CN108470061A - A kind of emotional semantic classification system for visual angle grade text - Google Patents
A kind of emotional semantic classification system for visual angle grade text Download PDFInfo
- Publication number
- CN108470061A CN108470061A CN201810250039.7A CN201810250039A CN108470061A CN 108470061 A CN108470061 A CN 108470061A CN 201810250039 A CN201810250039 A CN 201810250039A CN 108470061 A CN108470061 A CN 108470061A
- Authority
- CN
- China
- Prior art keywords
- text
- visual angle
- vector
- classification
- attention
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of emotional semantic classification systems for visual angle grade text.Including text data preprocessing module, for obtaining term vector dictionary；Feature extraction module, convolution memory network model extract the relevant feature in visual angle, and constantly update parameter according to error for building the relevant attention in visual angle；Classification results output module obtains the final emotional semantic classification result of text with classification function.The present invention can preferably carry out the visual angle grade analysis of viewpoint classification and versatile, can carry out sentiment analysis from the angle of data consumer.
Description
Technical field
The present invention relates to sentiment analysis and opining mining field, more particularly, to a kind of feelings for visual angle grade text
Feel categorizing system, can preferably carry out the visual angle grade analysis of viewpoint classification and versatile, can stand data consumer's
Angle carries out sentiment analysis.
Background technology
Currently, many technical methods can be used for text emotion classification.Traditional file classification method is merely from text point
The angle of class is set out, it is difficult to determine the difference of emotional category under different visual angles in text.It is different from traditional file classification method, such as
What classify for the text emotion at visual angle using the information of different visual angles in text, is asking for visual angle grade text emotion classification
Topic is crucial.The method that traditional research work is mainly based upon the machine learning of supervision, this method are related by building visual angle
Feature, use support vector machines（SVMs）Grader carries out emotional semantic classification, improves emotional semantic classification result.This one kind is based on spy
Although sign engineering and the method for shallow-layer linear model achieve certain effect, but need a large amount of Feature Engineering, take consumption
Power, and it is limited to the rule and the validity of feature and the learning ability of model of engineer, so its performance need
It is promoted.
Currently, with the fast development of deep learning, the expression learning model based on multilayer neural network is in semantic expressiveness
Advantage is had more with sentiment analysis utilization aspect.Text emotion of many researchers also by these models for visual angle grade is classified.
Neural network model can learn text representation, to avoid from data automatically compared with the machine learning method for having supervision
A large amount of Feature Engineering also has a better expansion between capture visual angle and context in terms of complicated semantic relation.But
Traditional neural network textual classification model is only simply used for model learning using visual angle as one of feature, does not examine fully
Consider the relationship between visual angle and context, each word cannot be weighed in the context at given visual angle to the language of visual angle grade emotional semantic classification
Adopted percentage contribution.Memory network model is widely used in natural language processing, can preferably be carried out to visual angle and context
Modeling, obtains the semantic distribution situation of context, and compared with Recognition with Recurrent Neural Network model, and this method is simpler, calculates
Speed is faster.But word order information is had ignored, is independent from each other between the word in context.Convolutional neural networks have study office
The ability of portion's feature can construct phrase feature, the contact between close context by convolutional layer.
Currently, grade sensibility classification method in visual angle underuses the Viewing-angle information in text, can not effectively excavate latent
Viewpoint information.In recent years, grade emotional semantic classification in visual angle has obtained the great attention of domestic and international many scholars and research institution, quilt
International semantic evaluation and test is classified as one of evaluation and test task, is the race of big data and computational intelligence contest that Chinese Computer Federation holds
One of and the new research focus of Chinese text tendentiousness evaluation and test meeting.In the grade text emotion classified use text of visual angle
The information of different visual angles classify for the text emotion at visual angle, can be carried out from the angle of data consumer specific
Sentiment analysis provides more fine-grained information, effectively improves the order of accuarcy of the analysis result of sentiment analysis system offer, has
Help study and judge true viewpoint and view that personnel become more apparent upon people to the various entities such as much-talked-about topic, tissue, product, to study and judge
Personnel provide more efficient and accurate information.This just proposes a challenge to visual angle grade text emotion sorting technique：How
The effective visual angle grade text emotion of structure one classifies prototype system to meet its needs.Therefore, people are highly desirable one kind
The visual angle grade text sentiment classification method of precise and high efficiency, this method can extract validity feature automatically, and be taken out to feature
As and combination, finally identify the different viewpoints classification under different visual angles in text.
Invention content
The purpose of the present invention is to provide a kind of emotional semantic classification systems for visual angle grade text, can extract automatically
Feature is imitated, and feature is abstracted and is combined, finally identifies the viewpoint classification for giving visual angle in text.
To achieve the above object, the technical scheme is that：A kind of emotional semantic classification system for visual angle grade text, packet
It includes：
One data preprocessing module obtains text term vector dictionary for segmenting text；
It is relevant to extract visual angle by the relevant attention in convolution memory network model construction visual angle for one feature extraction module
Feature, and parameter is constantly updated according to error；
One classification results output module, the text vector obtained according to feature extraction module obtain final sight with classification function
Point analysis result.
In an embodiment of the present invention, in the data preprocessing module, the acquisition of term vector dictionary is by means of increasing income
The Glove tools term vector that training obtains in big expectation in advance.
In an embodiment of the present invention, the feature extraction module includes convolutional layer and attention layer, to realize visual angle phase
The feature extraction of pass and attention structure.
In an embodiment of the present invention, the convolutional layer carries out convolution operation using convolution kernel, considers contextual information, will
Text message in specified window is handled and is mapped, and realizes the abstract of feature；Then, consider visual angle factor to text feelings
Visual angle vector sum text term vector is spliced in the influence for feeling classification.
In an embodiment of the present invention, the attention layer considers the semantic distribution of word order information and text, by attention
Three output vectors are obtained to get to the output vector of convolution memory network with text term vector weighted sum.
In an embodiment of the present invention, the attention layer further includes softmax layers, at obtained text term vector
Reason obtains the attention distribution of each word in text.
In an embodiment of the present invention, the system also includes several linear transformation layers, are carried out to visual angle vector linear
After transformation, after summation is added with the output vector of convolution memory network, cascade operation is carried out, feature is further abstracted and is obtained
Final text vector.
In an embodiment of the present invention, the classification results output module uses the obtained text of softmax function pairs
This Vector Processing predicts the viewpoint classification of each text.
In an embodiment of the present invention, in the training stage of model, weight matrix is all parameter, is passed according to the forward direction of information
It broadcasts the back-propagating with error will constantly be adjusted them, successive optimization object function.
Compared to the prior art, the invention has the advantages that：The present invention can extract validity feature automatically, and
Feature is abstracted and is combined, finally identifies the viewpoint classification for giving visual angle in text.
Description of the drawings
Fig. 1 is the schematic configuration view of one embodiment of the invention viewpoint analysis prototype system used by Chinese microblogging.
Specific implementation mode
Below in conjunction with the accompanying drawings, technical scheme of the present invention is specifically described.
A kind of emotional semantic classification system for visual angle grade text of the present invention, including：
One data preprocessing module obtains text term vector dictionary for segmenting text；
It is relevant to extract visual angle by the relevant attention in convolution memory network model construction visual angle for one feature extraction module
Feature, and parameter is constantly updated according to error；
One classification results output module, the text vector obtained according to feature extraction module obtain final sight with classification function
Point analysis result.
In the data preprocessing module, the acquisition of term vector dictionary is by means of the Glove tools increased income in advance big pre-
The term vector that training obtains in material.
The feature extraction module includes convolutional layer and attention layer, to realize the relevant feature extraction in visual angle and attention
Structure.The convolutional layer using convolution kernel carry out convolution operation, consider contextual information, by the text message in specified window into
Row processing and mapping, realize the abstract of feature；Then, consider the influence classified to text emotion of visual angle factor, by visual angle to
Amount and the splicing of text term vector.The attention layer considers the semantic distribution of word order information and text, by attention and text word
Vectorial weighted sum obtains three output vectors to get to the output vector of convolution memory network.The attention layer further includes
Softmax layers, to the processing of obtained text term vector, obtain the attention distribution of each word in text.
The system also includes several linear transformation layers, after carrying out linear transformation to visual angle vector, remember net with convolution
After the output vector of network is added summation, cascade operation is carried out, feature is further abstracted and obtains final text vector.
The classification results output module predicts each text using the obtained text vector processing of softmax function pairs
This viewpoint classification.
In the training stage of model, weight matrix is all parameter, according to the back-propagating of the propagated forward of information and error
Constantly they will be adjusted, successive optimization object function.
It is the specific implementation process of the present invention below.
As shown in Figure 1, the embodiment of the present invention includes in visual angle grade text emotion classification prototype system：Data preprocessing module
1, for obtaining term vector dictionary；Feature extraction module 2, convolution memory network model are used to build the relevant attention in visual angle,
The relevant feature in visual angle is extracted, and parameter is constantly updated according to error；Classification results output module 3, is obtained with classification function
The final emotional semantic classification result of text.The configuration of each module is described in detail separately below.
1）Data preprocessing module 1
First, how description data preprocessing module 1 will and obtain term vector dictionary.
Because the input data of neural network is usually vector, so as to the end-to-end training of model, it is therefore desirable to right first
Text data carries out vectorization expression.For the ease of the processing and analysis of data, in the data preprocessing module of the present invention, we
Participle operation is realized according to data set, and does not filter stop words.It, can be by text data from text by tabling look-up after pretreatment
Form is converted into vector form.
3）Feature extraction module 2
It is described below feature extraction module 2 is how the data for obtaining a upper module carry out feature extraction.This module is by rolling up
Two part of lamination and attention layer form.The core of convolutional layer is to carry out convolution operation using convolution kernel, is remembered each single in m
After column vector and the splicing of visual angle vector corresponding to word, the convolution kernel that sliding window size is 3 rolls up the vector in window
Product operation calculates the semantic information score of each word window in context.After process of convolution, highlight a part of semantic special
Sign, obtains semantic information score set, in order to enable this score can be used for the weighting of text semantic, needs to carry out base to it
In the normalization of softmax functions.Normalized result is each word window significance level score in context, as attention
Weight is used for the weighting of text semantic, and the corresponding each column vector of each word window for remembering m is weighted summation, and by line
Property variation after visual angle vector sum to obtain three output vectors of memory network and carry out cascade so that output vector is sequence-level
Other text representation.
4）Classification results output module 3
Finally, interpretive classification result output module 3.Module 2）Output vector obtained final text vector, classification results
Output module calculates gained vector using softmax classification functions one by one, according to the threshold value of setting obtain the text about to
Determine the emotional category predicted value at visual angle.In the training stage, need predicted value and desired value seeking error, and using under stochastic gradient
Drop method and back-propagating are iterated update to the parameter of whole system；Otherwise, obtained predicted value need to only be exported.
The above are preferred embodiments of the present invention, all any changes made according to the technical solution of the present invention, and generated function is made
When with range without departing from technical solution of the present invention, all belong to the scope of protection of the present invention.
Claims (9)
1. a kind of emotional semantic classification system for visual angle grade text, which is characterized in that including：
One data preprocessing module obtains text term vector dictionary for segmenting text；
It is relevant to extract visual angle by the relevant attention in convolution memory network model construction visual angle for one feature extraction module
Feature, and parameter is constantly updated according to error；
One classification results output module, the text vector obtained according to feature extraction module obtain final sight with classification function
Point analysis result.
2. system according to claim 1, which is characterized in that in the data preprocessing module, term vector dictionary obtains
It obtains by means of the Glove tools the increased income term vector that training obtains in big expectation in advance.
3. system according to claim 1, which is characterized in that the feature extraction module includes convolutional layer and attention
Layer, to realize that the relevant feature extraction in visual angle and attention are built.
4. system according to claim 3, which is characterized in that the convolutional layer carries out convolution operation using convolution kernel, examines
Consider contextual information, the text message in specified window is handled and mapped, realizes the abstract of feature；Then, consider
Visual angle vector sum text term vector is spliced in the influence that visual angle factor classifies to text emotion.
5. system according to claim 3, which is characterized in that the attention layer considers the language of word order information and text
Attention and text term vector weighted sum are obtained three output vectors to get to the output of convolution memory network by justice distribution
Vector.
6. system according to claim 5, which is characterized in that the attention layer further includes softmax layers, to acquired
Text term vector processing, obtain each word in text attention be distributed.
7. system according to claim 5, which is characterized in that the system also includes several linear transformation layers, to regarding
After angular amount carries out linear transformation, after summation is added with the output vector of convolution memory network, progress cascade operation, to feature into
One step abstracts to obtain final text vector.
8. system according to claim 1, which is characterized in that the classification results output module uses softmax letters
It is several that obtained text vector is handled, predict the viewpoint classification of each text.
9. system according to claim 1, which is characterized in that in the training stage of model, weight matrix is all parameter, root
It is believed that the back-propagating of the propagated forward and error of breath will constantly be adjusted them, successive optimization object function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810250039.7A CN108470061A (en) | 2018-03-26 | 2018-03-26 | A kind of emotional semantic classification system for visual angle grade text |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810250039.7A CN108470061A (en) | 2018-03-26 | 2018-03-26 | A kind of emotional semantic classification system for visual angle grade text |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108470061A true CN108470061A (en) | 2018-08-31 |
Family
ID=63265859
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810250039.7A Pending CN108470061A (en) | 2018-03-26 | 2018-03-26 | A kind of emotional semantic classification system for visual angle grade text |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108470061A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109271493A (en) * | 2018-11-26 | 2019-01-25 | 腾讯科技（深圳）有限公司 | A kind of language text processing method, device and storage medium |
| CN109783644A (en) * | 2019-01-18 | 2019-05-21 | 福州大学 | A kind of cross-cutting emotional semantic classification system and method based on text representation study |
| CN110362734A (en) * | 2019-06-24 | 2019-10-22 | 北京百度网讯科技有限公司 | Text recognition method, device, equipment and computer readable storage medium |
| CN110377741A (en) * | 2019-07-22 | 2019-10-25 | 成都深度智慧人工智能科技有限公司 | File classification method, intelligent terminal and computer readable storage medium |
| CN110569355A (en) * | 2019-07-24 | 2019-12-13 | 中国科学院信息工程研究所 | Viewpoint target extraction and target emotion classification combined method and system based on word blocks |
| CN110969011A (en) * | 2018-09-30 | 2020-04-07 | 北京国双科技有限公司 | Text emotion analysis method and device, storage medium and processor |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130218862A1 (en) * | 2009-12-01 | 2013-08-22 | Topsy Labs, Inc. | System and method for customizing analytics based on users media affiliation status |
| US20160085745A1 (en) * | 2014-09-24 | 2016-03-24 | International Business Machines Corporation | Perspective data analysis and management |
| CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
| CN107608961A (en) * | 2017-09-08 | 2018-01-19 | 广州汪汪信息技术有限公司 | Sentiment analysis method, electronic equipment, storage medium, system based on visual angle |
-
2018
- 2018-03-26 CN CN201810250039.7A patent/CN108470061A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130218862A1 (en) * | 2009-12-01 | 2013-08-22 | Topsy Labs, Inc. | System and method for customizing analytics based on users media affiliation status |
| US20160085745A1 (en) * | 2014-09-24 | 2016-03-24 | International Business Machines Corporation | Perspective data analysis and management |
| CN107341506A (en) * | 2017-06-12 | 2017-11-10 | 华南理工大学 | A kind of Image emotional semantic classification method based on the expression of many-sided deep learning |
| CN107608961A (en) * | 2017-09-08 | 2018-01-19 | 广州汪汪信息技术有限公司 | Sentiment analysis method, electronic equipment, storage medium, system based on visual angle |
Non-Patent Citations (4)
| Title |
|---|
| DUYU TANG ET AL.: "Aspect Level Sentiment Classification with Deep Memory Network", 《PROCEEDINGS OF THE 2016 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING》 * |
| PENG CHEN ET AL.: "Recurrent Attention Network on Memory for Aspect Sentiment Analysis", 《PROCEEDINGS OF THE 2017 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING》 * |
| 梁斌等: "基于多注意力卷积神经网络的特定目标情感分析", 《计算机研究与发展》 * |
| 王义真: "基于云环境的短文本情感分析", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110969011A (en) * | 2018-09-30 | 2020-04-07 | 北京国双科技有限公司 | Text emotion analysis method and device, storage medium and processor |
| CN110969011B (en) * | 2018-09-30 | 2023-04-07 | 北京国双科技有限公司 | Text emotion analysis method and device, storage medium and processor |
| CN109271493A (en) * | 2018-11-26 | 2019-01-25 | 腾讯科技（深圳）有限公司 | A kind of language text processing method, device and storage medium |
| CN109271493B (en) * | 2018-11-26 | 2021-10-08 | 腾讯科技（深圳）有限公司 | Language text processing method and device and storage medium |
| CN109783644A (en) * | 2019-01-18 | 2019-05-21 | 福州大学 | A kind of cross-cutting emotional semantic classification system and method based on text representation study |
| CN110362734A (en) * | 2019-06-24 | 2019-10-22 | 北京百度网讯科技有限公司 | Text recognition method, device, equipment and computer readable storage medium |
| CN110377741A (en) * | 2019-07-22 | 2019-10-25 | 成都深度智慧人工智能科技有限公司 | File classification method, intelligent terminal and computer readable storage medium |
| CN110569355A (en) * | 2019-07-24 | 2019-12-13 | 中国科学院信息工程研究所 | Viewpoint target extraction and target emotion classification combined method and system based on word blocks |
| CN110569355B (en) * | 2019-07-24 | 2022-05-03 | 中国科学院信息工程研究所 | Viewpoint target extraction and target emotion classification combined method and system based on word blocks |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108470061A (en) | A kind of emotional semantic classification system for visual angle grade text | |
| Samadi et al. | Deep contextualized text representation and learning for fake news detection | |
| CN109558487A (en) | Document Classification Method based on the more attention networks of hierarchy | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| CN109299268A (en) | A kind of text emotion analysis method based on dual channel model | |
| CN109710769A (en) | A kind of waterborne troops's comment detection system and method based on capsule network | |
| CN109241283A (en) | A kind of file classification method based on multi-angle capsule network | |
| CN110245229A (en) | A kind of deep learning theme sensibility classification method based on data enhancing | |
| CN110222163A (en) | A kind of intelligent answer method and system merging CNN and two-way LSTM | |
| CN104834747A (en) | Short text classification method based on convolution neutral network | |
| CN108038205A (en) | For the viewpoint analysis prototype system of Chinese microblogging | |
| CN102194001A (en) | Internet public opinion crisis early-warning method | |
| Wu et al. | Hierarchical attention-based multimodal fusion for video captioning | |
| CN106980650A (en) | A kind of emotion enhancing word insertion learning method towards Twitter opinion classifications | |
| CN110263822A (en) | A kind of Image emotional semantic analysis method based on multi-task learning mode | |
| CN111914553B (en) | Financial information negative main body judging method based on machine learning | |
| CN106339718A (en) | Classification method based on neural network and classification device thereof | |
| CN109783629A (en) | A kind of micro-blog event rumour detection method of amalgamation of global event relation information | |
| CN110569355A (en) | Viewpoint target extraction and target emotion classification combined method and system based on word blocks | |
| CN108154156A (en) | Image Ensemble classifier method and device based on neural topic model | |
| Stemle et al. | Using language learner data for metaphor detection | |
| CN109670169A (en) | A kind of deep learning sensibility classification method based on feature extraction | |
| CN111767720B (en) | Title generation method, computer and readable storage medium | |
| Lisjana et al. | Classifying complaint reports using rnn and handling imbalanced dataset | |
| CN111737467A (en) | Object-level emotion classification method based on segmented convolutional neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180831 |