CN112989033A - Microblog emotion classification method based on emotion category description - Google Patents
Microblog emotion classification method based on emotion category description Download PDFInfo
- Publication number
- CN112989033A CN112989033A CN202011408229.0A CN202011408229A CN112989033A CN 112989033 A CN112989033 A CN 112989033A CN 202011408229 A CN202011408229 A CN 202011408229A CN 112989033 A CN112989033 A CN 112989033A
- Authority
- CN
- China
- Prior art keywords
- emotion
- category
- microblog
- description
- question
- 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 160
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 230000004927 fusion Effects 0.000 claims abstract description 7
- 230000007935 neutral effect Effects 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 9
- 230000002996 emotional effect Effects 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 4
- 230000036651 mood Effects 0.000 claims description 3
- 238000003058 natural language processing Methods 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 14
- 238000002474 experimental method Methods 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 5
- 230000002457 bidirectional effect Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000008520 organization Effects 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 206010002368 Anger Diseases 0.000 description 1
- 244000097202 Rathbunia alamosensis Species 0.000 description 1
- 235000009776 Rathbunia alamosensis Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000007430 reference method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
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
-
- 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
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a microblog emotion classification method based on emotion category description, and belongs to the technical field of natural language processing. Firstly, proposing an emotion category description strategy, and expanding all emotion categories of microblogs to be classified into formal category descriptions; secondly, splicing the microblog text and the category description into a question-answer pair, and inputting the question-answer pair into a pre-trained BERT model; secondly, inputting a hidden state obtained by coding the question-answer pairs through a BERT model into a two-layer fully-connected neural network, and outputting a fusion semantic representation of the whole question-answer pairs; and finally, inputting the fusion semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
Description
Technical Field
The invention relates to a microblog emotion classification method based on emotion category description, and belongs to the technical field of natural language processing.
Background
Microblog sentiment classification is the basis of public opinion analysis. Unlike general emotion classification, which classifies the emotion of a text as subjective or positive, emotion classification requires recognition of more detailed emotion in the text, such as happiens, angers, sadness, and fear. Traditional supervised emotion classification methods generally convert categories into digital labels as a supervised signal to guide the learning process of the model. For example, a happy emotion is represented by "1" and an anger emotion is represented by "2". Specifically, the digital label will be represented as a one-hot vector for calculating the training loss and then minimizing the objective function by a back propagation algorithm, thereby achieving the purpose of training the model. The method only considers the characteristics of the microblog and does not combine the meanings of the emotion classes for modeling, so that the model does not know what is classified when learning, and the sample cannot be accurately classified into the corresponding class. The microblog emotion classification is a process of understanding whether certain emotion is contained or not by reading microblog content and combining the semantics of emotion categories. Based on the inspiration, the invention provides a microblog emotion classification model based on emotion category description.
Disclosure of Invention
The invention provides a microblog emotion classification method based on emotion category description, which is used for solving the problem that emotion category semantic information is not considered in a microblog emotion classification task in a traditional method.
The technical scheme of the invention is as follows: a microblog emotion classification method based on emotion category description includes the steps that firstly, an emotion category description strategy is proposed, and all emotion categories of microblogs to be classified are expanded into formal category descriptions; secondly, splicing the microblog text and the category description into a question-answer pair, and inputting the question-answer pair into a pre-trained BERT model; secondly, inputting a hidden state obtained by coding the question-answer pairs through a BERT model into a two-layer fully-connected neural network, and outputting a fusion semantic representation of the whole question-answer pairs; and finally, inputting the fused semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
As a further scheme of the invention, the method comprises the following specific steps:
step1, collecting microblog texts as an experimental data set, and carrying out preprocessing operation, wherein the manual label of each piece of data after preprocessing operation is one of a plurality of emotion categories;
step2, three emotion category description strategies are proposed, and a plurality of emotion categories are expanded into formal category descriptions; the three emotion category description strategies are respectively as follows: a category description strategy based on keywords, a category description strategy based on keyword expansion, and a category description strategy based on emotion definition;
step3, constructing a pre-trained BERT emotion classification model based on emotion category description: on the basis of Step1 and Step2, firstly, based on the thought of question answering, a microblog text is taken as a question, category descriptions of all emotion categories are spliced into a candidate answer set, the microblog text and the category descriptions form a question-answer pair together and are taken as input of a pre-training BERT model, and therefore rich context semantics and category semantic information are fused; secondly, inputting the hidden state of the 'CLS' token obtained by coding the question-answer pairs by the BERT model into a two-layer fully-connected neural network, and outputting the fusion semantic representation of the whole question-answer pairs; and finally, inputting the fused semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
Wherein, "[ CLS ]" represents a special classification token, and the last layer hidden state of "[ CLS ]" can be used for representing the semantic meaning of the text to perform a classification task; "[ SEP ]" represents a segmenter token for segmenting the two sentences in the question-answer pair.
As a further scheme of the invention, in Step1, preprocessing comprises removing duplicate, removing "//", "@" and URL data.
As a further aspect of the present invention, in Step1, the artificial label of each piece of data after the preprocessing operation is: happenses, anger, sadness, fear, and neutral are one of 5 mood categories.
As a further scheme of the present invention, in the keyword-based category description strategy mentioned in Step2, the keyword-based category description strategy directly uses chinese keywords corresponding to a plurality of emotion categories as emotion category descriptions.
As a further scheme of the present invention, in the category description strategy based on keyword expansion mentioned in Step2, synonyms corresponding to a plurality of emotion category keywords are queried according to the university of general organization emotion vocabulary ontology, and the synonyms corresponding to the category keywords are used together as emotion category description.
As a further scheme of the present invention, in the category description strategy based on keyword expansion mentioned in Step2, synonyms corresponding to 4 category keywords of "happy", "angry", "sad" and "fear" are queried according to the university of general organization emotional vocabulary ontology, the synonyms corresponding to the category keywords are used together as the emotional category description, and for the neutral category, the keywords "neutral" and "no emotion" are spliced together as the emotional category description.
As a further scheme of the present invention, in the category description policy based on emotion definition mentioned in Step2, the category description policy based on emotion definition queries specific definitions of each emotion category through encyclopedia, and after being manually adjusted, the category description policy is used as emotion category description.
The invention has the beneficial effects that: the invention firstly expands all emotion types into formalized type descriptions; secondly, based on the thought of question answering, the microblog and the category description are respectively used as questions and candidate answers, and a question-answer pair is formed and input into the BERT model to complete emotion classification. According to the method, the model can pay attention to the category information related to the microblog and the microblog information related to the category, so that rich context semantics and category semantics are fused, and the emotion of the microblog is better understood. Experiments on a microblog data set show that the method achieves the best effect, compared with a baseline method BERT, the macro-average and micro-average F1 values are respectively improved by 1.77% and 1.71%, and the effectiveness of the method is proved.
Drawings
Fig. 1 is a model diagram of a microblog emotion classification method based on emotion category description according to the present invention.
Detailed Description
Example 1: as shown in fig. 1, a microblog emotion classification method based on emotion category description specifically includes the following steps:
step1, collecting microblog data from the Sina microblog platform, and reserving 8306 microblogs as experimental data sets. And performing preprocessing operation, wherein the artificial label of each microblog is as follows: happenses, anger, sadness, fear, and neutral. Then, dividing the experimental data set into a training set, a verification set and a test set according to the proportion of approximately 8: 1, wherein the specific data set distribution is shown in table 1;
TABLE 1 Experimental data set partitioning (Unit: bar)
In Step1, data preprocessing is realized by adopting a python language writing program, and the operation of removing the duplication and deleting the characters such as '//', '@' and URL and the like is carried out on the text;
the design in the step is an important component of the invention, and is mainly used for constructing a microblog experimental data set, so that data support is provided for realizing microblog emotion classification based on emotion category description.
Step2, three emotion category description strategies are proposed, and 5 emotion categories of happoinsess, anger, sadness, fear and neutral are expanded into formal category descriptions; the three emotion category description strategies are respectively as follows: a category description strategy based on keywords, a category description strategy based on keyword expansion, and a category description strategy based on emotion definition;
as a further scheme of the present invention, in the keyword-based category description policy mentioned in Step2, the keyword-based category description policy directly uses chinese keywords respectively corresponding to 5 categories, namely happiess, anger, sadness, fear, and neutral, as emotion category descriptions.
As a further scheme of the present invention, in the category description strategy based on keyword expansion mentioned in Step2, synonyms corresponding to 4 category keywords of "happy", "angry", "sad" and "fear" are queried according to the university of general organization emotional vocabulary ontology, the synonyms corresponding to the category keywords are used together as the emotional category description, and for the neutral category, the keywords "neutral" and "no emotion" are spliced together as the emotional category description.
As a further scheme of the present invention, in the category description policy based on emotion definition mentioned in Step2, the category description policy based on emotion definition queries specific definitions of each emotion category through encyclopedia, and after being manually adjusted, the category description policy is used as emotion category description.
Finally, the construction examples of the three emotion category description strategies are shown in tables 2, 3 and 4, respectively.
TABLE 2 keyword-based Category description construction example
TABLE 3 Category description building example based on keyword expansion
Table 4 example of category description construction based on emotion definition
The optimal scheme design is an important component of the invention, mainly constructs emotion category description for the invention, and provides semantic information of emotion category for realizing microblog emotion classification based on emotion category description.
Step3, constructing a pre-trained BERT emotion classification model based on emotion category description: on the basis of Step1 and Step2, firstly, based on the thought of question answering, a microblog text is taken as a question, category descriptions of all emotion categories are spliced into a candidate answer set, the microblog text and the category descriptions form a question-answer pair together and are taken as input of a pre-training BERT model, and therefore rich context semantics and category semantic information are fused; secondly, inputting the hidden state of the 'CLS' token obtained by coding the question-answer pairs by the BERT model into a two-layer fully-connected neural network, and outputting the fusion semantic representation of the whole question-answer pairs; and finally, inputting the fused semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
As a preferred scheme of the present invention, in Step3, a construction method of a microblog text and emotion category description question-answer pair is as follows: giving a microblog text and all emotion classes { Yc|X}＝{Yc|x1,x2,…,xn},c＝1,2,…,N，YcDenotes N emotion categories, X ═ X1,x2,…,xnDenotes a microblog text. The microblog text is used as a question based on the idea of question-answering, and all emotion category descriptions are used as a group of candidate answer sets. Then, { Yc|X}＝{Yc|x1,x2,…,xn1,2, …, N can be expressed as a question-answer pair:
{Yc|X}＝“[CLS]X[SEP]Y1a category description of (1); y is2A category description of (1); …, respectively; y isNClass description of [ SEP ]]”
Wherein, "[ CLS ]" represents a special classification token, and the last layer hidden state of "[ CLS ]" can be used for representing the semantic meaning of the text to perform a classification task; "[ SEP ]" represents a segmenter token for segmenting the two sentences in the question-answer pair.
As a preferable embodiment of the present invention, Step3 mentioned above is the first "[ CLS ] of the present invention]Hidden state h of token[CLS]∈R768×1Outputting after passing through the full connection layer to represent the semantics of the whole input text, and marking as houtput∈R768×1：
houtput＝tanh(W1h[CLS]+b1) (1)
In the formula (1), W1∈R768×768As a weight matrix, b1∈R768×1Is a bias vector.
Finally, h isoutput∈R768×1Inputting the emotion category score vector s to the full connection layer to obtain the emotion category score vector s belonging to RN×1(ii) a And normalizing s by using a Softmax function to obtain a conditional probability distribution Pi(s), the formula is as follows:
s＝W2houtput+b2 (2)
in formulae (2) and (3), W2∈RN×768As a weight matrix, b2∈RN×1For the bias vector, N represents the number of emotion categories.
And Step4, carrying out an experiment on the constructed model aiming at the microblog data set, and verifying the effectiveness of the invention.
As a preferred embodiment of the present invention, in Step4, according to three proposed emotion classification description strategies, three models are constructed, which are respectively recorded as: BERT-KCD (BERT with Keyword-based Category Description) represents that Category Description based on keywords is introduced into a BERT model; BERT-KECD (BERT with Keyword Expansion-based Category Description) indicates that the Category Description based on the Keyword Expansion is introduced into a BERT model; BERT-EDCD (BERT with Emotion Definition-based Category Description) indicates that Category Description based on Emotion Definition is introduced in a BERT model.
Based on the experiment of the invention, microblog texts are taken as questions, the category descriptions of all emotion categories are spliced into a candidate answer set, the candidate answer set and the question-answer pair are formed together to be used as the input of a pre-training BERT model, and a 5-category emotion classifier is trained. For the experiment of the reference method (BERT), the microblog is directly used as the input of a pre-training BERT model, and a 5-class emotion classifier is trained. BERT-EDCD was chosen here as a representative model for the present invention and compared to the following baseline model:
MNB(Multinomial
bayes): performing well in many emotion classification tasks, the model's smoothing factor alpha is set to 1.0.
SVM (support Vector machines): is widely used in the emotion classification task and achieves excellent effect. The regularization constant C of the SVM model is set to 1.0 and the kernel function selects a linear kernel function linear.
BLSTM (Bidirectional Long Short-Term Memory): context-dependent textual features are extracted for emotion classification by bi-directional LSTM. The model adopts a single-layer bidirectional LSTM network, and the number of hidden layer units is 256.
CNN (volumetric Neural networks): the classic convolutional neural network proposed by Kim extracts deep semantic features through CNN to realize emotion classification of texts. The sizes of convolution kernels of the model are 3, 4 and 5, and the convolution kernels with different sizes are 100 respectively.
DPCNN (deep Pyramid capacitive networks): deep convolution operation is carried out at the word level, long-distance text features are extracted for emotion classification, and the best result at that time is obtained on a plurality of comment data sets such as Amazon.
HAN (cognitive attachment networks): the method comprises the steps of extracting word-level and sentence-level characteristics through a layered bidirectional GRU and an Attention mechanism, obtaining semantic representation of the whole text for emotion classification, and obtaining the best result in a plurality of emotion classification tasks.
BERT (bidirectional Encoder retrieval from transformations): the Google-sourced Chinese pre-training BERT model (BERT-Base, Chinese) refreshes a series of NLP task records including emotion classification tasks. The model has 12 layers of transformers, each layer of transformers has 12 self-attention heads, the hidden layer size is 768, and the total parameter number is 110M.
The microblog emotion classification experiment results and analysis are as follows:
TABLE 5 microblog emotion classification experiment results
Table 5 shows that the conventional machine learning models MNB and SVM are less effective than other deep learning models, and the Micro _ F measurement is only 70.11% and 70.23%, respectively. It can be seen that the Micro _ F measurements for BLSTM, CNN, DPCNN, HAN and BERT are 74.97%, 76.91%, 77.64%, 77.52% and 81.77%, respectively. The performance of these five deep learning based models is significantly better than MNB and SVM. Furthermore, the Micro _ F measure of BERT-EDCD is 83.48%, which is significantly better than the five deep learning models described above. Specifically, the Macro _ Precision, Macro _ Recall, Macro _ F-measure, and Micro _ F-measure of BERT-EDCD were increased by 11.57%, 19.13%, 18.76%, and 8.51%, respectively, as compared to BLSTM. Four indicators of BERT-EDCD were increased by 3.40%, 12.64%, 9.19% and 5.84%, respectively, compared to DPCNN. Four indices of BERT-EDCD were increased by 1.66%, 1.52%, 1.77%, and 1.71%, respectively, as compared to BERT. The effectiveness of the BERT-EDCD model is proved by experimental results, and the advantages of the model in microblog emotion classification are displayed.
In order to verify the effectiveness of the category description strategy provided by the invention, BERT-KCD, BERT-KECD and BERT-EDCD four models are compared, and the experimental results are shown in Table 6.
TABLE 6 validation of the effectiveness of the Category description policy
As can be seen from table 6, the introduction of three different class description strategies into the BERT model can improve the effect of the model. Compared with a BERT model, the Macro _ F1 and the Micro _ F1 of the BERT-KCD are respectively improved by 0.15 percent and 0.61 percent; the Macro _ F1 and the Micro _ F1 of BERT-KECD are respectively improved by 0.58% and 0.85%. We analyze that because the keyword-based category description reflects part of the semantic information of the tag, it can help the model to understand the emotion of the text. And, the richer the keywords, the more obvious the performance improvement of the model. In addition, the Macro _ F1 and the Micro _ F1 of BERT-EDCD are improved by 1.77% and 1.71%, respectively. It can be found that the performance improvement of BERT-EDCD is more significant than that of BERT-KCD and BERT-KECD. This is because the category description strategy based on emotion definition embodies semantic information of categories more completely, so that the model grasps meanings of each emotion category, thereby better understanding the emotion of the text. Experimental results show that the effect of the BERT model can be effectively improved by the three category description strategies provided by the invention, and the effectiveness of integrating the label semantic information in the microblog emotion classification task is proved.
To visually demonstrate the effectiveness of the models of the present invention, model predictions of BERT and BERT-EDCD were compared to manual labels, and the results are shown in Table 7.
TABLE 7 comparison of model predictions with Artificial tags
As can be seen from table 7, the present invention introduces the BERT-EDCD model based on mood-defined class descriptions to correctly predict all instances. The semantic information of the category can effectively guide the model to learn the meanings of various emotion categories, so that the emotion of the epidemic microblog text can be better understood. In contrast, BERT models that do not incorporate tagged semantic information are more prone to misclassification because BERT models only understand semantics and cannot assist in understanding the mood of text through category descriptions. Experimental results further prove the effectiveness and superiority of the microblog emotion classification method based on emotion category description.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Claims (8)
1. A microblog emotion classification method based on emotion category description is characterized by comprising the following steps: firstly, proposing a emotion category description strategy, and expanding all emotion categories of microblogs to be classified into formal category descriptions; secondly, splicing the microblog text and the category description into a question-answer pair, and inputting the question-answer pair into a pre-trained BERT model; secondly, inputting a hidden state obtained by coding the question-answer pairs through a BERT model into a two-layer fully-connected neural network, and outputting a fusion semantic representation of the whole question-answer pairs; and finally, inputting the fused semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
2. The microblog emotion classification method based on emotion classification description according to claim 1, wherein: the method comprises the following specific steps:
step1, collecting microblog texts as an experimental data set, and carrying out preprocessing operation, wherein the manual label of each piece of data after preprocessing operation is one of a plurality of emotion categories;
step2, three emotion category description strategies are proposed, and a plurality of emotion categories are expanded into formal category descriptions; the three emotion category description strategies are respectively as follows: a category description strategy based on keywords, a category description strategy based on keyword expansion, and a category description strategy based on emotion definition;
step3, constructing a pre-trained BERT emotion classification model based on emotion category description: on the basis of Step1 and Step2, firstly, based on the thought of question answering, a microblog text is taken as a question, category descriptions of all emotion categories are spliced into a candidate answer set, the microblog text and the category descriptions form a question-answer pair together and are taken as input of a pre-training BERT model, and therefore rich context semantics and category semantic information are fused; secondly, inputting a hidden state obtained by coding the question-answer pairs through a BERT model into a two-layer fully-connected neural network, and outputting a fusion semantic representation of the whole question-answer pairs; and finally, inputting the fused semantic representation of the question-answer pairs into a Softmax layer, outputting normalized emotion category probability distribution, and realizing emotion classification of the microblogs.
3. The microblog emotion classification method based on emotion classification description according to claim 2, wherein: in Step1, preprocessing includes deduplication, removal of "//", "@" and URL data.
4. The microblog emotion classification method based on emotion classification description according to claim 2, wherein: in Step1, the manual label of each piece of data after the preprocessing operation is as follows: happenses, anger, sadness, fear, and neutral are one of 5 mood categories.
5. The microblog emotion classification method based on emotion classification description according to claim 2, wherein: the keyword-based category description strategy mentioned in Step2 directly takes Chinese keywords corresponding to a plurality of emotion categories as emotion category descriptions.
6. The microblog emotion classification method based on emotion classification description according to claim 2, wherein: the category description strategy based on the keyword expansion mentioned in Step2 queries synonyms corresponding to a plurality of emotion category keywords according to the university of great courseware emotion vocabulary body based on the category description strategy based on the keyword expansion, and the synonyms corresponding to the category keywords are used as emotion category description together.
7. The microblog emotion classification method based on emotion classification description according to claim 4, wherein: according to the category description strategy based on keyword expansion mentioned in Step2, synonyms corresponding to 4 categories of keywords including 'happy', 'angry', 'sad' and 'fear' are inquired according to an emotional vocabulary body of university of large connecting studios, the synonyms corresponding to the category keywords are used as emotion category description together, and for the neutral category, the keywords 'neutral' and 'no emotion' are spliced together to be used as emotion category description.
8. The microblog emotion classification method based on emotion classification description according to claim 2, wherein: the category description strategy based on emotion definition mentioned in Step2 queries specific definitions of various emotion categories through Baidu encyclopedia, and the category description strategy based on emotion definition is manually adjusted to serve as emotion category description.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011408229.0A CN112989033B (en) | 2020-12-03 | 2020-12-03 | Microblog emotion classification method based on emotion category description |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011408229.0A CN112989033B (en) | 2020-12-03 | 2020-12-03 | Microblog emotion classification method based on emotion category description |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112989033A true CN112989033A (en) | 2021-06-18 |
| CN112989033B CN112989033B (en) | 2022-05-17 |
Family
ID=76344890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011408229.0A Active CN112989033B (en) | 2020-12-03 | 2020-12-03 | Microblog emotion classification method based on emotion category description |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112989033B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113220893A (en) * | 2021-07-09 | 2021-08-06 | 北京邮电大学 | Product feedback analysis system and method based on emotion analysis |
| CN113297366A (en) * | 2021-06-22 | 2021-08-24 | 中国平安人寿保险股份有限公司 | Multi-turn dialogue emotion recognition model training method, device, equipment and medium |
| CN113626577A (en) * | 2021-07-01 | 2021-11-09 | 昆明理工大学 | Chinese cross-language news event element extraction method based on reading understanding |
| CN117497140A (en) * | 2023-10-09 | 2024-02-02 | 合肥工业大学 | Multi-level depression state detection method based on fine granularity prompt learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794208A (en) * | 2015-04-24 | 2015-07-22 | 清华大学 | Sentiment classification method and system based on contextual information of microblog text |
| CN105205044A (en) * | 2015-08-26 | 2015-12-30 | 苏州大学张家港工业技术研究院 | Emotion and non-emotion question classifying method and system |
| CN106776581A (en) * | 2017-02-21 | 2017-05-31 | 浙江工商大学 | Subjective texts sentiment analysis method based on deep learning |
| CN108038166A (en) * | 2017-12-06 | 2018-05-15 | 武汉大学 | A kind of Chinese microblog emotional analysis method based on the subjective and objective skewed popularity of lexical item |
| CN110765245A (en) * | 2019-09-19 | 2020-02-07 | 平安科技（深圳）有限公司 | Emotion positive and negative judgment method, device and equipment based on big data and storage medium |
| CN111832312A (en) * | 2020-05-15 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Text processing method, device, equipment and storage medium |
-
2020
- 2020-12-03 CN CN202011408229.0A patent/CN112989033B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794208A (en) * | 2015-04-24 | 2015-07-22 | 清华大学 | Sentiment classification method and system based on contextual information of microblog text |
| CN105205044A (en) * | 2015-08-26 | 2015-12-30 | 苏州大学张家港工业技术研究院 | Emotion and non-emotion question classifying method and system |
| CN106776581A (en) * | 2017-02-21 | 2017-05-31 | 浙江工商大学 | Subjective texts sentiment analysis method based on deep learning |
| CN108038166A (en) * | 2017-12-06 | 2018-05-15 | 武汉大学 | A kind of Chinese microblog emotional analysis method based on the subjective and objective skewed popularity of lexical item |
| CN110765245A (en) * | 2019-09-19 | 2020-02-07 | 平安科技（深圳）有限公司 | Emotion positive and negative judgment method, device and equipment based on big data and storage medium |
| CN111832312A (en) * | 2020-05-15 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Text processing method, device, equipment and storage medium |
Non-Patent Citations (5)
| Title |
|---|
| WENQIANG LUO等: "Does ensemble really work when facing the twitter semantic classification?", 《2020 5TH INTERNATIONAL CONFERENCE ON COMPUTATIONAL INTELLIGENCE AND APPLICATIONS (ICCIA)》 * |
| 方英兰等: "基于BERT的文本情感分析方法的研究", 《信息技术与信息化》 * |
| 李丽华等: "基于深度学习的文本情感分析", 《湖北大学学报(自然科学版)》 * |
| 梁俊葛等: "基于集成一致性的多源跨领域情感分类模型", 《数据采集与处理》 * |
| 白旭栩: "基于特定事件的微博用户情感分析", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113297366A (en) * | 2021-06-22 | 2021-08-24 | 中国平安人寿保险股份有限公司 | Multi-turn dialogue emotion recognition model training method, device, equipment and medium |
| CN113626577A (en) * | 2021-07-01 | 2021-11-09 | 昆明理工大学 | Chinese cross-language news event element extraction method based on reading understanding |
| CN113220893A (en) * | 2021-07-09 | 2021-08-06 | 北京邮电大学 | Product feedback analysis system and method based on emotion analysis |
| CN117497140A (en) * | 2023-10-09 | 2024-02-02 | 合肥工业大学 | Multi-level depression state detection method based on fine granularity prompt learning |
| CN117497140B (en) * | 2023-10-09 | 2024-05-31 | 合肥工业大学 | Multi-level depression state detection method based on fine granularity prompt learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112989033B (en) | 2022-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111415740B (en) | Method and device for processing inquiry information, storage medium and computer equipment | |
| CN112989033B (en) | Microblog emotion classification method based on emotion category description | |
| Vateekul et al. | A study of sentiment analysis using deep learning techniques on Thai Twitter data | |
| CN109992783B (en) | Chinese word vector modeling method | |
| CN107967257B (en) | Cascading composition generating method | |
| Li et al. | Context-aware emotion cause analysis with multi-attention-based neural network | |
| CN108628828B (en) | Combined extraction method based on self-attention viewpoint and holder thereof | |
| WO2019153737A1 (en) | Comment assessing method, device, equipment and storage medium | |
| CN111275401B (en) | Intelligent interview method and system based on position relation | |
| CN110347836B (en) | Method for classifying sentiments of Chinese-Yue-bilingual news by blending into viewpoint sentence characteristics | |
| CN115048944B (en) | Open domain dialogue reply method and system based on theme enhancement | |
| CN112818106B (en) | Evaluation method for generating question and answer | |
| CN111914556A (en) | Emotion guiding method and system based on emotion semantic transfer map | |
| CN115659954A (en) | Composition automatic scoring method based on multi-stage learning | |
| CN115599902B (en) | Oil-gas encyclopedia question-answering method and system based on knowledge graph | |
| CN114756681B (en) | Evaluation and education text fine granularity suggestion mining method based on multi-attention fusion | |
| CN112231491B (en) | Similar test question identification method based on knowledge structure | |
| Aung et al. | Personality prediction based on content of Facebook users: A literature review | |
| Gan et al. | Unsupervised learning of sentence representations using convolutional neural networks | |
| CN112905736A (en) | Unsupervised text emotion analysis method based on quantum theory | |
| CN115659947A (en) | Multi-item selection answering method and system based on machine reading understanding and text summarization | |
| CN115329085A (en) | Social robot classification method and system | |
| Yuan | [Retracted] A Classroom Emotion Recognition Model Based on a Convolutional Neural Network Speech Emotion Algorithm | |
| CN115617960A (en) | Post recommendation method and device | |
| Cai et al. | XPROAX-local explanations for text classification with progressive neighborhood approximation |
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 |