Nothing Special   »   [go: up one dir, main page]

CN114722808B - Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network - Google Patents

Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network Download PDF

Info

Publication number
CN114722808B
CN114722808B CN202210245787.2A CN202210245787A CN114722808B CN 114722808 B CN114722808 B CN 114722808B CN 202210245787 A CN202210245787 A CN 202210245787A CN 114722808 B CN114722808 B CN 114722808B
Authority
CN
China
Prior art keywords
graph
sentence
emotion
words
collaborative
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.)
Active
Application number
CN202210245787.2A
Other languages
Chinese (zh)
Other versions
CN114722808A (en
Inventor
饶国政
喻开智
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN202210245787.2A priority Critical patent/CN114722808B/en
Publication of CN114722808A publication Critical patent/CN114722808A/en
Application granted granted Critical
Publication of CN114722808B publication Critical patent/CN114722808B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/045—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
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Machine Translation (AREA)

Abstract

The invention discloses a specific target emotion recognition method based on a multi-context and multi-word segment graph rolling network, which comprises the following steps: converting the target sentence into a vector, inputting the vector into an LSTM neural network for reinforcement learning, and outputting the sentence vector; carrying out graph convolution operation on the obtained local collaborative graph and sentence vector from the global collaborative graph to output a first graph convolution output result; performing graph convolution operation on sentence graphs and sentence vectors obtained by using a word segmentation tool space to output a second graph convolution output result; performing inner product operation on the first graph convolution output result and the second graph convolution output result by using a gating mechanism, and transferring the collaborative graph information to a graph representation; and (3) gathering target words in the output result of the gating mechanism by using the attention mechanism to obtain target information for classifying target emotion. The method based on multi-context and multi-word segment learning can automatically identify emotion information.

Description

Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network
Technical Field
The invention relates to the technical field of computer technology and natural language, in particular to a specific target emotion recognition model based on a multi-context and multi-word segment graph convolution network, which automatically learns emotion information by learning multi-sentence context and different word segments to realize emotion analysis tasks.
Background
The language is very important for human, the language brings people to people communication, the language is recorded, the words are the words, one person starts to learn the words from birth, understanding the content of the words becomes very important, in the internet age today, everyone is away from a mobile phone, people need to purchase favorite articles on shopping websites, comments of the articles are important to reference, and emotion information in the comment words is important to the invention, so that the emotion recognition analysis field is generated.
Before the deep learning algorithm appears, the emotion analysis task is based on a traditional feature statistics method, and a representative method of the emotion analysis task is a method adopting a support vector machine and text features, and the emotion analysis task has the advantages of being capable of completing the emotion analysis task, and has the disadvantages of needing a large amount of labor cost and being unfavorable for commercialization. The occurrence of deep learning solves the problem of high cost of manually constructed features, the deep learning adopts a special structure neural network to automatically learn the features, the neural network is based on a perceptron, each neuron represents a nonlinear function, the neural network can also realize vector conversion of different space dimensions, and the neural network becomes the important of the deep learning.
Deep learning presents its advantages over multiple tasks, achieving best results in the computer vision field, and also in the emotion analysis field, where the operational flow is generally vectorizing the input data, changing low-dimensional data to high-dimensional, the changing method has
Figure SMS_1
Figure SMS_2
In Chinese data processing, the usual method is +.>
Figure SMS_3
The tool performs word segmentation, and in English data processing, a parser is generally used to realize entity recognition, part-of-speech tagging and syntactic dependency parsing, such as
Figure SMS_4
Figure SMS_5
And->
Figure SMS_6
Disclosure of Invention
The invention aims at overcoming the technical defects in the prior art and provides a specific target emotion recognition method based on a multi-context and multi-word segment graph rolling network.
The technical scheme adopted for realizing the purpose of the invention is as follows:
a specific target emotion recognition method based on a multi-context and multi-word segment graph convolution network is realized based on a specific target emotion recognition model, and the processing steps of the model are as follows:
converting the target sentence into a vector, inputting the vector into an LSTM neural network for reinforcement learning, and outputting the sentence vector;
performing graph convolution operation on the obtained local collaborative graph from the global collaborative graph and the sentence vector to output a first graph convolution output result;
performing graph convolution operation on sentence graphs obtained by using a word segmentation tool space and sentence vectors to output a second graph convolution output result;
performing inner product operation of a matrix on the first graph rolling output result and the second graph rolling output result by using a gating mechanism, and transferring the collaborative graph information to a graph representation;
and (3) gathering target words in the output result of the gating mechanism by using the attention mechanism to obtain target information for classifying target emotion.
The global collaborative graph is a large matrix and is used for counting occurrence frequencies of different two words in a plurality of sentences; the local synergy graph is a small matrix, and each element value is derived from the global synergy graph.
Wherein, only the relation among open class words containing more emotion components is considered when counting the occurrence frequency of different two words in a plurality of sentences in the global collaborative graph.
The global collaborative map is used for sentence processing, maps words into corresponding part-of-speech tags, and is divided into an open class, a closed class and other three types, wherein the open class comprises adjectives and emotion words including adverbs.
The method comprises the steps of calculating the occurrence frequency of all words in a multi-sentence context by using a statistical method, forming a global collaborative graph among all words, then searching weights of corresponding positions of specific targets from the global collaborative graph by only considering the specific targets and open words according to specific sentences and specific targets, and putting the weights into the positions of the specific targets by using the obtained weights to obtain a local collaborative graph of a current sentence.
When the LSTM neural network is used for reinforcement learning, sentences are regarded as a set of word segments from left to right, an agent is defined as a hidden state of a word segment action selection layer, the state is evaluated at first, then the action of the word segment is selected, and the action refers to the selection of a certain emotion; training by adopting a strategy gradient method.
For the state evaluation layer, selecting a layer of LSTM neural network, inputting word vectors and states of the intelligent agents, randomly initializing the states of the intelligent agents during initialization, and outputting the state evaluation layer to the action selection layer; the action selection layer adopts three layers of LSTM neural networks, different neural networks LSTM are adopted for different actions, the action selection layer calculates the action with the highest probability on the input from the state evaluation layer, the most probable emotion selection of a batch of words is obtained, the emotion with the highest frequency in the batch is selected as emotion action selection, the emotion is input into the corresponding LSTM neural network, and the state of an intelligent body is output; the last moment of agent state output is used as the expression of the whole phrase, and the hidden state is mapped to the output probability distribution through a softmax output layer.
The multi-context and multi-word segment learning-based method provided by the invention can automatically identify emotion information experimental results
Figure SMS_7
Data set and +.>
Figure SMS_8
The accuracy of the data set reaches +.>
Figure SMS_9
And->
Figure SMS_10
Drawings
Fig. 1 is a block diagram of a sentence in which Spacy dependency parser analyzes sentence dependency.
Fig. 2 is a block diagram of the present invention based on a collaborative graph neural network.
FIG. 3 is a diagram of a circular memory neural network model for multi-word segment learning of the present invention.
Fig. 4 is a diagram showing the original data set.
Fig. 5 shows the data after sentence analysis using spacy dependency parser.
FIG. 6 shows a partially pre-trained 300-dimensional word vector of size 4907255 KB.
FIG. 7 is a graph of the loss and number of iterations of training of the model of the present invention.
Detailed Description
The invention is described in further detail below with reference to the drawings and the specific examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
1-2, the specific target emotion recognition method based on the multi-context and multi-word segment graph convolution network in the embodiment of the invention provides a collaborative graph model, wherein the collaborative graph can explore the relation between a specific target and a plurality of sentence contexts, and the collaborative graph is divided into two, namely a global collaborative graph and a local collaborative graph. The invention constructs a large global synergy graph which is a large matrix and can count the occurrence frequency of different two words in a plurality of sentences; aiming at different sentences, the invention constructs a local synergy graph (collaborative graph), wherein the local synergy graph is a small matrix, and each element value is derived from a global synergy graph. In the global collaborative graph, only the relation among open class words is considered, and because the open class words contain more emotion components, the invention combines the collaborative graph with the traditional graph network, and a novel gating mechanism is provided, and can control information transfer.
According to the embodiment of the invention, the global collaborative map and the local collaborative map can explore the relation between the specific target and the contexts of a plurality of sentences, and the defect that the specific target only considers isolated sentences is alleviated; the invention maps the words into corresponding part-of-speech labels, which are divided into open class, closed class and other three types, wherein the open class has adjectives, adverbs and other emotion words, and only the relation among the open class words containing emotion words is considered. In a collaborative graph, the present invention represents relationships by word to word frequency in a multi-sentence context, thus distinguishing from traditional syntactic graphs only considering whether or not there is syntactical relationship between words.
Firstly, the invention calculates the occurrence frequency of all words in the multi-sentence context by using a statistical method, forms a global collaborative graph T among all words, then only considers specific targets and open class words (adjective ADJ, adverb ADV, exclamatory word INTJ, name NOUN, pronoun PRON and VERB VERB) according to specific sentences and specific targets, retrieves the weight of the corresponding position of the specific targets from the global collaborative graph, and places the weight into the position of the specific targets by using the obtained weight, thereby obtaining the local collaborative graph of the current sentence.
As shown in fig. 2, the sentences "The appetizer are ok, but the service is slow" are input, and are input to a local collaborative graph (collaboral graph) and a syntax dependency graph (syntax graph) process, and simultaneously the sentences are converted into vectors by word embedding (word embedding) and are input to an LSTM neural network (long short term neural network) process, and sentence vectors are output. The overall process is shown below:
H＝LSTM(E) (1)
in the formula, E represents an embedded representation of a sentence, and H represents a sentence vector.
X＝gcn(H,LocalT) (2)
gcn(H,localT)＝σ(LocalT*H*W 1 +b 1 ) (3)
In the above formula, X represents the output of the local synergy graph (collaborative graph),Hthe sentence vector is represented as such,LocalTa partial synergy graph is shown.gcn()A graph convolution operation is shown and is described,W 1 andb 1 is the weight of the neural network, representing the outer product operation of the matrix.
Y＝gcn(H,G) (4)
Y is the output result of the syntax dependency graph (syntax graph), and is the output of the syntax graph calculated by using the sentence vector H and the sentence graph G obtained by using the word segmentation tool space.
The embodiment of the invention re-uses the gating mechanism of formula (5) to transfer the collaborative graph information to the graph representation,W y W x b y b x is the weight of the neural network,
Figure SMS_11
representing the inner product operation of the matrix, total of equation (5) is the model output.
Figure SMS_12
（5）
The embodiment of the invention uses an attention mechanism to gather emotion words in total to obtain information important for specific target emotion classification, and the invention can calculate the attention output by using the following formulas (6), (7) and (8).
Figure SMS_13
（6） />
Figure SMS_14
（7）
Figure SMS_15
（8）
H in equation (6) represents the output of equation 5,harepresenting a vector representation of a particular object, in equation (6)W 2 Andb 2 is the weight of the neural network, exp in equation (7) represents an exponential function.
FIG. 3 illustrates the reinforcement learning portion of the LSTM model of an embodiment of the invention. To further map word segments to emotion, the present invention uses a method similar to human reading articles, and from left to right the present invention treats sentences as a collection of word segments. The task of reinforcement learning is the current word segment selection action, in the invention, an agent is defined as the hidden state of a word segment action selection layer, the environment is virtual, the invention firstly evaluates the state, then selects the action of the word segment, the action refers to which emotion is selected, and the invention adopts a strategy gradient method for training.
For the state evaluation layer, the invention selects a layer of LSTM (long short term neural network), the input of the invention is word vector and the state of the intelligent agent, the state of the intelligent agent is initialized at random at first, the state evaluation layer outputs to the action selection layer, the calculation of the state evaluation layer is shown as a formula (9), and please refer to fig. 3.
Hidden，cell1＝LSTM(E，[state，cell1]) (9)
Where E represents the embedding of a sentence,hidden，cell1represents the hidden state and cells of LSTM, respectively;
the action selection layer adopts three layersLSTMUsing different actionsLSTMThe action selection layer calculates the action with the highest probability on the input from the state evaluation layer, obtains the most possible emotion selection of the words in one batch, selects the emotion with the highest frequency in the batch, inputs the emotion as emotion action selection into the corresponding LSTM model, outputs the state of the intelligent agent, and calculates a formula as formula (10).
Figure SMS_16
（10）
In the method, in the process of the invention,state，cell2indicating the hidden state and cells of LSTM, respectively.
The invention takes the state output of the agent at the last moment as the whole phrase, namely the sentence representation, and maps the hidden state to the output probability distribution through a softmax output layer.
Figure SMS_17
（11）
The method calculates the loss r by using a cross entropy method, and takes the loss r as the rewards of the intelligent agent;
Figure SMS_18
（12）
the invention uses strategy gradient to carry out error back transmission and parameter update.
Figure SMS_19
（13）
Figure SMS_20
（14）
Where J is the objective function of the strategy gradient, N is the number of sentences, B is the number of training batches, and θ represents the parameter.
The invention adopts a multitask learning method to train, the loss comprises that the collaborative graph model loss is loss_c, the attention model loss is loss_a, and the reinforcement learning loss is loss_r, thereby training the neural network.
Figure SMS_21
（15）
In the specific implementation, the method is realized through the processing steps of data preprocessing, word embedding, outputting of output specifications after model training, index evaluation and the like.
(1) Data preprocessing
The data uses text format, and since the data units are built on a single sentence, space is used for sentence and word segmentation. Because the use is tree structure network, the sentence is needed to be depended and analyzed, and the invention adopts an industrial natural language processing tool to analyze.
The identified object is to distinguish with special symbols in the sentence. Summarizing different words after data word segmentation, establishing a hash table to construct a dictionary, defining keys and values in the hash table according to the word occurrence sequence, wherein the keys correspond to the words, the values correspond to the positions of the words in the dictionary, and the dictionary is used for mapping input data from text to numbers. Referring to fig. 4, the original dataset is shown: fig. 5 shows the data after sentence analysis using spacy dependency parser.
(2) Word embedding
The experiment adopts a pre-trained word model Glove, the type of the word model Glove is gloove.42B.300d, words in a dictionary are mapped into the word model Glove to obtain corresponding vectors, each word is represented by using a 300-dimensional vector, and the initialization of the vectors adopts a uniform distribution method. Referring to FIG. 6, a partially pre-trained 300-dimensional word vector of size 4907255KB is shown:
(3) Training the model to generate normalized output information, and evaluating the index
After the word vector and the sentence vector containing the semantics are obtained, respectively placing the word vector and the sentence vector into a model to realize multi-context learning and multi-word segment learning, and then checking the effect of the recognition target of the model on a verification set, as shown in table 1; the super parameters used are shown in table 2; the loss and number of iterations of training are shown in fig. 7:
Figure SMS_22
TABLE 1
Figure SMS_23
TABLE 2
Through table 1, it was found that for emotion analysis of a specific target, an accuracy of 83.13% and an F1 value of 76.32 could be achieved on the resuurant dataset, an accuracy of 77.14% and an F1 value of 72.41% could be achieved on the Laptop dataset, the effect of removing some modules on the results of the present invention was also shown in table one, and after it was found that the relevant modules were removed, the experimental index of the method of the present invention was reduced, proving the effect of the method of the present invention. The training model is known to have good training speed on the training set through fig. 7. The optimizer Adam, at a learning rate of 0.001, the weight decay of 0.001 gives the model good performance.
Finally, the method not only can automatically analyze the emotion of the specific target in the SemEval 2014 Task 4, but also has very important significance compared with the training speed of the method compared with the speed block of the CDT and RGAT graph convolution network, and the specific target emotion analysis model based on the multi-context and multi-word-segment GCN.
The specific target recognition model based on the multi-context and multi-word segment GCN provided by the invention is used for experiments on two data sets on a Task 4 of the SemEval 2014, wherein the two data sets are a resuurant data set and a Laptop data set respectively, the resuurant data set is comment data about a Restaurant, the Laptop data set is comment data about a notebook computer, and the emotion information can be automatically recognized by the multi-context and multi-word segment learning based method provided by the invention, so that the accuracy of the resuurant data set and the Laptop data set can reach 83.13% and 77.14% respectively.
While the fundamental and principal features of the invention and advantages of the invention have been shown and described, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing exemplary embodiments, but may be embodied in other specific forms without departing from the spirit or essential characteristics thereof;
the present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a separate embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to the embodiments described in detail below, and that the embodiments described in the examples may be combined as appropriate to form other embodiments that will be apparent to those skilled in the art.

Claims (1)

1. The specific target emotion recognition method based on the multi-context and multi-word segment graph convolution network is characterized by being realized based on a specific target emotion recognition model, and the processing steps of the model are as follows:
converting the target sentence into a vector, inputting the vector into an LSTM neural network model for reinforcement learning, and outputting the sentence vector;
carrying out graph convolution operation on the local collaborative graph obtained from the global collaborative graph and the sentence vector to output a first graph convolution output result;
obtaining a sentence pattern on a target sentence by using a word segmentation tool space, wherein the sentence pattern is a structural diagram of the sentence of which the Spacy dependency parser is used for analyzing the sentence dependency relationship, and performing a graph convolution operation on the sentence pattern and the sentence vector to output a second graph convolution output result;
performing inner product operation of a matrix on the first graph rolling output result and the second graph rolling output result by using a gating mechanism, and transferring the collaborative graph information to a graph representation;
the attention mechanism is used for gathering target words in the output result of the gating mechanism, and target information for classifying target emotion is obtained;
the global synergy graph is a large matrix and is used for counting the occurrence frequency of different two words in a plurality of sentences; the local synergy graph is a small matrix, and each element value is derived from the global synergy graph;
only considering the relation among open class words containing more emotion components when counting the occurrence frequency of two different words in a plurality of sentences in the global collaborative graph;
the global collaborative map processes sentences, maps words into corresponding part-of-speech tags, and is divided into an open class, a closed class and other three types, wherein the open class comprises adjectives and emotion words including adverbs;
calculating the occurrence frequency of all words in a plurality of sentence contexts by using a statistical method, forming a global collaborative graph among all words, then searching the weight of the corresponding position of a specific target from the global collaborative graph by only considering the specific target and the open class words according to the specific sentences and the specific targets, and putting the weight into the position of the specific target by using the obtained weight to obtain a local collaborative graph of the current sentence;
when the LSTM neural network model is used for reinforcement learning, sentences are regarded as a set of word segments from left to right, an agent is defined as a hidden state of a word segment action selection layer, the state is firstly evaluated, and then the action of the word segment is selected, wherein the action refers to the selection of a certain emotion; training by adopting a strategy gradient method; for the state evaluation layer, selecting a layer of LSTM neural network, inputting word vectors and states of the intelligent agents, randomly initializing the states of the intelligent agents during initialization, and outputting the state evaluation layer to the action selection layer; the action selection layer adopts three layers of LSTM neural networks, different LSTM neural networks are adopted for different actions, the action selection layer calculates the action with the highest probability on the input from the state evaluation layer, the most probable emotion selection of a batch of words is obtained, the emotion with the highest frequency in the batch is selected as emotion action selection, the emotion is input into the corresponding LSTM neural network, and the state of an intelligent body is output; the last moment of agent state output is used as the expression of the whole phrase, and the hidden state is mapped to the output probability distribution through a softmax output layer.
CN202210245787.2A 2022-03-14 2022-03-14 Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network Active CN114722808B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210245787.2A CN114722808B (en) 2022-03-14 2022-03-14 Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210245787.2A CN114722808B (en) 2022-03-14 2022-03-14 Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network

Publications (2)

Publication Number Publication Date
CN114722808A CN114722808A (en) 2022-07-08
CN114722808B true CN114722808B (en) 2023-05-23

Family

ID=82237566

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210245787.2A Active CN114722808B (en) 2022-03-14 2022-03-14 Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network

Country Status (1)

Country Link
CN (1) CN114722808B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112001185A (en) * 2020-08-26 2020-11-27 重庆理工大学 Emotion classification method combining Chinese syntax and graph convolution neural network

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11481418B2 (en) * 2020-01-02 2022-10-25 International Business Machines Corporation Natural question generation via reinforcement learning based graph-to-sequence model
CN111339255B (en) * 2020-02-26 2023-04-18 腾讯科技（深圳）有限公司 Target emotion analysis method, model training method, medium, and device
CN112949684B (en) * 2021-01-28 2022-07-29 天津大学 Multimodal dialogue emotion information detection method based on reinforcement learning framework
CN112686056B (en) * 2021-03-22 2021-07-06 华南师范大学 Emotion classification method

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112001185A (en) * 2020-08-26 2020-11-27 重庆理工大学 Emotion classification method combining Chinese syntax and graph convolution neural network

Also Published As

Publication number Publication date
CN114722808A (en) 2022-07-08

Similar Documents

Publication Publication Date Title
CN111368996B (en) Retraining projection network capable of transmitting natural language representation
Tang et al. Sentiment embeddings with applications to sentiment analysis
CN112800776B (en) Bidirectional GRU relation extraction data processing method, system, terminal and medium
CN113535984A (en) Attention mechanism-based knowledge graph relation prediction method and device
Bekamiri et al. Patentsberta: A deep nlp based hybrid model for patent distance and classification using augmented sbert
CN111292848A (en) Bayesian estimation-based medical knowledge map assisted reasoning method
Tammina et al. Sentiment analysis on customer reviews using convolutional neural network
Tran et al. Analyzing knowledge graph embedding methods from a multi-embedding interaction perspective
CN113297351A (en) Text data labeling method and device, electronic equipment and storage medium
Dangi et al. An efficient model for sentiment analysis using artificial rabbits optimized vector functional link network
CN115081446B (en) Text matching method based on dynamic multi-mask and enhanced countermeasure
Yuwono et al. Automated scoring of chatbot responses in conversational dialogue
CN108875024B (en) Text classification method and system, readable storage medium and electronic equipment
Ruskanda et al. Simple sentiment analysis ansatz for sentiment classification in quantum natural language processing
CN116720519B (en) Seedling medicine named entity identification method
CN110334204B (en) Exercise similarity calculation recommendation method based on user records
CN114722808B (en) Specific target emotion recognition method based on multi-context and multi-word segment graph convolution network
Akhter et al. COVID-19 Fake News Detection using Deep Learning Model
WO2024007119A1 (en) Training method for text processing model, and text processing method and device
CN111859910B (en) Word feature representation method for semantic role recognition and fusing position information
CN114997155A (en) Fact verification method and device based on table retrieval and entity graph reasoning
Hong et al. High-quality noise detection for knowledge graph embedding with rule-based triple confidence
Mahimaidoss et al. Emotion Identification in Twitter Using Deep Learning Based Methodology
Ng et al. Sentiment analysis using learning-based approaches: A comparative study
Schwartz et al. Machine learning methods

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