CN114757183A - Cross-domain emotion classification method based on contrast alignment network - Google Patents
Cross-domain emotion classification method based on contrast alignment network Download PDFInfo
- Publication number
- CN114757183A CN114757183A CN202210373901.XA CN202210373901A CN114757183A CN 114757183 A CN114757183 A CN 114757183A CN 202210373901 A CN202210373901 A CN 202210373901A CN 114757183 A CN114757183 A CN 114757183A
- Authority
- CN
- China
- Prior art keywords
- domain
- cross
- target
- original
- comment
- 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 63
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000012549 training Methods 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 25
- 238000013145 classification model Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 10
- 238000012546 transfer Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 7
- 238000003062 neural network model Methods 0.000 abstract description 5
- 238000011160 research Methods 0.000 abstract description 3
- 239000000284 extract Substances 0.000 abstract description 2
- 238000003058 natural language processing Methods 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 7
- 230000006978 adaptation Effects 0.000 description 6
- 238000012800 visualization Methods 0.000 description 5
- 238000010606 normalization Methods 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007717 exclusion Effects 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 238000011065 in-situ storage Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008450 motivation Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000007 visual 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
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a cross-domain emotion analysis method based on a contrast alignment network, and belongs to the technical field of fine-grained emotion analysis in natural language processing. The invention researches an insufficiently explored scene of cross-domain emotion classification, namely a scene with few samples in a target domain. In this scenario, the present invention proposes a neural network model known as the Contrast Alignment Network (CAN). The model first randomly extracts two instances from the original domain and the target domain, and then trains the two instances according to the combined target domain and original domain. The first objective is to minimize classification errors on the original domain. The second is a pairwise comparison target, where the distance measure between the target domain instance and the original domain instance in a pair is minimized if they express the same emotion, otherwise the measure is maximized with a constant upper bound. The method solves the problem that the target field data resources are limited in the cross-field emotion classification task, and improves the use experience of users.
Description
Technical Field
The invention relates to a cross-domain emotion classification method, in particular to a cross-domain emotion analysis method based on a contrast alignment network, and belongs to the technical field of fine-grained emotion analysis in natural language processing.
Background
Cross-Domain Sentiment Classification (CDSC) is an important task, aiming at transferring learned knowledge from the original Domain to the target Domain. The CDSC enables the emotion classification model trained in the original domain with a large amount of labeled data to work well in the target domain data with limited training samples. This situation is common and challenging in the industry when the target domain has insufficient data and the original domain has sufficient data, with the main challenge being domain transfer (or distributed transfer) between the source domain and the target domain. The domain transfer problem is mainly the difference in distribution between any two domains, e.g., words used in the medical domain are very different from words used in the restaurant domain.
Domain transfer is an important problem in cross-domain emotion classification and can be alleviated to a large extent by domain adaptation methods. Currently, researchers have proposed various domain-adapted models. These models require a large amount of unlabeled data from the target domain so that they can learn a good representation of each target instance as the correct input to a classifier trained in the original domain.
Currently, Unsupervised Domain Adaptation (UDA) techniques have been used to solve the domain transfer problem. Essentially, UDAs take advantage of other unmarked data in the target domain to minimize domain skew by aligning statistics across domains. However, in practical applications, the large amount of unlabeled target domain data required for UDAs may not be sufficiently available, thereby limiting the applicability of UDAs.
Meanwhile, cross-domain emotion classification exists in a scene which is not fully explored at present, namely, a target domain is a few-sample scene, and the scene exists in many practical applications in the industry. Unlike unsupervised scenarios, the sample-less scenario does not require additional unlabeled target domain data, but only relies on scarce labeled data available in the target domain.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and solve the technical problem of data resource limitation in the target field, and creatively provides a cross-field emotion classification method based on a contrast alignment network. The invention researches an insufficiently explored scene of cross-domain emotion classification, namely a scene with few samples in a target domain. Unlike unsupervised scenarios, the sample-less scenario does not require additional unlabeled target domain data, but relies only on scarce labeled data available in the target domain. Under this scenario, the present invention proposes a neural network model known as a Contrast Alignment Network (CAN). The model first randomly extracts two instances from the original domain and the target domain, and then trains the two instances according to the combined target domain and original domain. The first objective is to minimize the classification errors on the original domain. The second is a pairwise comparison target, where the distance measure between the target domain instance and the original domain instance in a pair is minimized if they express the same emotion, otherwise the measure is maximized with a constant upper bound.
Notably, the motivation for aligning networks is not just to address the resource limitation issue. In fact, by deducing that the model provided by the invention has a more complete learning boundary, a theoretical basis is provided for the basic efficacy of the model.
Cross-domain emotion analysis CDSC requires an in-situ domain
Middle training model, original field
In which n issExample for target area
Emotion classification is performed. Xi sIs the ith sample sentence of the original field, yi sIs the label corresponding to the ith sample in the original field. In other words, the original domain
Example sample distribution compliance
Also, in the target area are
XtSample sentence, y, representing target field ttAnd a label corresponding to the sample sentence representing the target field t.
And
represents the domain offset between the joint probabilities of the source and target, respectively, and
the concept of domain adaptation is imposed to alleviate the domain transfer problem. In the case where there is not enough tagged data in the target domain, the unsupervised domain adaptation method utilizes a large amount of additional untagged target data
To align two fields, ntThe target domain is the number of marked data samples. However, the large amount of unlabeled data required may not always be available in the target domain. Therefore, the method focuses on a more practical case of cross-domain emotion analysis, namely a sample case, and uses a small amount of auxiliary mark target domain data instead 
nt＜＜ns，ntThe number of target field data samples is marked for a small number of aids.
Inspired by the adaptation work in the existing unsupervised field, the invention decomposes the learning objective into two parts: and calculating the risk of the discriminant original domain and the regularized domain transfer. In order to further minimize the risk of the original field, the invention provides a method for comparing and aligning the original field and the target field by taking example-level classification information as a condition.
The technical method adopted by the invention is as follows.
An emotion classification method based on a contrast alignment network comprises the following steps:
first, text data preprocessing is performed.
And loading the comment corpus set and the pre-training language model, and performing text preprocessing and text data formatting on the comment text data in the comment corpus set. The pre-training language model may adopt a BERT model, a RoBERTa model, or the like.
And then, constructing a cross-domain emotion classification model based on the contrast alignment network.
The cross-domain emotion classification model f based on the contrast alignment network comprises a coder gθAnd a classifier hφOn the basis of the architecture, an original domain classification target loss function is introduced
And comparing the target domain classification target loss functions
Wherein, the encoder g θAnd using the pre-training language model as a base for coding the context information of the comment sentence. Preferably, the encoder uses CLS (sentence vector representation) full-sentence representation of the pre-trained language model as a context-hidden-state representation vector H of the entire comment sentence, H ═ H1，h2，...，hn}，hnThe hidden state representing the nth token represents a vector.
Classifier hφConsists of a multi-layer perceptron MLP and softmax layers (soft maximization normalization layers). Wherein, multilayer perceptron includes the four layers, does in proper order: a fully connected layer, a ReLU (linear rectification function) activation function layer, a dropout layer (random discard layer), and a fully connected layer. To be passed through MLPThe output representation is passed to the softmax layer (soft maximization normalization layer) from which the corresponding losses are calculated.
Then, the discriminative original domain risk is calculated.
In the method, for the risk of the discriminant original field, the empirical classification loss term of the original field is adopted to model the classification target as the loss based on the cross entropy
Wherein n issThe number of original field data samples; y isiIs the label of the ith sample of the original domain data,
is a prediction label of the model on the ith sample of the original domain data.
Then, the original domain and the target domain are aligned by comparing the example-level classification information.
The goal is to minimize the domain bias under limited target domain data. Although distribution level alignment is difficult to achieve with limited target data, instance level alignment can be merged and can be achieved with a certain probability.
In the method of the invention, contrast loss is introduced to minimize the distance measure between the original and target domain instances of the sampling pair (if they share the same emotion), otherwise maximize the measure. Therefore, the emotion labels can be accurately assigned regardless of the field. Although the distance measure between the original domain and the target domain can be directly applied to the input x, the hidden representation Z is aligned by following most domain adaptive methods, which is more abstract and can capture more semantic information.
In particular, any pair is given
Specific loss of contrast
The calculation is as follows:
wherein Xi sSample comment sentence, X, representing the ith sample of the original fieldj tSample comment sentence, y, j, representing target field ti sA label, y, corresponding to the comment statement of the ith sample of the original field sj tRepresenting a label corresponding to the jth sample comment statement in the target field t;
representing a distance measure between the original domain instance and the target domain instance; 
Indicating a function for the equation; m is a predefined constant.
The contrast loss objective pushes the same emotion polarity instances closer, rejecting the different emotion polarity instances. Furthermore, the goal is not to push the different clusters to infinity, but rather to limit the exclusion range to a constant, as a relaxation to the learning algorithm.
Then, regularized domain transfer is performed.
The overall target of the method comprises a cross entropy loss function of original field data
Comparing and aligning original domain and target domain loss functions
And pass through
Regularization minimization of cross entropy loss function of raw domain data
And comparing the aligned original domain and the target domain loss function
Wherein the overall objective function is:
where α is a trade-off between classification and contrast objectives and λ is a regularization coefficient for all model parameters Θ ═ θ, Φ.
Then, model training is performed. Using standard batch random gradient descent algorithm to perform overall objective function
And (5) training.
Specifically, batch iterative training is carried out on all training samples in the training set, and a trained cross-domain emotion classification model based on a contrast alignment network is obtained.
And finally, performing cross-domain emotion classification by using the trained cross-domain emotion classification model based on the contrast alignment network.
Advantageous effects
Compared with the prior art, the method has the following advantages:
1. the method solves the problem that the target field data resources are limited in the cross-field emotion classification task. The invention researches an insufficiently explored scene of cross-domain emotion classification, namely a scene with few samples in a target domain. Unlike unsupervised scenarios, the sample-less scenario does not require additional unlabeled target domain data, but relies only on scarce labeled data available in the target domain.
2. The neural network model of the contrast alignment network provided by the method comprises two targets, wherein the first target is to minimize the classification error in the original field. The second is a pairwise comparison target, where the distance measure between the target domain instance and the original domain instance in a pair is minimized if they express the same emotion, otherwise the measure is maximized with a constant upper bound.
3. The performance of the comparison and alignment network model provided by the invention on the cross-domain emotion classification task is obviously superior to that of a corresponding baseline model. Other baseline models have a dramatic drop in performance across the domain data set. In contrast, the contrast-aligned neural network model proposed by the present invention is more robust than other baseline models.
4. The contrast alignment neural network model provided by the invention not only solves the problem of resource limitation in the target field, but also has a more complete learning boundary, and provides a theoretical basis for the basic efficacy of the provided model.
5. The method improves the problems in the cross-domain emotion classification of the existing fine-grained emotion analysis, and can well improve the use experience of the user.
Drawings
FIG. 1 is an overall flow diagram of the process of the present invention.
FIG. 2 is an exemplary diagram of a cross-domain emotion classification task.
FIG. 3 is a visual representation of the effect of the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and examples.
Examples
As shown in FIG. 1, a cross-domain emotion classification method based on a contrast alignment network includes the following steps:
step 1: and (5) preprocessing the text.
First, a corpus of comments and a pre-trained language model are loaded. The pre-training language model may be a BERT model, or may be another model (e.g., RoBERTa model).
And then, performing text preprocessing and text data formatting on the comment corpus.
Specifically, the method comprises the following steps:
step 1.1: and extracting attribute words, viewpoint words and position information thereof from each comment sentence.
Step 1.2: pre-segmenting words of the comment sentence by using an nltk word segmenter, and separating segmented token words by using spaces.
Step 1.3: adding two special token words after the comment sentence segmentation token sequence: [ CLS]、[SEP]Thereby, a general input form is constructed: s { [ CLS { []，w1，w2...，wn，[SEP]N denotes the total number of token words of the comment sentence, wnThe nth token representing the comment sentence.
Step 1.4: and formatting the text data.
And (5) performing padding (padding) processing on each comment sentence token word sequence to enable the length of each comment sentence token word sequence to be 128. Token operation is performed on each token word in the comment sentence using the tokenizer of the pre-trained language model. And dividing the processed data set into a training set, a verification set and a test set, and constructing the training set, the verification set and the test set into a batch data form.
Step 2: and constructing a cross-domain emotion classification model based on a contrast alignment network.
A cross-domain emotion classification model f based on a contrast alignment network is composed of a coder gθAnd a classifier hφAnd (4) forming. Based on encoder gθAnd a classifier hφArchitecture for introducing domain-specific objective loss functions
And comparing the target domain classification target loss functions
Wherein, the encoder gθEncoding context information of the comment sentence. For the emotion classification task, emotion information of a sentence is contained in the context of the sentence, and it is particularly important to perform context modeling on the whole sentence. Thus, the full-sentence CLS representation of the pre-trained language model is used as a context-hidden state representation of the entire comment sentence Vector H, H ═ H { H }1，h2，...，hn}，hnThe hidden state representing the nth token represents a vector. Each token word in the sequence of comment sentences is mapped into an encoding vector.
Classifier hφConsisting of a multi-layer perceptron MLP and a softmax layer (soft maximization normalization layer). Wherein, the multilayer perceptron includes four layers: a fully connected layer, a ReLU (linear rectification function) activation function layer, a dropout layer (random drop layer), and a fully connected layer. The output representation passing through the fully-connected layer is sent to the softmax layer (soft maximization normalization layer), so that the corresponding label is predicted, and the corresponding target loss is calculated.
And 3, step 3: and calculating the risk of the discriminative original field.
For the discriminative original field risk, the invention adopts the experience classification loss term of the original field and then models the classification target as the loss based on the cross entropy
Wherein n issIs the number of original domain data samples, yiIs the label of the ith sample of the original domain data,
is a prediction label of the model on the ith sample of the original domain data.
And 4, step 4: and comparing and aligning the original domain and the target domain by using example-level classification information.
The domain migration is minimized under limited target domain data. Although distribution level alignment is difficult to achieve with limited target data, instance level alignment can be merged and with a certain probability.
Thus, the present invention introduces a loss of contrast to minimize the distance measure between the original and target domain instances of the sampling pair if they share the same emotion, otherwise maximize the measure. The emotion labels can be accurately assigned regardless of the field.
Although the distance measure between the original domain and the target domain can be directly applied to the input x, the present invention follows most domain adaptation methods, aligning the hidden representation Z. This is more abstract and can capture more semantic information.
In particular, any pair is given
The specific contrast loss was calculated as:
wherein, the first and the second end of the pipe are connected with each other,
a distance measure between the original domain instance and the target domain instance is represented.
Indicating a function for the equation. m is a predefined constant.
The contrast loss objective pushes the instances of the same emotion polarity closer, rejecting the instances of different emotion polarities. Furthermore, the goal is not to push the different clusters to infinity, but rather to limit the exclusion range to a constant, as a relaxation of the learning algorithm.
And 5: regularization domain transfer.
The overall target of the method comprises a cross entropy loss function of original field data 
Comparing and aligning original domain and target domain loss functions
By passing
Regularization minimization of cross entropy loss function of original domain data
And comparing the aligned original domain and the target domain loss function
The overall objective function is:
where α is a trade-off between classification and contrast objectives, λ is a regularization coefficient for all model parameters Θ ═ θ, Φ, θ denotes the parameters of the encoder, and Φ denotes the parameters of the classifier.
Step 6: and training a cross-domain emotion classification model based on a contrast alignment network.
Using standard batch random gradient descent algorithm to perform overall objective function
And carrying out optimization training.
Specifically, batch iterative training is carried out on all training samples in the training set, and a trained cross-domain emotion classification model based on a contrast alignment network is obtained.
And 7: and performing cross-domain emotion classification by using the trained cross-domain emotion classification model based on the contrast alignment network.
Further, the method can be evaluated. And after the training is finished in the training set, performing verification test in the used verification set. The evaluation indexes used include:
for cross-domain emotion classification, the accuracy and the F1 value are used as evaluation indexes;
The optimal model is updated for each round of validation and saved.
Test verification
The method was tested. Firstly, loading the optimal model parameters and the test data which are stored before, and then converting the test data into the required format to be input into the optimal model for testing. The evaluation index is the same as the evaluation index used in the verification.
As shown in fig. 2, for the original field that is the comment field of the notebook computer, the comment sentence "the brand of the notebook computer is too jammed, the use experience is very poor", the target field that is the comment sentence "the dishes of other households are general, the services are also general" in the restaurant field, and there is a fine-grained field deviation in the data of the original field and the target field, so that most of the cross-field emotion classification methods and the unsupervised adaptive methods cannot perform another migration well, and the emotion polarity of the comment sentence in the target field cannot be effectively determined.
As shown in fig. 3, a cross-domain (laptop review domain and restaurant review domain) effect visualization representation of the emotion classification method based on the contrast alignment network, the left graph is a cross-domain effect visualization graph without the contrast alignment network, and the right graph is a cross-domain effect visualization graph through the contrast alignment network. It can be seen that the model effect visualization with the right graph with the contrast aligned network has a more elegant manifold than the effect visualization without the contrast aligned network, which means that the contrast aligned network has generalization capability even if the target data is small.
The above description is a preferred embodiment of the present invention, and the present invention should not be limited to the disclosure of the embodiment and the drawings. Equivalents and modifications may be made without departing from the spirit of the disclosure, which is to be considered as within the scope of the invention.
Claims (5)
1. A cross-domain emotion classification method based on a contrast alignment network is characterized by comprising the following steps:
step 1: loading a comment corpus set and a pre-training language model, and performing text preprocessing and text data formatting on comment text data in the comment corpus set;
step 2: constructing a cross-domain emotion classification model based on a contrast alignment network;
wherein, the cross-domain emotion classification model f based on the contrast alignment network comprises a coder gθAnd a classifier hφOn the basis of the architecture, an original domain classification target loss function is introduced
And comparing the target domain classification target loss functions
Encoder gθUsing a pre-training language model as a base for encoding context information of the comment statement; classifier hφThe sensor consists of a multi-layer sensor MLP and a softmax layer; sending the output representation passing through the multilayer perceptron to the softmax layer, thereby calculating the corresponding loss;
and step 3: calculating the risk of the discriminant original field;
For the risk of the discriminant original field, the empirical classification loss item of the original field is adopted to model the classification target into the loss based on the cross entropy
Wherein n issThe number of data samples in the original field; y isiIs the label of the ith sample of the original domain data,
a prediction label of the model on the ith sample of the original field data;
and 4, step 4: comparing and aligning the original field and the target field by using example-level classification information;
given an arbitrary pair
Specific loss of contrast
The calculation is as follows:
wherein, Xi sSample comment sentence, X, representing the ith sample of the original fieldj tSample comment sentence, y, j, representing target field ti sA label, y, corresponding to the comment statement of the ith sample of the original field sj tRepresenting a label corresponding to the jth sample comment statement in the target field t;
representing a distance measure between the original domain instance and the target domain instance;
indicating a function for an equation; m is a predefined constant;
and 5: carrying out regularization domain transfer; the overall objective includes a cross-entropy loss function of the original domain data
Comparing and aligning original domain and target domain loss functions
And pass through
Regularization minimization of cross entropy loss function of raw domain data
And comparing the aligned original domain and the target domain loss function 
Wherein the overall objective function is:
where α is a trade-off between classification and contrast objectives, and λ is a regularization coefficient for all model parameters Θ ═ θ, Φ };
step 6: using standard batch random gradient descent algorithm to perform overall objective function
Training to obtain a trained cross-domain emotion classification model based on a contrast alignment network;
and 7: and performing cross-domain emotion classification by using the trained cross-domain emotion classification model based on the contrast alignment network.
2. A cross-domain emotion classification method based on a contrast alignment network is characterized in that a coder uses CLS full sentence representation through a pre-trained language model as a context hidden state representation vector H of an entire comment sentence, and H is { H ═ H }1,h2,…,hn}，hnThe hidden state representing the nth token represents a vector.
3. A cross-domain emotion classification method based on a contrast alignment network is characterized in that in step 2, a multilayer perceptron comprises four layers, which are sequentially as follows: the system comprises a full connection layer, a ReLU activation function layer, a dropout layer and a full connection layer.
4. A cross-domain emotion classification method based on a contrast alignment network is characterized in that the step 1 comprises the following steps:
Step 1.1: extracting attribute words, viewpoint words and position information thereof from each comment sentence;
step 1.2: pre-segmenting words of the comment sentence by using an nltk word segmenter, and separating segmented token words by using spaces;
step 1.3: adding two special token words after the comment sentence word segmentation token sequence: [ CLS]、[SEP]Thereby, a general input form is constructed: s { [ CLS { [],w1,w2…,wn,[SEP]N denotes the total number of token words of the comment sentence, wnAn nth token representing a comment sentence;
step 1.4: formatting text data;
filling up each comment sentence token word sequence to make the length of each comment sentence token word sequence be 128; using a tokenizer of a pre-training language model to perform tokenize operation on each token word in the comment sentence; and dividing the processed data set into a training set, a verification set and a test set, and constructing the training set, the verification set and the test set into a batch data form.
5. A cross-domain emotion classification method based on a contrast alignment network is characterized in that after training is completed in a training set, verification testing is performed in a used verification set, and evaluation indexes include:
for cross-domain emotion classification, the accuracy and the F1 value are used as evaluation indexes;
the optimal model is updated and saved for each round of validation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210373901.XA CN114757183B (en) | 2022-04-11 | 2022-04-11 | Cross-domain emotion classification method based on comparison alignment network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210373901.XA CN114757183B (en) | 2022-04-11 | 2022-04-11 | Cross-domain emotion classification method based on comparison alignment network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114757183A true CN114757183A (en) | 2022-07-15 |
| CN114757183B CN114757183B (en) | 2024-05-10 |
Family
ID=82328537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210373901.XA Active CN114757183B (en) | 2022-04-11 | 2022-04-11 | Cross-domain emotion classification method based on comparison alignment network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114757183B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115239937A (en) * | 2022-09-23 | 2022-10-25 | 西南交通大学 | Cross-modal emotion prediction method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308318A (en) * | 2018-08-14 | 2019-02-05 | 深圳大学 | Training method, device, equipment and the medium of cross-domain texts sentiment classification model |
| CN109753566A (en) * | 2019-01-09 | 2019-05-14 | 大连民族大学 | The model training method of cross-cutting sentiment analysis based on convolutional neural networks |
| CN110032646A (en) * | 2019-05-08 | 2019-07-19 | 山西财经大学 | The cross-domain texts sensibility classification method of combination learning is adapted to based on multi-source field |
| CN110503092A (en) * | 2019-07-22 | 2019-11-26 | 天津科技大学 | The improvement SSD monitor video object detection method adapted to based on field |
| AU2020100710A4 (en) * | 2020-05-05 | 2020-06-11 | Chen, Dadu Mr | A method for sentiment analysis of film reviews based on deep learning and natural language processing |
| CN112163091A (en) * | 2020-09-25 | 2021-01-01 | 大连民族大学 | CNN-based aspect-level cross-domain emotion analysis method |
| CN113326378A (en) * | 2021-06-16 | 2021-08-31 | 山西财经大学 | Cross-domain text emotion classification method based on parameter migration and attention sharing mechanism |
-
2022
- 2022-04-11 CN CN202210373901.XA patent/CN114757183B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308318A (en) * | 2018-08-14 | 2019-02-05 | 深圳大学 | Training method, device, equipment and the medium of cross-domain texts sentiment classification model |
| CN109753566A (en) * | 2019-01-09 | 2019-05-14 | 大连民族大学 | The model training method of cross-cutting sentiment analysis based on convolutional neural networks |
| CN110032646A (en) * | 2019-05-08 | 2019-07-19 | 山西财经大学 | The cross-domain texts sensibility classification method of combination learning is adapted to based on multi-source field |
| CN110503092A (en) * | 2019-07-22 | 2019-11-26 | 天津科技大学 | The improvement SSD monitor video object detection method adapted to based on field |
| AU2020100710A4 (en) * | 2020-05-05 | 2020-06-11 | Chen, Dadu Mr | A method for sentiment analysis of film reviews based on deep learning and natural language processing |
| CN112163091A (en) * | 2020-09-25 | 2021-01-01 | 大连民族大学 | CNN-based aspect-level cross-domain emotion analysis method |
| CN113326378A (en) * | 2021-06-16 | 2021-08-31 | 山西财经大学 | Cross-domain text emotion classification method based on parameter migration and attention sharing mechanism |
Non-Patent Citations (1)
| Title |
|---|
| 赵传君;王素格;李德玉;: "基于集成深度迁移学习的多源跨领域情感分类", 山西大学学报(自然科学版), no. 04, 4 April 2018 (2018-04-04) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115239937A (en) * | 2022-09-23 | 2022-10-25 | 西南交通大学 | Cross-modal emotion prediction method |
| CN115239937B (en) * | 2022-09-23 | 2022-12-20 | 西南交通大学 | Cross-modal emotion prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114757183B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112214995B (en) | Hierarchical multitasking term embedded learning for synonym prediction | |
| CN110298037B (en) | Convolutional neural network matching text recognition method based on enhanced attention mechanism | |
| CN111241837B (en) | Theft case legal document named entity identification method based on anti-migration learning | |
| CN111738003B (en) | Named entity recognition model training method, named entity recognition method and medium | |
| CN106599032B (en) | Text event extraction method combining sparse coding and structure sensing machine | |
| CN111046670B (en) | Entity and relationship combined extraction method based on drug case legal documents | |
| Kulkarni et al. | Deep learning for NLP | |
| CN112732916A (en) | BERT-based multi-feature fusion fuzzy text classification model | |
| CN111597340A (en) | Text classification method and device and readable storage medium | |
| CN112749274B (en) | Chinese text classification method based on attention mechanism and interference word deletion | |
| CN114818717A (en) | Chinese named entity recognition method and system fusing vocabulary and syntax information | |
| CN113626589A (en) | Multi-label text classification method based on mixed attention mechanism | |
| CN111582506A (en) | Multi-label learning method based on global and local label relation | |
| CN114281931A (en) | Text matching method, device, equipment, medium and computer program product | |
| CN113934835B (en) | Retrieval type reply dialogue method and system combining keywords and semantic understanding representation | |
| CN115563314A (en) | Knowledge graph representation learning method for multi-source information fusion enhancement | |
| CN117574904A (en) | Named entity recognition method based on contrast learning and multi-modal semantic interaction | |
| CN114757183B (en) | Cross-domain emotion classification method based on comparison alignment network | |
| CN111723572B (en) | Chinese short text correlation measurement method based on CNN convolutional layer and BilSTM | |
| CN114022687A (en) | Image description countermeasure generation method based on reinforcement learning | |
| Hua et al. | A character-level method for text classification | |
| CN116956228A (en) | Text mining method for technical transaction platform | |
| CN116662924A (en) | Aspect-level multi-mode emotion analysis method based on dual-channel and attention mechanism | |
| CN116049349A (en) | Small sample intention recognition method based on multi-level attention and hierarchical category characteristics | |
| CN115455144A (en) | Data enhancement method of completion type space filling type for small sample intention recognition |
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 |