CN114840665A - Rumor detection method and device based on emotion analysis and related medium - Google Patents
Rumor detection method and device based on emotion analysis and related medium Download PDFInfo
- Publication number
- CN114840665A CN114840665A CN202210419377.5A CN202210419377A CN114840665A CN 114840665 A CN114840665 A CN 114840665A CN 202210419377 A CN202210419377 A CN 202210419377A CN 114840665 A CN114840665 A CN 114840665A
- Authority
- CN
- China
- Prior art keywords
- rumor
- text
- comment
- emotion
- emotional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000008451 emotion Effects 0.000 title claims abstract description 172
- 238000001514 detection method Methods 0.000 title claims abstract description 76
- 238000004458 analytical method Methods 0.000 title claims abstract description 38
- 230000002996 emotional effect Effects 0.000 claims abstract description 122
- 238000000034 method Methods 0.000 claims abstract description 29
- 230000007246 mechanism Effects 0.000 claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 230000036651 mood Effects 0.000 claims description 2
- 230000001953 sensory effect Effects 0.000 claims 1
- 238000002474 experimental method Methods 0.000 description 10
- 238000011176 pooling Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000009977 dual effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000002679 ablation Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 3
- 238000013145 classification model Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000012552 review Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000002349 favourable effect Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000019580 granularity Nutrition 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation 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
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a rumor detection method, a rumor detection device and a related medium based on emotion analysis, wherein the method comprises the following steps: obtaining comment texts of rumors to be detected, and extracting emotional features of the comment texts from the comment texts through an emotional dictionary; forward learning and backward learning are respectively carried out on the emotional features by utilizing a Bi-GRU network, and comment emotional feature representation of the comment text is obtained; acquiring a rumor text, and respectively performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain a rumor characteristic representation of the rumor text; and performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism, and performing rumor detection on the comment text according to correlation learning results. According to the method, relevance learning is carried out on the comment emotion characteristic representation and the rumor characteristic representation, so that reasonable explanation is generated in terms of emotion, and therefore the rumor detection accuracy is improved.
Description
Technical Field
The invention relates to the technical field of computer software, in particular to a rumor detection method and device based on emotion analysis and a related medium.
Background
Rumor detection is generally classified into rumor content-based, social context-based, and mixed-feature-based methods. The rumor content can be divided into two aspects of text and vision, and the text aspect refers to extraction of text features according to the language style, writing style and the like of the rumor. The language style based approach detects rumors by capturing differences in writing styles because rumors tend to prefer capitalized words over proper nouns. The prior art centers, by analyzing the data set of the fake news web, found that people express their emotions or opinions of fake news through social media posts, with a greater proportion of neutral returns than positive and negative returns for real news. Statistical findings on FakeNewsNet (false news data set) indicate that the emotional polarity of reviews under false news is greater than under real news. Therefore, the potential emotion of the user is integrated into an end-to-end deep embedded framework to detect false news. And three neural networks are used to process news images, news text and user profiles, while a countermeasure mechanism is introduced to maintain semantic similarity and enhance the consistency of representation between text and images. Finally, the user's emotion is modeled and integrated into the proposed framework. One novelty of this work is the use of adversarial learning to discover semantic correlations between different patterns in news content. The resulting false news detection system is superior to other classical and deep learning based classifiers. The above experiments show that the emotion analysis has the greatest impact on the system performance. Furthermore, visual features may extract features from a video or picture.
Social context-based detection methods can be generally classified into user-based and network-based. The user-based approach models the characteristics of rumor publishers and forwarding users. The characteristics mainly comprise user gender, fan number and user configuration. Network-based methods address rumor detection through forwarding or structure of interest features in social networks. Includes a tree-shaped cyclic neural network for embedding the learning rumor propagation structure and learning the embedded rumor propagation by using the digraph neural network model.
Mixed feature-based methods tend to merge multimodal or multiplex features for rumor detection. I.e., learning the correlation between rumors and comments using the common attention mechanism and providing explanation using the common attention mechanism. There is also a Graph-aware collaborative Attention network model (Graph-aware Co-Attention Networks) based rumor detection using rumor text, forwarding user sequences and user interaction features, respectively, with two collaborative Attention mechanisms to generate interpretations by highlighting suspicious forwarders and the utterances they are interested in. Meanwhile, the prior art also considers that most of the existing research on rumor detection is based on emotional characteristics of content transmitted by a publisher, but rarely pays attention to emotion caused in crowds, namely emotional characteristics of comments, and the emotion, social emotion and emotion of the publisher are known through an emotion dictionary and are used as supplementary characteristics of a false news detector, so that a good effect is achieved. The prior art also considers that rumor detection usually needs complex reasoning capability, so a fine-grained reasoning model with human information processing capability is provided, which not only can better reflect the logic process of human thinking, but also can realize fine-grained modeling of subtle clues to improve accuracy and interpretability.
In summary, in the current prior art, while the accuracy of rumor detection is improved, it can be explained that a rumor is "why" is detected as a rumor, but in the existing interpretable rumor detection model, the problem of interpretable rumor detection from the emotional point of view is ignored.
Disclosure of Invention
The embodiment of the invention provides a rumor detection method, a rumor detection device, computer equipment and a storage medium based on emotion analysis, and aims to improve the rumor detection precision by combining emotion analysis.
In a first aspect, an embodiment of the present invention provides a rumor detection method based on emotion analysis, including:
obtaining comment texts of rumors to be detected, and extracting emotional features of the comment texts from the comment texts through an emotional dictionary;
forward learning and backward learning are respectively carried out on the emotional features by utilizing a Bi-GRU network, and comment emotional feature representation of the comment text is obtained;
acquiring a rumor text, and respectively performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain a rumor characteristic representation of the rumor text;
and performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism, and performing rumor detection on the comment text according to correlation learning results.
In a second aspect, an embodiment of the present invention provides a rumor detection apparatus based on emotion analysis, including:
the comment text acquisition unit is used for acquiring comment texts of rumors to be detected and extracting emotional features of the comment texts from the comment texts through an emotional dictionary;
the first learning unit is used for respectively carrying out forward learning and backward learning on the emotional features by utilizing a Bi-GRU network to obtain comment emotional feature representation of the comment text;
the system comprises a rumor text acquisition unit, a rumor text analysis unit and a rumor text analysis unit, wherein the rumor text acquisition unit is used for acquiring a rumor text, and performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain rumor feature representation of the rumor text;
and the rumor detection unit is used for performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism and performing rumor detection on the comment text according to a correlation learning result.
In a third aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor, when executing the computer program, implements the rumor detection method based on emotion analysis according to the first aspect.
In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the rumor detection method based on emotion analysis according to the first aspect.
The embodiment of the invention provides a rumor detection method, a rumor detection device, computer equipment and a storage medium based on emotion analysis, wherein the method comprises the following steps: obtaining comment texts of rumors to be detected, and extracting emotional features of the comment texts from the comment texts through an emotional dictionary; forward learning and backward learning are respectively carried out on the emotional features by utilizing a Bi-GRU network, and comment emotional feature representation of the comment text is obtained; acquiring a rumor text, and respectively performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain a rumor characteristic representation of the rumor text; and performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism, and performing rumor detection on the comment text according to correlation learning results. According to the embodiment of the invention, relevance learning is carried out on the comment emotion characteristic representation and the rumor characteristic representation, so that reasonable explanation is generated in terms of emotion angle, and the rumor detection precision is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic flow chart of a rumor detection method based on emotion analysis according to an embodiment of the present invention;
fig. 2 is a schematic sub-flow chart of a rumor detection method based on emotion analysis according to an embodiment of the present invention;
FIG. 3 is a schematic view of another sub-process of a rumor detection method based on emotion analysis according to an embodiment of the present invention;
FIG. 4 is a diagram of a model framework of a rumor detection method based on emotion analysis according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating experimental results of a rumor detection method based on emotion analysis according to an embodiment of the present invention;
fig. 6 is a schematic block diagram of a rumor detection apparatus based on emotion analysis according to an embodiment of the present invention;
FIG. 7 is a block diagram illustrating a rumor detection apparatus based on emotion analysis according to an embodiment of the present invention;
fig. 8 is another sub-schematic block diagram of a rumor detection apparatus based on emotion analysis according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be further understood that the term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
Referring to fig. 1, fig. 1 is a diagram illustrating a rumor detection method based on emotion analysis according to an embodiment of the present invention, which specifically includes: steps S101 to S104.
S101, obtaining a comment text of a rumor to be detected, and extracting emotional features of the comment text from the comment text through an emotional dictionary;
s102, forward learning and backward learning are respectively carried out on the emotional features by utilizing a Bi-GRU network, and comment emotional feature representation of the comment text is obtained;
s103, acquiring a rumor text, and performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain a rumor feature representation of the rumor text;
s104, performing relevance learning on the comment emotion characteristic representation and the rumor characteristic representation through a collaborative attention mechanism, and performing rumor detection on the comment text according to a relevance learning result.
In this embodiment, Bi-directional learning (i.e., forward learning and backward learning) is performed on the comment text and the rumor text through a Bi-GRU network to obtain corresponding comment emotion feature representation and rumor feature representation, so that a Co-attentive force mechanism (Co-attentionsecuisman) is used to perform correlation learning on the comment emotion feature representation and the rumor feature representation, thereby obtaining a weight ratio of the comment emotion feature representation and the rumor feature representation, and performing rumor detection on the comment text.
With reference to fig. 4, in this embodiment, in combination with emotion analysis of rumor comments, according to feature points at which Co-Attention Mechanism can learn two feature correlations, an interpretable rumor detection model based on emotion analysis and common Attention Mechanism is proposed, which obtains the correlation between rumor content and comments by using Co-Attention, gives higher weight to words and comments which are favorable for rumor detection, extracts corresponding text features and emotion features, and provides interpretation from an emotion perspective by using the weight of Co-Attention. Specifically, a Bi-directional gate recycling unit (Bi-GRU) is used for solving the problem of information loss of a long-sequence text, fully learning context information and obtaining rumor text characteristics; secondly, obtaining the emotional characteristics of the comments through various emotional dictionaries, and obtaining context information of the comments of the user by using the Bi-GRU; and finally, assigning different weights to rumor detection through the correlation between the Co-Attention learning comment emotional characteristics and the rumor text, and providing explanation by using the Co-Attention weights.
In the embodiment, relevance learning is performed on the comment emotion feature representation and the rumor feature representation, so that reasonable explanation is generated in terms of emotion, and therefore the rumor detection accuracy is improved.
In one embodiment, the step S101 includes:
and respectively extracting emotion vocabulary, emotion intensity, emotion value and auxiliary characteristics from the comment text through an emotion dictionary, and comprehensively using the emotion vocabulary, the emotion intensity, the emotion value and the auxiliary characteristics as the emotion characteristics.
Further, in an embodiment, as shown in fig. 2, the extracting, by an emotion dictionary, emotional vocabulary, emotional intensity, emotional value, and auxiliary features in the comment text, respectively, and integrating as the emotional features includes: steps S201 to S207.
S201, selecting the negative words and the degree adverbs in the comment text, and calculating scores of the negative words and the degree adverbs according to the following formula:
s(c i )＝D(c i )*neg(c i ,w)*deg(c i ,w)
wherein D is an emotion dictionary, c i For words in text, w denotes the context range, neg (c) i ,w)，deg(c i W) are the corresponding negative word and degree adverb values, respectively, where:
S202selecting the emotion words in the comment text, and calculating the emotion value emo of the comment text according to the following formula by combining the negative words and the degree adverbs z ：
Wherein L is the length of the comment text;
s203, calculating the score of the emotion vocabulary according to the following formula:
s204, connecting all the emotion vocabulary scores in a connection mode according to the following formula to obtain an emotion vocabulary total score emo y ：
S205, calculating the emotion intensity corresponding to each emotion vocabulary according to the following formula:
in the formula, int (c) i ) Indicates the degree of intensiveness of the ith emotional vocabulary if c i In the dictionary, int (c) is calculated from the emotion dictionary i ) Otherwise, the value is 0;
s206, according to the following formula, the emotion intensity characteristics emo are obtained by connecting the emotion intensity corresponding to each emotion vocabulary d ：
S207, passing through the assistant feature emo f Capturing emoticons, punctuation marks and emotions in the comment textAnd connecting the emotion value, the total score of the emotion vocabulary, the emotion intensity characteristic and the auxiliary characteristic of the comment text according to the following formula to obtain the emotion characteristic x:
in this embodiment, an emotion vocabulary, emotion intensity, emotion value, and auxiliary features are respectively extracted from the comment text through an emotion dictionary, and are comprehensively used as the emotional features of the comment text.
The individual scores of the words of each comment in the comment text in the emotion dictionary are firstly calculated, and the values of the negative words and the degree words are matched and calculated by using the existing emotion dictionary.
s(c i )＝D(c i )*neg(c i ,w)*deg(c i ,w)
Wherein D is an emotion dictionary, c i For words in text, w denotes the context range, neg (c) i ,w)，deg(c i W) are respectively the corresponding negative word and degree adverb values, c) i-w A vocabulary representing the upper and lower ranges of the core word, wherein:
calculating to obtain an emotion value of a certain type in the text according to the emotion words, the degree adverbs and the negative words in the comment, namely calculating to obtain the emotion value of the comment text:
next, the features of a certain emotion of the comment text are calculated:
where T denotes comment text, q i Representing a certain mood;
all the emotion vocabulary extraction features are obtained in a connected mode:
according to the dictionary and the comments corresponding to each emotion, the emotional intensity of the emotion is calculated first,
where T denotes comment text, q denotes each emotion dictionary, int (c) i ) Indicates the strength of the vocabulary if c i In the dictionary, int (c) is calculated from the emotion dictionary i ) Otherwise, it is 0. Emotional intensity characteristics are obtained by linking the intensity of various emotions:
finally, the emoticons, punctuation marks, emotional words, human pronouns and the like are captured through the auxiliary features, especially in comments, the emoticons are mostly used for replacing characters, and a plurality of emotion dictionaries are used for obtaining the auxiliary features emo f Then, connecting the features to obtain the emotional features of each comment in the comment text as follows:
the emotional characteristics of the comments comprise emotional vocabulary, emotional intensity, emotional value and auxiliary characteristics, wherein each part of the emotional characteristics is connected with the upper and lower characteristics in a certain way, and the emotional intensity can be influenced by the size of the emotional value.
In one embodiment, the step S102 includes:
forward learning and backward learning are respectively carried out on the emotional characteristics according to the following formula:
in the formula,
for the forward hidden state of the affective features,
for the hidden state of the emotional feature,
is the emotional characteristic of the ith comment in the comment text, i is the single emotional characteristic in the comment,
for the last forward hidden state of the packet,
the state is the last backward hidden state, and m is the number of comments;
connecting the forward hiding state of the emotional feature with the backward hiding state of the emotional feature to obtain comment emotional feature representation of the comment text
In the present embodiment, for the obtained emotional characteristics, Bi-GRU is used in consideration of the correlation between the emotional characteristics and the upper and lower emotionsThe network learns the representation of the emotional features. Specifically, the emotional characteristic of the comment text is set as X ═ X 1 ,x 2 ,x 3 ,...,x M ]，x i The emotional characteristics of the ith comment in the comments are respectively learned in the forward direction and the backward direction through Bi-GRU:
forward hidden state over connection
And a backward hidden state
Obtaining the comment emotional characteristic representation of the comment text
In one embodiment, the step S103 includes:
forward learning and backward learning are respectively performed on the rumor text by using a Bi-GRU network according to the following formula:
in the formula,
for the forward hidden state of the rumor text,
for the backward hidden state of the rumor text,
the j is the j word in the rumor text, and j represents a single word in the rumor text;
connecting the forward hidden state of the rumor text with the backward hidden state of the rumor text to obtain a rumor feature representation of the rumor text
In this example, since RNN is theoretically able to capture long-term dependence, in practice, old memory will disappear as the sequence gets longer. To capture the long term dependence of RNN, Bi-GRU is used to ensure more persistent memory. The vocabulary in the text is often associated with the context and has strong bidirectional semantic dependence, so that the reverse order processing is very necessary, and the Bi-directional Bi-GRU is adopted to model from two directions of the vocabulary. A rumor text is preprocessed, and the rumor s composed of n words is ═ w 1 ,w 2 ,w 3 ,...,w n ]Then, we can get:
then forward hidden state by concatenating rumor text
Backward hidden states of rumor text
Rumor characterization to rumor text
In one embodiment, as shown in fig. 3, the step S104 includes: steps S301 to S304.
S301, constructing a similar matrix F according to the following formula:
F＝tanh(E T W es S)
in the formula,
is a parameter matrix which can be learnt, E is comment emotional characteristic representation, and E ═ E 1 ,e 2 ,...,e M ]S is rumor character, S ═ S 1 ,s 2 ,...,s N ]M, N, respectively, indicating the number of sentences in the comment text and the number of sentences in the rumor text;
s302, cooperatively mapping the comment emotion characteristic representation and the rumor characteristic representation by using the similarity matrix according to the following formula:
H e ＝tanh(W e E+(W s S)F T )
in the formula,
and
are all a matrix of parameters that can be learned,
s303, learning attention weights of the comment emotion characteristic expression and the rumor characteristic expression respectively according to the following formula:
in the formula,
attention probability vectors for each vocabulary in the rumor text and each emotional feature of the comment text respectivelyThe attention probability vector of the review is,
are all learnable weights;
s304, respectively generating the comment emotional characteristic representations in a weighted sum mode according to the following formula
Attention vector of sum rumor feature
Further, in an embodiment, the step S104 further includes:
rumor detection was performed on the comment text using softmax function according to the following formula:
in the formula,
the rumor test results.
In this embodiment, since the user comment may contain relevant information explaining an important aspect of why a piece of news is fake, they are small in information amount and loud in noise. Therefore, it is weak to use the rumors themselves to detect the rumors and explain the truth of the rumors, but the comments have rich emotions, so that the emotional characteristics are more prominent than the semantic characteristics, which is more beneficial to the rumors detection, and the emotions can reflect the reason of the truth of the rumors. Therefore, the embodiment learns the relevance of the emotion and the rumor in the comment text through the cooperative attention mechanism, and utilizes the attention weight of the emotion and the vocabulary in the rumor to perform rumor detection and rumor explanation.
The rumor is expressed as: s ═ S 1 ,s 2 ,...,s N ]The comment emotional characteristics are expressed as: e ═ E 1 ,e 2 ,...,e M ]。
First, a similarity matrix is calculated
F＝tanh(E T W es S) in which
Is a learnable parameter matrix. With the similarity matrix as a feature, the collaborative mapping of rumor text and comment sentiment can be learned:
H s ＝tanh(W s S+(W e E)F)
H e ＝tanh(W e E+(W s S)F T )
wherein,
all the parameters are learnable parameter matrixes, the attention weight of rumor texts and comment emotional characteristics can be learnt, F is a similar matrix, H s ,H e Attention mapping representing rumor text sentences and comment text, respectively:
wherein
Respectively, the attention probability vector for each word in the rumor and for each comment in the emotional character of the comment.
Are all learnableAnd (4) weighting. Finally, a rumor and attention vector commenting on sentiment are generated by weighted sum.
Then
Respectively, rumor text and comment emotional characteristics learned through a cooperative attention mechanism.
And then, performing prediction judgment on the spliced rumor text characteristics and the emotional characteristics of the comments through a softmax function.
In a specific embodiment, the rumor detection method based on emotion analysis provided in this embodiment is tested. The experiment used one Chinese dataset Weibo-20 and two English datasets rumor to test Twitter15 and Twitter 16. Weibo-20 is based on Weibo-16, deduplication is performed through a clustering algorithm, error information identified by a microblog community management center is added, and a new data set comprising information such as rumors, comments and labels is formed. The Twitter dataset (i.e., Twitter15 and Twitter16) selects "true" and "false" tag data, both of which contain rumor content, user reviews, and corresponding forward user sequences. Rumor content and user comments were taken as input. The statistics of the data set are shown in table 1.
TABLE 1
(1) Experimental setup
Adam is used for updating parameters in the experiment, the initial learning rate is 0.001, and a self-adaptive decreasing strategy is adopted for updating the learning rate. The english word vector dimension is set to 100 and the chinese word vector dimension is set to 300.
To highlight the advancement of this embodiment, experiments are performed on the two data sets, and the experimental results are compared and analyzed with the reference model. Wherein the reference model includes:
RNN: an RNN-based approach models social context information as a time series of variable length for continuous representation of learning rumors.
text-CNN: a text classification model based on a convolutional neural network utilizes a plurality of convolutional filters to capture text features of different granularities.
HAN: a hierarchical attention network based document classification model utilizes word-level attention and sentence-level attention to learn news content representations.
dEFEND: a false news detection model based on a collaborative attention mechanism learns the correlation between news content and user comments.
GCAN: a false news detection model based on double common attention can learn correlation between source short text tweets, and takes news content and a forward reply sequence as input.
Dual Emotion: a false news detection model based on dual emotional characteristics is characterized in that rumor emotional characteristics, comment emotional characteristics and emotional characteristic differences are used as supplementary characteristics of a false news detector.
In the experiment, according to the actual conditions of the Twitter15 and Twitter16 data sets, the number of the text sentences of the dEFEND model reradial is 1, the length of the text sentences is 32, and 12 and 9 comment sentences are respectively selected; the GCAN model quotes the heavy experimental results of the original paper; for fair comparison, the Dual annotation model extracts text features by using bert (bidirectional encoded representation from transformations), and selects 12 and 9 comments to extract emotional features respectively; in the model presented herein, the number of rumor sentences is 1, the length is 32, and the number of comments is 12 and 9 respectively.
(2) Results of the experiment
70% of the data were randomly selected for training and 30% were used for testing, with a 1: 1 sample ratio in each data set. Common evaluation indexes are set as follows: Accuracy-Accuracy, Accuracy-Precision, Recall-Recall, and F1. The results of the experiments on Twitter15 and Twitter16 are shown in tables 2 and 3.
| Method | accuracy | precision | recall | F1 |
| RNN | 0.720 | 0.716 | 0.715 | 0.713 |
| Text-CNN | 0.756 | 0.732 | 0.731 | 0.730 |
| HAN | 0.811 | 0.814 | 0.813 | 0.811 |
| dEFEND | 0.845 | 0.845 | 0.846 | 0.845 |
| GCAN | 0.876 | 0.825 | 0.829 | 0.825 |
| DualEmotion | 0.851 | 0.851 | 0.851 | 0.851 |
| The invention | 0.901 | 0.896 | 0.897 | 0.898 |
TABLE 2
| Method | accuracy | precision | recall | F1 |
| RNN | 0.653 | 0.652 | 0.653 | 0.653 |
| Text-CNN | 0.674 | 0.672 | 0.673 | 0.677 |
| HAN | 0.723 | 0.712 | 0.712 | 0.716 |
| dEFEND | 0.743 | 0.756 | 0.774 | 0.741 |
| GCAN | 0.908 | 0.763 | 0.759 | 0.759 |
| DualEmotion | 0.812 | 0.821 | 0.817 | 0.812 |
| The invention | 0.892 | 0.882 | 0.883 | 0.889 |
TABLE 3
It can be found from tables 2 and 3 that the method based on the mixed features is obviously superior to the method based on the text content features, which indicates that the method based on the mixed features is more favorable for rumor detection features to have better effect. In the method based on mixed features, the model provided by the embodiment is superior to a dEFEND model, which shows that Co-Attention is also adopted to extract feature correlation, and the correlation model of texts and emotions is superior to the model of texts and comment texts; the model provided by the embodiment is superior to a Dual Emotion model, which shows that text and comment Emotion are used as characteristics of a rumor detector, but the Co-Attention extracted characteristic correlation is more advantageous; compared with the GCAN model, the model provided by the embodiment adopts Co-orientation once, uses fewer features and obtains the optimal result, which shows that the selection of more effective features is the key for detecting rumors.
On the two data sets of Twitter15 and Twitter16, the model provided by the embodiment is significantly better than other models in each index, the performance of the model on Twitter15 is improved by about 5%, and the performance of the model on Twitter16 is improved by about 7%. This fully demonstrates the effectiveness of the method proposed by this embodiment, and the effect of the individual modules on the model performance will be elaborated in the ablation experimental section.
(3) Ablation experiment
To explore the effect of the Co-Attention mechanism and emotional characteristics on model performance, the results of ablation experiments on the Twitter data set are shown in FIG. 5. First, the Co-Attention is replaced by a pooling layer, and the splicing characteristics are sent to a full-connection layer classifier. Secondly, replacing the comment emotional features with the semantic features of the comment, namely the dEFEND model to explore the influence of the emotional features on the model. The maximum pooling (Max-pooling) is denoted by "M", the average pooling (Avg-pooling) is denoted by "A", the dEFEND model is denoted by "D", and the model proposed in this example is denoted by "O", respectively. Specifically, as shown in tables 4 and 5:
| Method | accuracy | precision | Recall | F1 |
| Max-pooling | 0.892 | 0.882 | 0.882 | 0.889 |
| Avg-pooling | 0.878 | 0.864 | 0.872 | 0.866 |
| No-emotion | 0.845 | 0.845 | 0.846 | 0.845 |
| the invention | 0.901 | 0.896 | 0.897 | 0.898 |
TABLE 4
| Method | accuracy | precision | recall | F1 |
| Max-pooling | 0.875 | 0.863 | 0.899 | 0.850 |
| Avg-pooling | 0.854 | 0.815 | 0.844 | 0.805 |
| No-emotion | 0.743 | 0.756 | 0.774 | 0.741 |
| The invention | 0.892 | 0.882 | 0.883 | 0.889 |
TABLE 5
Experimental results show that the Co-Attention can effectively highlight important features through interaction of tweets, comment contents and emotional features, and further improve the performance of the model. Experiments show that the posts and the features generated do not interact with emotional features, so that the performance of the classification model is reduced, and the effectiveness of focusing on the important features of the tweet by using emotional information is proved.
Fig. 6 is a schematic block diagram of a rumor detection apparatus 600 based on emotion analysis according to an embodiment of the present invention, where the apparatus 600 includes:
the comment text acquisition unit 601 is used for acquiring a comment text of a rumor to be detected and extracting emotional features of the comment text from the comment text through an emotional dictionary;
a first learning unit 602, configured to perform forward learning and backward learning on the emotion features by using a Bi-GRU network, respectively, to obtain comment emotion feature representations of the comment text;
a rumor text obtaining unit 603, configured to obtain a rumor text, and perform forward learning and backward learning on the rumor text by using a Bi-GRU network, respectively, to obtain a rumor feature representation of the rumor text;
a rumor detecting unit 604, configured to perform correlation learning on the comment emotion feature representation and the rumor feature representation through a collaborative attention mechanism, and perform rumor detection on the comment text according to a correlation learning result.
In one embodiment, the comment text acquiring unit 601 includes:
and the comprehensive setting unit is used for extracting emotion words, emotion intensity, emotion values and auxiliary features from the comment text through an emotion dictionary and comprehensively taking the emotion words, the emotion intensity, the emotion values and the auxiliary features as the emotion features.
In one embodiment, as shown in fig. 7, the integrated setting unit includes:
a first selecting unit 701, configured to select a negative word and a degree adverb in the comment text, and calculate scores of the negative word and the degree adverb according to the following formula:
s(c i )＝D(c i )*neg(c i ,w)*deg(c i ,w)
wherein D is an emotion dictionary, c i For words in text, w denotes the context range, neg (c) i ,w)，deg(c i W) are respectively the corresponding negative word and degree adverb values, c) i-w A vocabulary representing the upper and lower ranges of the core word, wherein:
a second selecting unit 702, configured to select an emotion vocabulary in the comment text, and obtain an emotion value emo of the comment text according to the following formula by combining the negative word and the degree adverb z ：
Wherein L is the length of the comment text;
a score calculating unit 703, configured to calculate a score of the emotion vocabulary according to the following formula:
wherein T represents comment text, q i Represents a certain emotion;
a score connection unit 704 for connecting all the emotion vocabulary scores in a connection manner according to the following formula to obtain emotionTotal vocabulary score emo y ：
The emotion intensity calculating unit 705 is configured to calculate an emotion intensity corresponding to each emotion vocabulary according to the following formula:
in the formula, T denotes comment text, q denotes each emotion dictionary, int (c) i ) Indicates the degree of intensiveness of the ith emotional vocabulary if c i In the dictionary, int (c) is calculated from the emotion dictionary i ) Otherwise, the value is 0;
an emotion intensity connecting unit 706 for connecting the emotion intensities corresponding to the emotion words to obtain an emotion intensity characteristic emo according to the following formula d ：
An auxiliary capture unit 707 for passing an auxiliary feature emo f Capturing emoticons, punctuations, emotional words and/or human pronouns in the comment text, and connecting the emotional value, the total score of emotional words, the emotional intensity characteristic and the auxiliary characteristic of the comment text according to the following formula to obtain the emotional characteristic x:
in an embodiment, the first learning unit 602 includes:
the bidirectional learning unit is used for respectively carrying out forward learning and backward learning on the emotional characteristics according to the following formula:
in the formula,
for the forward hidden state of the affective features,
for the hidden state of the emotional feature,
is the emotional characteristic of the ith comment in the comment text, i is the single emotional characteristic in the comment,
for the last forward hidden state of the packet,
the state is the last backward hidden state, and m is the number of comments;
a first state connecting unit, configured to connect the forward hidden state of the emotion feature and the backward hidden state of the emotion feature to obtain a comment emotion feature representation of the comment text
In one embodiment, the rumor text acquisition unit 603 includes:
a second learning unit, configured to perform forward learning and backward learning on the rumor text by using a Bi-GRU network according to the following formula:
in the formula,
for the forward hidden state of the rumor text,
for the backward hidden state of the rumor text,
the j is the j word in the rumor text, and j represents a single word in the rumor text;
a second state connecting unit for connecting the forward hidden state of the rumor text with the backward hidden state of the rumor text to obtain the rumor feature representation of the rumor text
In one embodiment, as shown in fig. 8, the rumor detection unit 604 comprises:
a matrix construction unit 801, configured to construct a similar matrix F according to the following formula:
F＝tanh(E T W es S)
in the formula,
is a parameter matrix which can be learnt, E is comment emotional characteristic representation, and E ═ E 1 ,e 2 ,...,e M ]S is rumor character, S ═ S 1 ,s 2 ,...,s N ]M, N, respectively, indicating the number of sentences in the comment text and the number of sentences in the rumor text;
a collaborative mapping unit 802, configured to perform collaborative mapping on the comment emotion feature representation and the rumor feature representation by using the similarity matrix according to the following formula:
H e ＝tanh(W e E+(W s S)F T )
in the formula,
and
are all learnable parameter matrices, F is a similarity matrix, H s ,H e An attention map representing rumor text sentences and comment text, respectively;
a weight learning unit 803, configured to learn attention weights of the comment emotion feature representation and the rumor feature representation respectively according to the following formula:
in the formula,
an attention probability vector for each vocabulary in the rumor text and an attention probability vector for each comment in the emotional feature of the comment text respectively,
are all learnable weights;
a vector generating unit 804, configured to generate the comment emotional feature representations by weighted sum according to the following formula
Attention vector of sum rumor feature
In one embodiment, the rumor detection unit 604 further comprises:
a function detection unit, configured to perform rumor detection on the comment text by using a softmax function according to the following formula:
in the formula,
the rumor test results.
Since the embodiments of the apparatus portion and the method portion correspond to each other, please refer to the description of the embodiments of the method portion for the embodiments of the apparatus portion, which is not repeated here.
Embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the steps provided by the above embodiments can be implemented. The storage medium may include: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The embodiment of the present invention further provides a computer device, which may include a memory and a processor, where the memory stores a computer program, and the processor may implement the steps provided in the above embodiments when calling the computer program in the memory. Of course, the computer device may also include various network interfaces, power supplies, and the like.
The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the description of the method part. It should be noted that, for those skilled in the art, it is possible to make several improvements and modifications to the present application without departing from the principle of the present application, and such improvements and modifications also fall within the scope of the claims of the present application.
It is further noted that, in the present specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Claims (10)
1. A rumor detection method based on emotion analysis is characterized by comprising the following steps:
obtaining comment texts of rumors to be detected, and extracting emotional features of the comment texts from the comment texts through an emotional dictionary;
forward learning and backward learning are respectively carried out on the emotional features by utilizing a Bi-GRU network, and comment emotional feature representation of the comment text is obtained;
acquiring a rumor text, and respectively performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain a rumor characteristic representation of the rumor text;
and performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism, and performing rumor detection on the comment text according to correlation learning results.
2. The rumor detection method based on emotion analysis of claim 1, wherein the obtaining of comment texts of rumors to be detected and the extracting of emotion features of the comment texts in the comment texts through an emotion dictionary comprise:
and respectively extracting emotion vocabulary, emotion intensity, emotion value and auxiliary characteristics from the comment text through an emotion dictionary, and comprehensively using the emotion vocabulary, the emotion intensity, the emotion value and the auxiliary characteristics as the emotion characteristics.
3. The rumor detection method based on emotion analysis as claimed in claim 2, wherein said extracting emotion vocabulary, emotion intensity, emotion value, auxiliary features from said comment text by emotion dictionary respectively, and integrating them as said emotion features comprises:
selecting negative words and degree adverbs in the comment text, and calculating scores of the negative words and the degree adverbs according to the following formula:
s(c i )＝D(c i )*neg(c i ,w)*deg(c i ,w)
wherein D is an emotion dictionary, c i For words in text, w denotes the context range, neg (c) i ,w)，deg(c i W) are respectively the corresponding negative word and degree adverb values, c) i-w A vocabulary representing the upper and lower ranges of the core word, wherein:
selecting the emotion words in the comment text, and calculating the emotion value emo of the comment text according to the following formula by combining the negative words and the degree adverbs z ：
Wherein L is the length of the comment text;
and calculating the score of the emotion vocabulary according to the following formula:
wherein T represents comment text, q i Representing a certain mood;
connecting all the emotion vocabulary scores in a connection mode according to the following formula to obtain an emotion vocabulary total score emo y ：
Calculating the corresponding emotional intensity of each emotional vocabulary according to the following formula:
in the formula, T denotes comment text, q denotes each emotion dictionary, int (c) i ) Indicates the degree of intensiveness of the ith emotional vocabulary if c i In the dictionary, int (c) is calculated from the emotion dictionary i ) Otherwise, the value is 0;
according to the following formula, the emotional intensity characteristics emo are obtained by connecting the emotional intensity corresponding to each emotional vocabulary d ：
By assist feature emo f Capturing emoticons, punctuations, emotional words and/or human pronouns in the comment text, and connecting the emotional value, the total score of emotional words, the emotional intensity characteristic and the auxiliary characteristic of the comment text according to the following formula to obtain the emotional characteristic x:
4. the rumor detection method based on emotion analysis of claim 3, wherein the obtaining of the comment emotional feature representation of the comment text by respectively performing forward learning and backward learning on the emotional features by using the Bi-GRU network comprises:
forward learning and backward learning are respectively carried out on the emotional characteristics according to the following formula:
in the formula,
for the forward hidden state of the affective features,
for the hidden state of the emotional feature,
is the emotional characteristic of the t-th comment in the comment text, i is the single emotional characteristic in the comment,
for the last forward hidden state of the packet,
the state is the last backward hidden state, and m is the number of comments;
will give the said feelingConnecting the forward hidden state of the sensory features with the backward hidden state of the emotional features to obtain comment emotional feature representation of the comment text
5. The method of claim 4, wherein the obtaining a rumor text and performing forward learning and backward learning on the rumor text using a Bi-GRU network to obtain a rumor feature representation of the rumor text comprises:
forward learning and backward learning are respectively performed on the rumor text by using a Bi-GRU network according to the following formula:
in the formula,
for the forward hidden state of the rumor text,
for the backward hidden state of the rumor text,
the t-th vocabulary in the rumor text, and j represents a single vocabulary in the rumor text;
connecting the forward hidden state of the rumor text with the backward hidden state of the rumor text to obtain a rumor feature representation of the rumor text
6. The rumor detection method based on emotion analysis of claim 5, wherein the performing correlation learning on the comment emotion feature representation and the rumor feature representation through a cooperative attention mechanism and performing rumor detection on the comment text according to the correlation learning result comprises:
a similarity matrix F is constructed as follows:
F＝tanh(E T W es S)
in the formula,
is a parameter matrix which can be learnt, E is comment emotional characteristic representation, and E ═ E 1 ,e 2 ,...,e M ]S is rumor character, S ═ S 1 ,s 2 ,...,s N ]M, N, respectively, indicating the number of sentences in the comment text and the number of sentences in the rumor text;
and cooperatively mapping the comment emotion feature representation and the rumor feature representation by using the similarity matrix according to the following formula:
H s ＝tanh(W s S+(W e E)F)
H e ＝tanh(W e E+(W s S)F T )
in the formula,
and
the parameters are learnable parameter matrixes, F is a similar matrix, and Hs and He respectively represent attention mapping of rumor text sentences and comment texts;
the attention weights of the commentary emotion feature representation and rumor feature representation were learned separately according to the following formula:
in the formula,
an attention probability vector for each vocabulary in the rumor text and an attention probability vector for each comment in the emotional feature of the comment text respectively,
are all learnable weights;
respectively generating the comment emotional feature representations in a weighted sum mode according to the following formula
Attention vector of sum rumor feature
7. The emotion analysis-based rumor detection method of claim 6, wherein the comment emotion feature representation and the rumor feature representation are subjected to correlation learning through a cooperative attention mechanism, and the comment text is subjected to rumor detection according to a correlation learning result, further comprising:
rumor detection was performed on the comment text using softmax function according to the following formula:
in the formula,
the rumor test results.
8. A rumor detection device based on emotion analysis, comprising:
the comment text acquisition unit is used for acquiring comment texts of rumors to be detected and extracting emotional features of the comment texts from the comment texts through an emotional dictionary;
the first learning unit is used for respectively carrying out forward learning and backward learning on the emotional features by utilizing a Bi-GRU network to obtain comment emotional feature representation of the comment text;
the system comprises a rumor text acquisition unit, a rumor text analysis unit and a rumor text analysis unit, wherein the rumor text acquisition unit is used for acquiring a rumor text, and performing forward learning and backward learning on the rumor text by utilizing a Bi-GRU network to obtain rumor feature representation of the rumor text;
and the rumor detection unit is used for performing correlation learning on the comment emotion characteristic representation and the rumor characteristic representation through a cooperative attention mechanism and performing rumor detection on the comment text according to a correlation learning result.
9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor when executing the computer program implements the rumor detection method based on emotion analysis according to any one of claims 1 to 7.
10. A computer-readable storage medium, having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the rumor detection method based on emotion analysis according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210419377.5A CN114840665A (en) | 2022-04-20 | 2022-04-20 | Rumor detection method and device based on emotion analysis and related medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210419377.5A CN114840665A (en) | 2022-04-20 | 2022-04-20 | Rumor detection method and device based on emotion analysis and related medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114840665A true CN114840665A (en) | 2022-08-02 |
Family
ID=82565058
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210419377.5A Pending CN114840665A (en) | 2022-04-20 | 2022-04-20 | Rumor detection method and device based on emotion analysis and related medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114840665A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116431760A (en) * | 2023-01-10 | 2023-07-14 | 重庆理工大学 | Social network rumor detection method based on emotion perception and graph convolution network |
-
2022
- 2022-04-20 CN CN202210419377.5A patent/CN114840665A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116431760A (en) * | 2023-01-10 | 2023-07-14 | 重庆理工大学 | Social network rumor detection method based on emotion perception and graph convolution network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ishaq et al. | Aspect-based sentiment analysis using a hybridized approach based on CNN and GA | |
| CN108363753B (en) | Comment text emotion classification model training and emotion classification method, device and equipment | |
| Gupta et al. | Enhanced twitter sentiment analysis using hybrid approach and by accounting local contextual semantic | |
| Daelemans | Explanation in computational stylometry | |
| CN112131350B (en) | Text label determining method, device, terminal and readable storage medium | |
| Peng et al. | Phonetic-enriched text representation for Chinese sentiment analysis with reinforcement learning | |
| Abdullah et al. | Emotions extraction from Arabic tweets | |
| Zhang et al. | Multi-modal multi-label emotion recognition with heterogeneous hierarchical message passing | |
| Zhou et al. | Set-CNN: A text convolutional neural network based on semantic extension for short text classification | |
| CN107515855B (en) | Microblog emotion analysis method and system combined with emoticons | |
| Hyun et al. | Target-aware convolutional neural network for target-level sentiment analysis | |
| CN112256866A (en) | Text fine-grained emotion analysis method based on deep learning | |
| CN111507093A (en) | Text attack method and device based on similar dictionary and storage medium | |
| CN110321918A (en) | The method of public opinion robot system sentiment analysis and image labeling based on microblogging | |
| Peng et al. | Seq2Emoji: A hybrid sequence generation model for short text emoji prediction | |
| Liebeskind et al. | Deep learning for period classification of historical Hebrew texts | |
| CN111159405B (en) | Irony detection method based on background knowledge | |
| Diao et al. | Multi-granularity bidirectional attention stream machine comprehension method for emotion cause extraction | |
| Dehghani et al. | Sentiment analysis of Persian political tweets using ParsBERT embedding model with convolutional neural network | |
| Biswas et al. | A new ontology-based multimodal classification system for social media images of personality traits | |
| CN114817541A (en) | Rumor detection method and device based on dual-emotion perception | |
| CN114840665A (en) | Rumor detection method and device based on emotion analysis and related medium | |
| Abdi et al. | Using an auxiliary dataset to improve emotion estimation in users’ opinions | |
| Wang et al. | CA-CD: context-aware clickbait detection using new Chinese clickbait dataset with transfer learning method | |
| CN113806545B (en) | Comment text emotion classification method based on label description generation |
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 |