusing local structural information: each node aggregates the feature information of the neighbor nodes, namely the feature summarizing matrix obtained in the last step is multiplied by the weight value matrix W(l)Obtaining an aggregation matrix;
(3-4) nonlinear transformation: performing nonlinear transformation on the information aggregated in the step (3-3), wherein sigma in the propagation rule formula is a nonlinear activation function;
and (3-5) adding a pooling layer after the convolution layer.
4. The method for analyzing text emotion based on deep learning of claim 3, wherein: the nonlinear activation function is a ReLU function or a Sigmoid function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911189487.1A CN110889282B (en) | 2019-11-28 | 2019-11-28 | Text emotion analysis method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911189487.1A CN110889282B (en) | 2019-11-28 | 2019-11-28 | Text emotion analysis method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110889282A true CN110889282A (en) | 2020-03-17 |
| CN110889282B CN110889282B (en) | 2023-03-21 |
Family
ID=69749199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911189487.1A Active CN110889282B (en) | 2019-11-28 | 2019-11-28 | Text emotion analysis method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110889282B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111445963A (en) * | 2020-03-31 | 2020-07-24 | 桂林电子科技大学 | Subgraph isomorphic constraint solving method based on graph node information aggregation |
| CN111489192A (en) * | 2020-03-27 | 2020-08-04 | 北京理工大学 | Product share trend prediction method integrating ICT supply chain network topological features |
| CN111563164A (en) * | 2020-05-07 | 2020-08-21 | 成都信息工程大学 | Specific target emotion classification method based on graph neural network |
| CN112784011A (en) * | 2021-01-04 | 2021-05-11 | 南威软件股份有限公司 | Emotional problem processing method, device and medium based on CNN and LSTM |
| CN113220893A (en) * | 2021-07-09 | 2021-08-06 | 北京邮电大学 | Product feedback analysis system and method based on emotion analysis |
| CN114048305A (en) * | 2021-11-05 | 2022-02-15 | 山东大学 | Plan recommendation method for administrative penalty documents based on graph convolution neural network |
| CN115858725A (en) * | 2022-11-22 | 2023-03-28 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Method and system for screening text noise based on unsupervised graph neural network |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101398814A (en) * | 2007-09-26 | 2009-04-01 | 北京大学 | Method and system for simultaneously abstracting document summarization and key words |
| CN105894088A (en) * | 2016-03-25 | 2016-08-24 | 苏州赫博特医疗信息科技有限公司 | Medical information extraction system and method based on depth learning and distributed semantic features |
| CN106991127A (en) * | 2017-03-06 | 2017-07-28 | 西安交通大学 | A kind of knowledget opic short text hierarchy classification method extended based on topological characteristic |
| CN107609634A (en) * | 2017-08-21 | 2018-01-19 | 哈尔滨工程大学 | A kind of convolutional neural networks training method based on the very fast study of enhancing |
| AU2019100371A4 (en) * | 2019-04-05 | 2019-05-16 | Ba, He Mr | A Sentiment Analysis System Based on Deep Learning |
| CN109977232A (en) * | 2019-03-06 | 2019-07-05 | 中南大学 | A kind of figure neural network visual analysis method for leading figure based on power |
| CN109992779A (en) * | 2019-03-29 | 2019-07-09 | 长沙理工大学 | A kind of sentiment analysis method, apparatus, equipment and storage medium based on CNN |
| CN110245874A (en) * | 2019-03-27 | 2019-09-17 | 中国海洋大学 | A kind of Decision fusion method based on machine learning and knowledge reasoning |
| CN110277165A (en) * | 2019-06-27 | 2019-09-24 | 清华大学 | Aided diagnosis method, device, equipment and storage medium based on figure neural network |
| CN110297889A (en) * | 2019-06-28 | 2019-10-01 | 南京冰鉴信息科技有限公司 | A kind of enterprise's Sentiment orientation analysis method based on Fusion Features |
| CN110349185A (en) * | 2019-07-12 | 2019-10-18 | 安徽大学 | A kind of training method and device of RGBT target following model |
| CN110399857A (en) * | 2019-08-01 | 2019-11-01 | 西安邮电大学 | A kind of brain electricity emotion identification method based on figure convolutional neural networks |
| CN110472003A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | Social networks text emotion fine grit classification method based on figure convolutional network |
-
2019
- 2019-11-28 CN CN201911189487.1A patent/CN110889282B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101398814A (en) * | 2007-09-26 | 2009-04-01 | 北京大学 | Method and system for simultaneously abstracting document summarization and key words |
| CN105894088A (en) * | 2016-03-25 | 2016-08-24 | 苏州赫博特医疗信息科技有限公司 | Medical information extraction system and method based on depth learning and distributed semantic features |
| CN106991127A (en) * | 2017-03-06 | 2017-07-28 | 西安交通大学 | A kind of knowledget opic short text hierarchy classification method extended based on topological characteristic |
| CN107609634A (en) * | 2017-08-21 | 2018-01-19 | 哈尔滨工程大学 | A kind of convolutional neural networks training method based on the very fast study of enhancing |
| CN109977232A (en) * | 2019-03-06 | 2019-07-05 | 中南大学 | A kind of figure neural network visual analysis method for leading figure based on power |
| CN110245874A (en) * | 2019-03-27 | 2019-09-17 | 中国海洋大学 | A kind of Decision fusion method based on machine learning and knowledge reasoning |
| CN109992779A (en) * | 2019-03-29 | 2019-07-09 | 长沙理工大学 | A kind of sentiment analysis method, apparatus, equipment and storage medium based on CNN |
| AU2019100371A4 (en) * | 2019-04-05 | 2019-05-16 | Ba, He Mr | A Sentiment Analysis System Based on Deep Learning |
| CN110277165A (en) * | 2019-06-27 | 2019-09-24 | 清华大学 | Aided diagnosis method, device, equipment and storage medium based on figure neural network |
| CN110297889A (en) * | 2019-06-28 | 2019-10-01 | 南京冰鉴信息科技有限公司 | A kind of enterprise's Sentiment orientation analysis method based on Fusion Features |
| CN110349185A (en) * | 2019-07-12 | 2019-10-18 | 安徽大学 | A kind of training method and device of RGBT target following model |
| CN110399857A (en) * | 2019-08-01 | 2019-11-01 | 西安邮电大学 | A kind of brain electricity emotion identification method based on figure convolutional neural networks |
| CN110472003A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | Social networks text emotion fine grit classification method based on figure convolutional network |
Non-Patent Citations (3)
| Title |
|---|
| KIPF THOMAS N. 等: "Semi-supervised classification with graph convolutional networks", 《网页在线公开:HTTPS://ARXIV.ORG/ABS/1609.02907》 * |
| ZHANG TONG 等: "GCB-Net: Graph Convolutional Broad Network and Its Application in Emotion Recognition", 《IEEE TRANSACTIONS ON AFFECTIVE COMPUTING》 * |
| 王勇: "社会网络结构及影响力分析方法研究", 《中国优秀博硕士学位论文全文数据库(博士)基础科学辑》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111489192A (en) * | 2020-03-27 | 2020-08-04 | 北京理工大学 | Product share trend prediction method integrating ICT supply chain network topological features |
| CN111445963A (en) * | 2020-03-31 | 2020-07-24 | 桂林电子科技大学 | Subgraph isomorphic constraint solving method based on graph node information aggregation |
| CN111445963B (en) * | 2020-03-31 | 2023-03-10 | 桂林电子科技大学 | Subgraph isomorphic constraint solving method based on graph node information aggregation |
| CN111563164A (en) * | 2020-05-07 | 2020-08-21 | 成都信息工程大学 | Specific target emotion classification method based on graph neural network |
| CN111563164B (en) * | 2020-05-07 | 2022-06-28 | 成都信息工程大学 | Specific target emotion classification method based on graph neural network |
| CN112784011B (en) * | 2021-01-04 | 2023-06-30 | 南威软件股份有限公司 | Emotion problem processing method, device and medium based on CNN and LSTM |
| CN112784011A (en) * | 2021-01-04 | 2021-05-11 | 南威软件股份有限公司 | Emotional problem processing method, device and medium based on CNN and LSTM |
| CN113220893A (en) * | 2021-07-09 | 2021-08-06 | 北京邮电大学 | Product feedback analysis system and method based on emotion analysis |
| CN113220893B (en) * | 2021-07-09 | 2021-10-01 | 北京邮电大学 | Product feedback analysis system and method based on emotion analysis |
| CN114048305A (en) * | 2021-11-05 | 2022-02-15 | 山东大学 | Plan recommendation method for administrative penalty documents based on graph convolution neural network |
| CN114048305B (en) * | 2021-11-05 | 2024-06-07 | 山东大学 | Class case recommendation method of administrative punishment document based on graph convolution neural network |
| CN115858725A (en) * | 2022-11-22 | 2023-03-28 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Method and system for screening text noise based on unsupervised graph neural network |
| CN115858725B (en) * | 2022-11-22 | 2023-07-04 | 广西壮族自治区通信产业服务有限公司技术服务分公司 | Text noise screening method and system based on unsupervised graph neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110889282B (en) | 2023-03-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110889282B (en) | Text emotion analysis method based on deep learning | |
| CN107679580B (en) | Heterogeneous migration image emotion polarity analysis method based on multi-mode depth potential correlation | |
| CN109271537B (en) | Text-to-image generation method and system based on distillation learning | |
| CN107025284A (en) | The recognition methods of network comment text emotion tendency and convolutional neural networks model | |
| CN106776713A (en) | It is a kind of based on this clustering method of the Massive short documents of term vector semantic analysis | |
| Ju et al. | An efficient method for document categorization based on word2vec and latent semantic analysis | |
| CN110175221B (en) | Junk short message identification method by combining word vector with machine learning | |
| CN101561805A (en) | Document classifier generation method and system | |
| CN108733647B (en) | Word vector generation method based on Gaussian distribution | |
| Gottapu et al. | Entity resolution using convolutional neural network | |
| CN108038106B (en) | Fine-grained domain term self-learning method based on context semantics | |
| WO2013118435A1 (en) | Semantic similarity level computation method, system and program | |
| CN109992784B (en) | Heterogeneous network construction and distance measurement method fusing multi-mode information | |
| CN107391565B (en) | Matching method of cross-language hierarchical classification system based on topic model | |
| CN107688870A (en) | A kind of the classification factor visual analysis method and device of the deep neural network based on text flow input | |
| CN111859961A (en) | Text keyword extraction method based on improved TopicRank algorithm | |
| CN111191031A (en) | Entity relation classification method of unstructured text based on WordNet and IDF | |
| Ramasundaram et al. | Text categorization by backpropagation network | |
| Han et al. | Vertical domain text classification: towards understanding IT tickets using deep Neural Networks | |
| CN108549718A (en) | A kind of general theme incorporation model joint training method | |
| CN114564563A (en) | End-to-end entity relationship joint extraction method and system based on relationship decomposition | |
| Nabil et al. | Cufe at semeval-2016 task 4: A gated recurrent model for sentiment classification | |
| Parvathi et al. | Identifying relevant text from text document using deep learning | |
| CN107301426A (en) | A kind of multi-tag clustering method of shoe sole print image | |
| CN112231476B (en) | Improved graphic neural network scientific literature big data classification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |