CN112507723A - News emotion analysis method based on multi-model fusion - Google Patents
News emotion analysis method based on multi-model fusion Download PDFInfo
- Publication number
- CN112507723A CN112507723A CN202011400913.4A CN202011400913A CN112507723A CN 112507723 A CN112507723 A CN 112507723A CN 202011400913 A CN202011400913 A CN 202011400913A CN 112507723 A CN112507723 A CN 112507723A
- Authority
- CN
- China
- Prior art keywords
- news
- emotion
- english
- text
- emotion analysis
- 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
- 230000008451 emotion Effects 0.000 title claims abstract description 86
- 238000004458 analytical method Methods 0.000 title claims abstract description 49
- 230000004927 fusion Effects 0.000 title claims abstract description 17
- 238000002372 labelling Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 230000009193 crawling Effects 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 25
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 230000002457 bidirectional effect Effects 0.000 claims description 16
- 230000002996 emotional effect Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 10
- 238000007499 fusion processing Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000000034 method Methods 0.000 description 13
- 238000012360 testing method Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 241000288105 Grus Species 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- 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 Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a news emotion analysis method based on multi-model fusion, which comprises the following steps: crawling English news related to an official news website to obtain an original news corpus; preprocessing the crawled English news text; establishing an entity knowledge base according to English news text information in a preprocessed news corpus, carrying out emotion tendency labeling on the English news text by using the entity knowledge base, and dividing the English news text labeled as an emotion sentence into three levels for labeling; training an emotion analysis model by using the English news text labeled in the third stage, so that the emotion analysis model has a function of carrying out three-stage classification on the preprocessed English news text; and inputting the English news sentence marked as the emotion sentence into the trained emotion analysis model to obtain a classification result. The invention can save the network training time and improve the efficiency while keeping almost the same accuracy.
Description
Technical Field
The invention belongs to the field of computer artificial intelligence, and particularly relates to a news emotion analysis method based on multi-model fusion.
Background
Emotion classification, also known as emotion tendency analysis, refers to identifying whether the tendency of subjective text is positive or negative, or positive or negative, for a given text, and is the most studied field of emotion analysis. Web text typically has a large amount of subjective text and objective text. The objective text is an objective description of things and has no emotional color and emotional tendency, and the subjective text is the opinion or idea of the author to various things and has emotional tendency such as likes and dislikes of the author. The object of emotion classification is subjective text with emotion tendency, so emotion classification is to perform subjective and objective classification of the text first. The subjective and objective classification of the text mainly takes emotion word identification as a main part, different text characteristic representation methods and classifiers are used for identification and classification, and the subjective and objective classification is carried out on the web text in advance, so that the speed and the accuracy of emotion classification can be improved. In the prior research work of subjective text emotion tendency analysis, the main research thought is divided into an emotion dictionary method based on semantics and a method based on machine learning.
The method based on the emotion dictionary requires manual marking and construction of the emotion dictionary, and the analysis result is positively correlated with the quality of the dictionary and has great limitation; the traditional machine learning method needs manual screening of emotional characteristics, and the workload is huge; meanwhile, emotion analysis research of the current deep learning method mostly aims at improving the classification accuracy rate, and often ignores the training rate of the network.
Disclosure of Invention
The invention aims to provide a news emotion analysis method based on multi-model fusion.
The technical scheme for realizing the purpose of the invention is as follows: a news emotion analysis method based on multi-model fusion comprises the following steps:
step 1: crawling English news related to official news websites by using a crawler technology facing the events to obtain an original news corpus;
step 2: preprocessing the crawled English news text;
and step 3: establishing an entity knowledge base facing to countries, characters, organizations and events according to English news text information in the preprocessed news corpus, carrying out emotion tendency labeling on the preprocessed English news text by utilizing the established entity knowledge base, and dividing the English news text labeled as emotion sentences into three levels for labeling;
and 4, step 4: training an emotion analysis model by using the English news text labeled in the third stage, so that the emotion analysis model has a function of carrying out three-stage classification on the preprocessed English news text;
and 5: preprocessing and emotion tendency labeling are carried out on the crawled English news text according to the steps 2 and 3, and the English news sentence labeled as the emotion sentence is input into the trained emotion analysis model to obtain a classification result.
Preferably, preprocessing the crawled English news text comprises performing sentence segmentation, stop word removal and standardization on the crawled news corpus.
Preferably, the entity knowledge base stores characters, English formal names of organizations, alternative names, country names and hot events.
Preferably, the principle of labeling emotion tendencies of the preprocessed English news text by using the established entity knowledge base is as follows:
when n knowledge base entities appear in a sentence of news, the news is marked as an emotional sentence, and n is an adjustable parameter.
Preferably, the emotion analysis model (CNN-BiGRU) comprises a word embedding layer, a Dropout layer, a convolutional neural network, a pooling layer, a bidirectional gated cyclic unit, and an output layer, wherein the word embedding layer is used for converting an input sentence into a vector; the Dropout layer is arranged behind the word embedding layer; the convolutional neural network is used for performing convolution operation on the word vectors output by the Dropout layer to obtain local characteristics among words; the pooling layer is used for pooling the convolved features; the bidirectional gating circulation unit comprises a forward GRU unit and a backward GRU unit, and the output layer is used for inputting the feature vectors into the classifier after full connection to obtain a classification result.
Preferably, the local features obtained by the convolution operation of the convolutional neural network are:
ci=f(w·xi:i+h-1+b)
where b represents an offset, f (-) represents a nonlinear convolution kernel, and xi:i+h-1Representing the i to i + h-1 th rows of the generated vector matrix, and w representing the weight matrix.
Preferably, the GRU unit is used for calculating a text feature vector, and the specific calculation formula is:
zt=σ(wz·[ht-1,xt])
rt=σ(wr·[ht-1,xt])
wherein x istIndicates the input at time t, ht-1Represents the output of the GRU unit cell at time t-1, w represents the weight matrix, ztIs the gating of control updates, rtIs a gate that controls the reset of the reset,represents a candidate hidden state, htRepresenting the final output text feature vector.
Preferably, the sentence characteristic f output to the convolutional neural network by using the Concatenate modecSentence characteristic f output by bidirectional gate control circulation unitgAnd inputting and outputting the layer after fusion processing, wherein the characteristics after fusion processing are as follows:
preferably, the output of the classifier is:
wherein the content of the first and second substances,is a matrix of the weights that is,is the deviation of the weight, and,is the probability of each class.
Compared with the prior art, the invention has the following remarkable advantages: firstly, the static local features of the text are preliminarily extracted by using CNN, the sequence features and the context semantic information of the text are further extracted by a bidirectional gating circulation unit (GRU), and then two layers of unidirectional GRUs are connected, so that the features do not need to be extracted manually in the whole process; compared with a long-time memory network (LSTM), the gating cycle unit (GRU) adopted by the invention is one less gating unit, only a reset gate and an update gate are needed, the parameters are less, the network training speed is higher, the network training time is saved while almost the same accuracy is maintained, and the efficiency is improved.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a flowchart of emotion analysis model training.
FIG. 3 is a diagram of a CNN-BiGRU structure.
FIG. 4 is a flowchart of emotion analysis model testing.
The specific implementation mode is as follows:
a news emotion analysis method based on multi-model fusion constructs emotion analysis models according to emotion corpora facing characters and organizations, can perform emotion sentence recognition and emotion three-level judgment on crawled English news, and comprises the following steps as shown in figure 1:
step 1: and crawling an official news website such as a world wide web and the like to crawl relevant English news by using a crawler technology for the event to obtain an original news corpus.
In some embodiments, the first step in processing the news text is to obtain the news text data, and there are two main methods: the existing corpus which is finished completely by others is directly used, or the corpus data required by others is acquired from the network through tools such as a crawler and the like. The invention is directed to domain specific news emotions, and the second method is more suitable when linguistic data under some specified topics are needed.
Step 2: preprocessing the crawled large-section English news text, including sentence segmentation, stop word removal, standardization and the like on the crawled news corpus. First, in! Is there a … shows the segmentation flag to make sentence break for English news text. Secondly, the English text has some invalid words, such as some short words of "a" and "for", punctuation marks, etc., when performing text analysis, these contents often have little meaning, regard them as stop words, and establish a stop word list to complete the removal of the stop words, which is beneficial to more efficient text analysis. In addition, sometimes words are not in the correct format. For example, "I looooveeee you" shall be "I love you". It is standardized using simple rules and regular expressions.
And step 3: establishing an entity knowledge base facing to countries, characters, organizations and events according to information in the preprocessed news corpus, thereby realizing automatic emotion tendency labeling of news segmented into sentences by utilizing the established entity knowledge base, and then manually labeling news corpora which are automatically labeled as emotion sentences, wherein the method comprises the following specific steps:
step 3.1: establishing an entity knowledge base facing to countries, organizations, characters and events according to information in the preprocessed news corpus, wherein related entities such as English formal names, alternative names, country names, hot events and the like of the characters and the organizations are stored in the entity knowledge base;
step 3.2: and completing automatic labeling of a sentence of news based on the entity knowledge base. The automatic labeling rule is: and when n knowledge base entities appear in a sentence of news, marking the news as an emotional sentence, otherwise, the sentence of news is not the emotional sentence, and n is an adjustable parameter. The traditional corpus construction and labeling method is usually based on manual labeling, and both time cost and labor cost are high. In this way, a large amount of emotional sentences and non-emotional sentence corpus can be obtained relatively easily.
Step 3.3: and (3) marking the emotion sentence corpus on the basis of the step 3.2, and dividing the emotion sentence corpus into three levels (1-negative, 2-neutral and 3-positive). The method combines automatic labeling and manual labeling, finishes the labeling of whether the news segmented into sentences is emotion sentences through a machine, and carries out emotion three-level labeling through a manual mode on the basis of the labeling, thereby greatly reducing the labor cost of emotion labeling of a corpus.
And 4, step 4: training an emotion analysis model by using the English news text labeled in the third stage, so that the emotion analysis model has a function of carrying out three-stage classification on the preprocessed English news text;
in some embodiments, a multi-model fusion method is adopted, a Convolutional Neural Network (CNN) and a bidirectional gated cyclic unit (GRU) are combined to form a deep learning model (CNN-BiGRU), feature information of a text is obtained through the Convolutional Neural Network (CNN) and the bidirectional gated cyclic unit (GRU), and finally a Softmax classifier is used for emotion classification.
And (3) obtaining a data set of emotion sentence three-level classification according to the step 3.3, dividing the data set into a training set and a testing set according to a twenty-eight principle, and training an emotion analysis model by using the training set, wherein the model training process is shown in fig. 2.
The emotion analysis model (CNN-BiGRU) comprises a word embedding layer, a Dropout layer, a convolutional neural network, a pooling layer, a bidirectional gating circulation unit and an output layer.
The word embedding layer is used for converting an input sentence into a vector, and in some embodiments, the most common word2vec algorithm is selected to map each English word into an m-dimensional real number vector. The final result of inputting a sentence is a real number matrix representing the sentence, each row represents a word vector (default dim ═ m), and the row number (default dim ═ n) represents the number of english words in the sentence. Therefore, the input text is subjected to word embedding to obtain n x m dimensional vectors;
the Dropout layer is placed after the word embedding layer to mitigate overfitting during training, and in some embodiments the Dropout layer size is 0.3.
The convolutional neural network represents S by the word vector of the sentence randomly obtained at the dropout layer1,S2,Si,Sn(SiRepresenting the ith word in the sentence), the features of the sentence can be convolved to find local features between words in the input sample. The local features resulting from the convolutional neural network convolution operation can be expressed as:
ci=f(w·xi:i+h-1+b)
where b represents an offset, f (-) represents a nonlinear convolution kernel, and xi:i+h-1Representing the i to i + h-1 th rows of the generated vector matrix, and W represents a weight matrix. From this, one can obtain a feature representation C:
C=[c1,c2,c3…cn-h+1]
the pooling layer is used for pooling the convolved features to extract more remarkable features, maximum pooling (Max boosting) is selected for feature selection, and the final text vector feature generated is fcAnd serves as the input for the next layer.
The bidirectional gating circulation unit is a bidirectional structure adopted on a basic GRU structure, each input low-dimensional word vector is transmitted into a forward GRU unit and a backward GRU unit, and then a text feature vector h output by the forward GRU unit and the backward GRU unittCombining by concat to obtain output f of bidirectional GRUg. At time t, the GRU unit body calculates the text feature vector in the following way:
zt=σ(wz·[ht-1,xt])
rt=σ(wr·[ht-1,xt])
wherein xtIndicates the input at time t, ht-1Represents the output of the GRU unit cell at time t-1, w represents the weight matrix, ztIs the gating of control updates, rtIs a gate that controls the reset of the reset,represents a candidate hidden state, htRepresenting the hidden state of the last output, i.e. the text feature vector of the last output.
By output h of the last momentt-1And input x of the current timetTo obtain two gating states. Updating gating ztDeciding which information to discard and which to add, resetting the gating rtThe extent to which information is discarded is determined. σ is a softmax function by which data can be transformed into a value in the range of 0-1 to act as a gating signal.
After the gating signal is obtained, reset gating is first used to obtain the data r after "resett*ht-1And is then combined with xtAnd splicing, and then scaling the data to the range of-1 to 1 through a tanh activation function. I.e. to obtain candidate hidden state values
Last ztControl h required from the last momentt-1How much information is forgotten and how much current time needs to be addedFinally obtaining htAnd directly obtaining the finally output hidden state.
Inputting each sentence into an emotion analysis model, respectively passing through a convolutional neural network and a bidirectional gate control circulation unit, and outputting a sentence characteristic f through the convolutional neural networkcThe sentence characteristic output by the bidirectional gate control circulation unit is fgAnd performing fusion processing on the characteristics of the sentences by using a concatemate mode to obtain a characteristic vector S', which is expressed as:
the output layer is to input the spliced feature vectors into a classifier after full connection, and because the invention researches three categories of emotions, a softmax function is selected as an output classifier, and the output result is mapped to a value of (0, 1) and expressed as:
wherein the content of the first and second substances,is a matrix of the weights that is,is the deviation of the weight, and,is the output of the last layer, the output is the probability of each class. The probability under different categories is obtained through a softmax classifier, and the emotional tendency of the news sentence is analyzed to be positive, or neutral, or negative.
And (4) testing an emotion analysis model. In the emotion analysis algorithm model test stage, a classification result is obtained by the test set obtained in the step 4 according to the twenty-eight principle through the trained model, and accuracy and performance analysis are performed. The flow of the model test is shown in fig. 4.
And 5: preprocessing and emotion tendency labeling are carried out on the crawled English news text according to the steps 2 and 3, and the English news sentence labeled as the emotion sentence is input into the trained emotion analysis model to obtain a classification result.
The method analyzes the emotion levels (positive, negative and neutral) of English news based on the multi-model fusion deep learning model. Aiming at the challenges of public sentiment analysis and fine-grained sentiment analysis in a specific field, automatic sentiment tendency labeling of news divided into sentences based on an entity knowledge base is designed, and sentiment three-level labeling is carried out in a manual mode on the basis of the automatic sentiment tendency labeling, so that the labor cost of sentiment labeling of a corpus is greatly reduced.
The invention designs an emotion classification model of CNN-BiGRU by combining a bidirectional gate control circulation unit neural network on the basis of a convolutional neural network. The convolutional neural network has the property of automatically extracting and learning abstract features in sentences, so that the trouble of manually extracting the features is overcome. The method has the advantages of unique memory and selection characteristics of a unique processing sequence problem of a gated cyclic unit (GRU) based on long-time memory network evolution, and good performance in processing word serialization problems.
Example 1
The present embodiment assumes that a piece of news is crawled from the world news web: "Lam saidthat short full supported Yeung's state and transformed hi actions in evaluating with the textual books. Global Times 57% of responses in the US disparity of word ways relationships with China, a receiver Gallup summary, carried out July 30to Aug 12, shared. Preprocessing the crawled English news comprises the steps of carrying out sentence segmentation, data cleaning, denoising and stop word removal on crawled news corpus, removing words irrelevant to themes and the like, wherein the result after processing is as follows:
1)Lam support Yeung statementpraise action deal textbooks
2)Global Times 57%respondents US disapprove Trump handle relations China Gallup survey carry July 30Aug 12show
the preprocessed news sentences are automatically labeled through an entity knowledge base to obtain sentences of which the news sentences are all emotional tendencies. And then, carrying out emotion three-level classification by adopting the trained emotion analysis model, and judging whether the emotional tendency of the two sentences of news is positive, negative or neutral. In the above example, it can be obtained that the polarity of the emotion sentence Lam support Yeung statement action default texts is positive; the polarity of the emotion sentence Global Times 57% responses US dispoprovided Trump handle relationships China Gallup surveiy study July 30Aug 12show is negative.
Claims (9)
1. A news emotion analysis method based on multi-model fusion is characterized by comprising the following steps:
step 1: crawling English news related to official news websites by using a crawler technology facing the events to obtain an original news corpus;
step 2: preprocessing the crawled English news text;
and step 3: establishing an entity knowledge base facing to countries, characters, organizations and events according to English news text information in the preprocessed news corpus, carrying out emotion tendency labeling on the preprocessed English news text by utilizing the established entity knowledge base, and dividing the English news text labeled as emotion sentences into three levels for labeling;
and 4, step 4: training an emotion analysis model by using the English news text labeled in the third stage, so that the emotion analysis model has a function of carrying out three-stage classification on the preprocessed English news text;
and 5: preprocessing and emotion tendency labeling are carried out on the crawled English news text according to the steps 2 and 3, and the English news sentence labeled as the emotion sentence is input into the trained emotion analysis model to obtain a classification result.
2. The news emotion analysis method based on multi-model fusion as claimed in claim 1, wherein the preprocessing of the crawled English news text comprises sentence segmentation, word stop and standardization of the crawled news corpus.
3. The news emotion analysis method based on multi-model fusion as claimed in claim 1, wherein characters, English formal names of organizations, alternative names, country names, and hot events are stored in the entity knowledge base.
4. The news emotion analysis method based on multi-model fusion as claimed in claim 1, wherein the principle of labeling emotion tendencies of the preprocessed English news text by using the established entity knowledge base is as follows:
when n knowledge base entities appear in a sentence of news, the news is marked as an emotional sentence, and n is an adjustable parameter.
5. The news emotion analysis method based on multi-model fusion as claimed in claim 1, wherein the emotion analysis model (CNN-BiGRU) includes a word embedding layer for converting an input sentence into a vector, a Dropout layer, a convolutional neural network, a pooling layer, a bidirectional gated round-robin unit, an output layer; the Dropout layer is arranged behind the word embedding layer; the convolutional neural network is used for performing convolution operation on the word vectors output by the Dropout layer to obtain local characteristics among words; the pooling layer is used for pooling the convolved features; the bidirectional gating circulation unit comprises a forward GRU unit and a backward GRU unit, and the output layer is used for inputting the feature vectors into the classifier after full connection to obtain a classification result.
6. The news emotion analysis method based on multi-model fusion as claimed in claim 5, wherein the local features obtained by the convolution operation of the convolutional neural network are:
ci=f(w·xi:i+h-1+b)
where b represents an offset, f (-) represents a nonlinear convolution kernel, and xi:i+h-1Representing the i to i + h-1 th rows of the generated vector matrix, and w representing the weight matrix.
7. The news emotion analysis method based on multi-model fusion as claimed in claim 5, wherein the GRU unit is used for calculating text feature vectors, and the specific calculation formula is as follows:
zt=σ(wz·[ht-1,xt])
rt=σ(wr·[ht-1,xt])
wherein x istIndicates the input at time t, ht-1Represents the output of the GRU unit cell at time t-1, w represents the weight matrix, ztIs the gating of control updates, rtIs a gate that controls the reset of the reset,represents a candidate hidden state, htRepresenting the final output text feature vector.
8. The news emotion analysis method based on multi-model fusion as claimed in claim 5, wherein the sentence feature f outputted to the convolutional neural network is outputted in a Concatenate mannercSentence characteristic f output by bidirectional gate control circulation unitgAnd inputting and outputting the layer after fusion processing, wherein the characteristics after fusion processing are as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011400913.4A CN112507723B (en) | 2020-12-03 | 2020-12-03 | News emotion analysis method based on multi-model fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011400913.4A CN112507723B (en) | 2020-12-03 | 2020-12-03 | News emotion analysis method based on multi-model fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112507723A true CN112507723A (en) | 2021-03-16 |
CN112507723B CN112507723B (en) | 2024-06-18 |
Family
ID=74969790
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011400913.4A Active CN112507723B (en) | 2020-12-03 | 2020-12-03 | News emotion analysis method based on multi-model fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112507723B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116450840A (en) * | 2023-03-22 | 2023-07-18 | 武汉理工大学 | Deep learning-based field emotion dictionary construction method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472052A (en) * | 2019-07-31 | 2019-11-19 | 西安理工大学 | A kind of Chinese social platform sentiment analysis method based on deep learning |
CN110807320A (en) * | 2019-11-11 | 2020-02-18 | 北京工商大学 | Short text emotion analysis method based on CNN bidirectional GRU attention mechanism |
CN111027329A (en) * | 2019-11-06 | 2020-04-17 | 上海大学 | User emotion analysis system and method of logistics intelligent customer service system based on deep learning |
CN111310476A (en) * | 2020-02-21 | 2020-06-19 | 山东大学 | Public opinion monitoring method and system using aspect-based emotion analysis method |
CN111368086A (en) * | 2020-03-17 | 2020-07-03 | 昆明理工大学 | CNN-BilSTM + attribute model-based sentiment classification method for case-involved news viewpoint sentences |
KR20200127555A (en) * | 2019-05-03 | 2020-11-11 | 주식회사 자이냅스 | A program for an automatic sentiment information labeling to news articles for providing sentiment information |
-
2020
- 2020-12-03 CN CN202011400913.4A patent/CN112507723B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20200127555A (en) * | 2019-05-03 | 2020-11-11 | 주식회사 자이냅스 | A program for an automatic sentiment information labeling to news articles for providing sentiment information |
CN110472052A (en) * | 2019-07-31 | 2019-11-19 | 西安理工大学 | A kind of Chinese social platform sentiment analysis method based on deep learning |
CN111027329A (en) * | 2019-11-06 | 2020-04-17 | 上海大学 | User emotion analysis system and method of logistics intelligent customer service system based on deep learning |
CN110807320A (en) * | 2019-11-11 | 2020-02-18 | 北京工商大学 | Short text emotion analysis method based on CNN bidirectional GRU attention mechanism |
CN111310476A (en) * | 2020-02-21 | 2020-06-19 | 山东大学 | Public opinion monitoring method and system using aspect-based emotion analysis method |
CN111368086A (en) * | 2020-03-17 | 2020-07-03 | 昆明理工大学 | CNN-BilSTM + attribute model-based sentiment classification method for case-involved news viewpoint sentences |
Non-Patent Citations (1)
Title |
---|
崔昕阳;龙华;熊新;邵玉斌;杜庆治;: "基于并行双向门控循环单元与自注意力机制的中文文本情感分类", 北京化工大学学报(自然科学版), no. 02, 20 March 2020 (2020-03-20) * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116450840A (en) * | 2023-03-22 | 2023-07-18 | 武汉理工大学 | Deep learning-based field emotion dictionary construction method |
Also Published As
Publication number | Publication date |
---|---|
CN112507723B (en) | 2024-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108363743B (en) | Intelligent problem generation method and device and computer readable storage medium | |
CN109543084B (en) | Method for establishing detection model of hidden sensitive text facing network social media | |
CN107944559B (en) | Method and system for automatically identifying entity relationship | |
CN108804612B (en) | Text emotion classification method based on dual neural network model | |
CN106569998A (en) | Text named entity recognition method based on Bi-LSTM, CNN and CRF | |
CN107247702A (en) | A kind of text emotion analysis and processing method and system | |
CN110502753A (en) | A kind of deep learning sentiment analysis model and its analysis method based on semantically enhancement | |
CN110188781B (en) | Ancient poetry automatic identification method based on deep learning | |
CN110287323B (en) | Target-oriented emotion classification method | |
CN110222178A (en) | Text sentiment classification method, device, electronic equipment and readable storage medium storing program for executing | |
CN111985247B (en) | Microblog user interest identification method and system based on multi-granularity text feature representation | |
CN110807320A (en) | Short text emotion analysis method based on CNN bidirectional GRU attention mechanism | |
CN108647225A (en) | A kind of electric business grey black production public sentiment automatic mining method and system | |
CN111522908A (en) | Multi-label text classification method based on BiGRU and attention mechanism | |
CN111078833A (en) | Text classification method based on neural network | |
CN106682089A (en) | RNNs-based method for automatic safety checking of short message | |
CN114756681B (en) | Evaluation and education text fine granularity suggestion mining method based on multi-attention fusion | |
CN110489554B (en) | Attribute-level emotion classification method based on location-aware mutual attention network model | |
CN112163091A (en) | CNN-based aspect-level cross-domain emotion analysis method | |
CN115952292B (en) | Multi-label classification method, apparatus and computer readable medium | |
CN111581364B (en) | Chinese intelligent question-answer short text similarity calculation method oriented to medical field | |
CN110297986A (en) | A kind of Sentiment orientation analysis method of hot microblog topic | |
CN111339772B (en) | Russian text emotion analysis method, electronic device and storage medium | |
CN112100212A (en) | Case scenario extraction method based on machine learning and rule matching | |
CN107818173B (en) | Vector space model-based Chinese false comment filtering 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 |