CN106776554A - A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph - Google Patents
A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph Download PDFInfo
- Publication number
- CN106776554A CN106776554A CN201611128388.9A CN201611128388A CN106776554A CN 106776554 A CN106776554 A CN 106776554A CN 201611128388 A CN201611128388 A CN 201611128388A CN 106776554 A CN106776554 A CN 106776554A
- Authority
- CN
- China
- Prior art keywords
- hypergraph
- microblogging
- modal
- node
- text
- 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
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000002996 emotional effect Effects 0.000 title claims abstract description 31
- 230000008451 emotion Effects 0.000 claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims description 32
- 230000000007 visual effect Effects 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 4
- 235000019227 E-number Nutrition 0.000 claims description 3
- 239000004243 E-number Substances 0.000 claims description 3
- 239000012467 final product Substances 0.000 claims description 3
- 201000011243 gastrointestinal stromal tumor Diseases 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 18
- 238000004458 analytical method Methods 0.000 abstract description 16
- 230000006870 function Effects 0.000 description 14
- 238000007477 logistic regression Methods 0.000 description 6
- 238000013398 bayesian method Methods 0.000 description 5
- 244000097202 Rathbunia alamosensis Species 0.000 description 4
- 235000009776 Rathbunia alamosensis Nutrition 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 206010016256 fatigue Diseases 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000003012 network analysis Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph, belongs to multi-modal sentiment analysis field.For problem present in the emotion prediction on microblogging multi-channel content, there is provided a kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph.Comprise the following steps:1) the multi-modal feature of microblogging is extracted;2) distance between microblogging is calculated;3) multi-modal hypergraph model is built;4) hypergraph study.The independence between different mode solution mode is more preferably associated, has relatively good effect in microblog emotional prediction.
Description
Technical field
The invention belongs to multi-modal sentiment analysis field, more particularly, to a kind of microblogging feelings based on the study of multi-modal hypergraph
Sense Forecasting Methodology.
Background technology
Recently, developing rapidly with the large-scale social platform such as Sina weibo, the multi-medium data rule of daily social networks
Mould constantly increases, and by taking Sina weibo as an example, ends in May, 2014, and Sina weibo month any active ues reach 1.4 hundred million, compared to 2013
December in year increases 10.9%.Used as one of most popular platform, Sina weibo enables that Internet user feels at them
The topic following table of interest reaches their emotion.Therefore, it has attracted the substantial amounts of research excavated on emotion information, these researchs
Being related to some emerging applications includes that event detection, social network analysis and business are recommended.
One obvious characteristic of microblogging development is the growth of multi-modal information, such as image, video, short text and rich
Rich emoticon.Main reason is that increasing Internet user issues content using more and more various equipment,
Also therefore, issue picture and emoticon turn into a kind of convenient form, rather than text for no reason.But for sentiment analysis with
Prediction, at present most research is in single text passage, rather than abundant multi-modal information.And according to cognitive section
Theory, for sentiment analysis, the otherness between different modalities is very big, it is impossible to make simple fusion, therefore multimode
State analysis is necessary.
Currently, microblog emotional analysis method mainly utilizes plain text channel information, such as《One kind is special based on large-scale corpus
Levy microblog emotional analysis method (Chinese patent CN201510310710.9) of study》、《The Chinese of rule-based and statistical model
Microblog emotional analysis method (Chinese patent CN201510127310.4)》、《A kind of Chinese microblogging for merging dominant and recessive character
Sentiment analysis method (Chinese patent CN201410723617.6)》、《A kind of Sentiment orientation analysis method of Chinese microblogging (China
Patent CN201310072472.3)》.It is single from simple the features such as content is less however, because microblogging text has structure arbitrarily
Plain text passage carry out that microblog emotional category analysis difficulty is big, the degree of accuracy of emotion prediction is low.《It is a kind of towards microblogging short text
Sentiment analysis method (Chinese patent CN201210088366.X)》Propose a kind of short text sentiment analysis method, but its face
To specific area and particular topic, without universality.《A kind of side for carrying out Sentiment orientation classification to microblogging using emoticon
Method (Chinese patent CN201310664725.6)》Disposition in proposing to be built based on emoticon Dictionary use Nae Bayesianmethod
The method of sense grader and polarity emotion classifiers carries out microblog emotional classification, but the microblogging containing emoticon is only accounted for
32%, carry out microblog emotional prediction using emoticon single channel and be difficult to be applied to all microbloggings.《One kind is based on microblogging group rings
The multi-modal sentiment analysis method (Chinese patent CN201410006867.8) of microblogging in border》The multi-modal sentiment analysis method for proposing
It is also based on microblogging original text and comment text on single text passage.
Prior art is analyzed mainly for the microblog emotional of single text passage, and there is the text of microblogging microblogging text to have
There is the features such as structure is random, and content is less, single to carry out microblog emotional category analysis difficulty greatly from simple plain text passage, emotion
The degree of accuracy of prediction is low.
The content of the invention
The purpose of the present invention be directed on microblogging multi-channel content (multi-modal) emotion prediction present in problem, carry
For a kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph.
The present invention is comprised the following steps:
Step 1 extracts the multi-modal feature of microblogging (Feature Extraction), and specific method is as follows:
Step 1.1 is entered first by Chinese Academy of Sciences automatic word segmentation instrument ICTCLAS for text modality to microblogging content of text
Row participle (Text segment), then using the Chinese sentiment dictionary (Text word dictionary) after treatment to participle
Every microblogging text afterwards builds bag of words (Bag-of-textual-word), special as the text emotion after last screening
Levy, the Chinese sentiment dictionary is made up of Hownet HowNet Chinese sentiment dictionaries and Taiwan Univ. NTUSD Chinese sentiment dictionaries, and
2547 emotion words of the higher-frequency degree occurred in microblogging corpus of text are filtered out, Chinese sentiment dictionary is constituted.I-th
Microblogging Text Representation is Fi botw;
Step 1.2 extracts picture bottom visual signature (low-level visual first for visual modalities
Feature), including local binary patterns feature (LBP), color histogram feature (color histogram), GIST features with
And PHOW describes sub- bag of words feature, then using adjective-verb to (ANP) detector library SentiBank to being extracted bottom
Every microblog picture of visual signature carries out the extraction of middle level features (mid-level visual feature), obtains 1200
The confidence value of adjective-verb pair, and more than 0.8 confidence value is retained in, remaining is set to 0, so final to obtain vision
Affective characteristics (Bag-of-visual-word).I-th microblogging image feature representation is Fi bovw;
Step 1.3 collects all of emoticon in corpus of text first for emoticon affective characteristics, then screens
Go out 49 emoticons that high frequency is used, be finally that every microblogging builds emoticon bag of words emoticon feature the most
(Bag-of-emoticon-word).I-th microblogging emoticon character representation is Fi boew;
Step 2 calculates distance (Distance computing) between microblogging, and specific method is as follows:
For text, vision, three modal characteristics of emoticon, all calculated using Euclidean distance, obtained all
Three kinds of distances of mode between two microbloggings of meaning, are expressed as Dist with distance matrix respectivelybotw,DistbovwAnd Distboew;
Step 3 builds multi-modal hypergraph model (Hypergraph construction), and specific method is as follows:
Similarity under step 3.1 calculating different modalities between sample, specific method is as follows:It is calculated using Euclidean distance
Three kinds of mode under emotion distance between any two microbloggings i and j, and then calculate its similarity s (i, j).
Step 3.2 calculates the incidence matrix of hypergraph.Hypergraph can be expressed as G={ V, E, w }, and wherein V represents all node (samples
This) set, E represents that all super sides (institute's set a little in the range of the k neighbours put centered on certain node) are gathered, and w is represented
The weight set on super side.Build hypergraph incidence matrix H:If node viIn super side ej(central node is node vi) in, then H (vi,
ej)=s (i, j);Conversely, H (vi,ej)=0.
Step 3.3 calculates the number of degrees on the node number of degrees and super side:
The node v number of degrees are calculated as follows:For all super side e comprising node v, by corresponding weights w (e) of super side e with
Node v relating value h (v, e) corresponding with super side e are multiplied and add up in incidence matrix H, obtain the node v number of degrees;
The super side e number of degrees are calculated as follows:Super side e relating value h (v, e) corresponding with all node v that it is included are tired out
Plus, obtain the number of degrees of super side e.
Step 3.4 calculates hypergraph Laplacian Matrix (Laplacian Matrix) Δ and Laplace regularization Ω:
Ω (f)=fTΔf
Wherein, Dv、De, W and I represent that node number of degrees diagonal matrix, super edge degree number diagonal matrix, super side right weight are diagonal respectively
Matrix and unit matrix, f represent the emotional category vector of prediction.Hypergraph Laplacian Matrix reflects the association feelings of each node
Condition, Laplace regularization Xiang Ze reflects the association situation of the node with different classes of label.
Step 3.5 builds object function, that is, minimize by Laplace regularization Ω, expected loss Remp(f) and
Loss function on W regularization terms composition, tries to achieve optimal f and W.
Wherein Remp(f)=| | f-y | |2, the expected loss between prediction categorization vector f and label vector y is represented,
Represent the positive regular terms of L2, wiRepresent i-th on W diagonal, neRepresent the number on super side, λ and μ be respectively loss term coefficient with
Regularization coefficient;Step 4 is introduced and hypergraph study how is carried out on object function obtains optimal f and W.
Step 4 hypergraph learns (Hypergraph learning), and specific method is as follows:
The object function in step 3.5 is optimized using alternating iteration optimization method, i.e., each iteration, first fixes W,
Make object function seek partial derivative to f, try to achieve the optimal f of current iteration, then the f of fixed optimization, makes object function seek local derviation to W
Number, tries to achieve the optimal W of current iteration.The value of the f optimized after each iteration and W is initialized f and W in next iteration, so
Iterate optimization, until the convergence of loss function value, obtains final product the f and W of final optimal.Wherein f is included to ameleia label microblogging
Emotion prediction.
Experimental evaluation standard is the degree of accuracy (Accuracy), reflects the microblog emotional polar categories and mark in advance of prediction
Emotional category between consistent degree.
The experiment effect of various methods compares referring to table 1 in text modality.
Table 1
Wherein, NB represent Nae Bayesianmethod (Bayes), LR represents logistic regression method (Logistic
Regression), SVM is support vector machine method, and HG_text is hypergraph learning method in text single mode, similarly hereinafter.
The experiment effect of multi-modal various methods compares referring to table 2.
Table 2
Wherein, CBM-NB represents cross-module state Nae Bayesianmethod, and CBM-LR represents cross-module state logistic regression method,
CBM-SVM represents cross-module state support vector machine method, MHG be it is proposed that multi-modal hypergraph learning method, similarly hereinafter.
Tables 1 and 2 contrast reflects prediction effect of the multi-modal prediction effect significantly better than plain text mode, while
See it is proposed that multi-modal hypergraph learning method effect it is best.
Fig. 4 and Fig. 5 are respectively distinct methods on 2 class feeling polarities class predictions and 3 class feeling polarities class predictions
Effect compares.
Effect of the common methods in mode independence experimentally compares referring to table 3.
Table 3
As can be seen from Table 3, multi-modal hypergraph learning method proposed by the present invention more preferably associates different mode and solves mould
Independence between state.
Brief description of the drawings
Fig. 1 is that different modalities combine upper microblogging-microblogging to similar number, and wherein TV, TE, TVE and TOTAL is represented respectively
In text and visual modalities combination, text and emoticon modality combinations, vision and emoticon modality combinations and data set
All microbloggings pair.
Fig. 2 is that different modalities combine upper microblogging-microblogging to similar ratio, and wherein TV, TE, TVE and TOTAL is represented respectively
In text and visual modalities combination, text and emoticon modality combinations, vision and emoticon modality combinations and data set
All microbloggings pair.
Fig. 3 is the microblog emotional Forecasting Methodology schematic flow sheet based on the study of multi-modal hypergraph.
Fig. 4 compares for effect of the distinct methods on 2 class feeling polarities class predictions.
Fig. 5 compares for effect of the distinct methods on 3 class feeling polarities class predictions.
Fig. 6 is the image of example 1.
Fig. 7 is the image of example 2.
Fig. 8 is the image of example 3.
Fig. 9 is the image of example 4.
Figure 10 is the image of example 5.
Figure 11 is the image of example 6.
Figure 12 is the image of example emoticon 1.
Figure 13 is the image of example emoticon 2.
Figure 14 is the image of example emoticon 3.
Figure 15 is the image of example emoticon 4.
Figure 16 is the image of example emoticon 5.
Figure 17 is the image of example emoticon 6.
Specific embodiment
The embodiment of the present invention is comprised the following steps:
Step 1 extracts the multi-modal feature of microblogging (Feature Extraction), and specific method is as follows:
Step 1.1 is entered first by Chinese Academy of Sciences automatic word segmentation instrument ICTCLAS for text modality to microblogging content of text
Row participle (Text segment), then using the Chinese sentiment dictionary (Text word dictionary) after treatment to participle
Every microblogging text afterwards builds bag of words (Bag-of-textual-word), special as the text emotion after last screening
Levy, the Chinese sentiment dictionary is made up of Hownet HowNet Chinese sentiment dictionaries and Taiwan Univ. NTUSD Chinese sentiment dictionaries, and
2547 emotion words of the higher-frequency degree occurred in microblogging corpus of text are filtered out, Chinese sentiment dictionary is constituted.I-th
Microblogging Text Representation is Fi botw;
Step 1.2 extracts picture bottom visual signature (low-level visual first for visual modalities
Feature), including local binary patterns feature (LBP), color histogram feature (color histogram), GIST features with
And PHOW describes sub- bag of words feature, then using adjective-verb to (ANP) detector library SentiBank to being extracted bottom
Every microblog picture of visual signature carries out the extraction of middle level features (mid-level visual feature), obtains 1200
The confidence value of adjective-verb pair, and more than 0.8 confidence value is retained in, remaining is set to 0, so final to obtain vision
Affective characteristics (Bag-of-visual-word).I-th microblogging image feature representation is Fi bovw;
Step 1.3 collects all of emoticon in corpus of text first for emoticon affective characteristics, then screens
Go out 49 emoticons that high frequency is used, be finally that every microblogging builds emoticon bag of words emoticon feature the most
(Bag-of-emoticon-word).I-th microblogging emoticon character representation is Fi boew;
Step 2 calculates distance (Distance computing) between microblogging, and specific method is as follows:
For text, vision, three modal characteristics of emoticon, all calculated using Euclidean distance, obtained all
Three kinds of distances of mode between two microbloggings of meaning, are expressed as Dist with distance matrix respectivelybotw,DistbovwAnd Distboew;
Step 3 builds multi-modal hypergraph model (Hypergraph construction), and specific method is as follows:
Similarity under step 3.1 calculating different modalities between sample, specific method is as follows:It is calculated using Euclidean distance
Three kinds of mode under emotion distance between any two microbloggings i and j, and then calculate its similarity s (i, j).
Step 3.2 calculates the incidence matrix of hypergraph.Hypergraph can be expressed asWhereinRepresent all nodes
(sample) is gathered,Represent that all super sides (institute's set a little in the range of the k neighbours put centered on certain node) are gathered, w tables
Show the weight set on super side.Build hypergraph incidence matrix H:If node viIn super side ej(central node is node vi) in, then H
(vi,ej)=s (i, j);Conversely, H (vi,ej)=0.
Step 3.3 calculates the number of degrees on the node number of degrees and super side:
The node v number of degrees are calculated as follows:For all super side e comprising node v, by corresponding weights w (e) of super side e with
Node v relating value h (v, e) corresponding with super side e are multiplied and add up in incidence matrix H, obtain the node v number of degrees;
The super side e number of degrees are calculated as follows:Super side e relating value h (v, e) corresponding with all node v that it is included are tired out
Plus, obtain the number of degrees of super side e.
Step 3.4 calculates hypergraph Laplacian Matrix (Laplacian Matrix) Δ and Laplace regularization Ω:
Ω (f)=fTΔf
Wherein, Dv、De, W and I represent that node number of degrees diagonal matrix, super edge degree number diagonal matrix, super side right weight are diagonal respectively
Matrix and unit matrix, f represent the emotional category vector of prediction.Hypergraph Laplacian Matrix reflects the association feelings of each node
Condition, Laplace regularization Xiang Ze reflects the association situation of the node with different classes of label.
Step 3.5 builds object function, that is, minimize by Laplace regularization Ω, expected loss Remp(f) and
Loss function on W regularization terms composition, tries to achieve optimal f and W.
Wherein Remp(f)=| | f-y | |2, the expected loss between prediction categorization vector f and label vector y is represented,
Represent the positive regular terms of L2, wiRepresent i-th on W diagonal, neRepresent the number on super side, λ and μ be respectively loss term coefficient with
Regularization coefficient;Step 4 is introduced and hypergraph study how is carried out on object function obtains optimal f and W.
Step 4 hypergraph learns (Hypergraph learning), and specific method is as follows:
The object function in step 3.5 is optimized using alternating iteration optimization method, i.e., each iteration, first fixes W,
Make object function seek partial derivative to f, try to achieve the optimal f of current iteration, then the f of fixed optimization, makes object function seek local derviation to W
Number, tries to achieve the optimal W of current iteration.The value of the f optimized after each iteration and W is initialized f and W in next iteration, so
Iterate optimization, until the convergence of loss function value, obtains final product the f and W of final optimal.Wherein f is included to ameleia label microblogging
Emotion prediction.
Experimental evaluation standard is the degree of accuracy (Accuracy), reflects the microblog emotional polar categories and mark in advance of prediction
Emotional category between consistent degree.
The experiment effect of various methods compares referring to table 1 in text modality.
Table 1
Wherein, NB represent Nae Bayesianmethod (Bayes), LR represents logistic regression method (Logistic
Regression), SVM is support vector machine method, and HG_text is hypergraph learning method in text single mode, similarly hereinafter.
The experiment effect of multi-modal various methods compares referring to table 2.
Table 2
Wherein, CBM-NB represents cross-module state Nae Bayesianmethod, and CBM-LR represents cross-module state logistic regression method,
CBM-SVM represents cross-module state support vector machine method, MHG be it is proposed that multi-modal hypergraph learning method, similarly hereinafter.
Tables 1 and 2 contrast reflects prediction effect of the multi-modal prediction effect significantly better than plain text mode, while
See it is proposed that multi-modal hypergraph learning method effect it is best.
Fig. 1 provides the upper microblogging-microblogging of different modalities combination to similar number, wherein TV, TE, TVE and TOTAL difference table
Show text and visual modalities combination, text and emoticon modality combinations, vision and emoticon modality combinations and data set
In all microbloggings pair.Fig. 2 provides the upper microblogging-microblogging of different modalities combination to similar ratio, wherein TV, TE, TVE and TOTAL
Represent respectively text and visual modalities combination, text and emoticon modality combinations, vision and emoticon modality combinations and
All microbloggings pair in data set.Fig. 3 provides the microblog emotional Forecasting Methodology schematic flow sheet based on the study of multi-modal hypergraph.
Figure 4 and 5 are respectively effect of the distinct methods on 2 class feeling polarities class predictions and 3 class feeling polarities class predictions
Fruit is compared.
Effect of the common methods in mode independence experimentally compares referring to table 3.
Table 3
As can be seen from Table 3, multi-modal hypergraph learning method proposed by the present invention more preferably associates different mode and solves mould
Independence between state.
Some multi-modal microblog emotionals predict example referring to table 4.
The multi-modal microblog emotional of table 4 predicts example
Wherein, gt represents class label (ground true).It can be seen from the results that the present invention is predicted in microblog emotional
On have relatively good effect.
Claims (5)
1. it is a kind of based on multi-modal hypergraph study microblog emotional Forecasting Methodology, it is characterised in that comprise the following steps:
1) the multi-modal feature of microblogging is extracted;
2) distance between microblogging is calculated;
3) multi-modal hypergraph model is built;
4) hypergraph study.
2. as claimed in claim 1 it is a kind of based on multi-modal hypergraph study microblog emotional Forecasting Methodology, it is characterised in that in step
In rapid 1, the specific method for extracting the multi-modal feature of microblogging is as follows:
Step 1.1 is divided microblogging content of text for text modality first by Chinese Academy of Sciences automatic word segmentation instrument ICTCLAS
Word, then builds bag of words using the Chinese sentiment dictionary after treatment to every microblogging text after participle, used as last sieve
Text emotion feature after choosing, the Chinese sentiment dictionary is by Hownet HowNet Chinese sentiment dictionaries and Taiwan Univ. NTUSD
Literary sentiment dictionary composition, and 2547 emotion words of the higher-frequency degree occurred in microblogging corpus of text are filtered out, constituted
Chinese sentiment dictionary;I-th microblogging Text Representation is Fi botw;
Step 1.2 extracts picture bottom visual signature first for visual modalities, including local binary patterns feature, colour are directly
Square figure feature, GIST features and PHOW describe sub- bag of words feature, then using adjective-verb to detector library
SentiBank carries out the extraction of middle level features to every microblog picture for being extracted bottom visual signature, obtains 1200 and describes
The confidence value of word-verb pair, and more than 0.8 confidence value is retained in, remaining is set to 0, so final to obtain visual emotion
Feature;I-th microblogging image feature representation is Fi bovw;
Step 1.3 collects all of emoticon in corpus of text first for emoticon affective characteristics, then filters out height
49 emoticons that frequency is used, are finally that every microblogging builds emoticon bag of words emoticon feature the most;I-th
Microblogging emoticon character representation is Fi boew。
3. as claimed in claim 1 it is a kind of based on multi-modal hypergraph study microblog emotional Forecasting Methodology, it is characterised in that in step
In rapid 2, the specific method of distance is as follows between the calculating microblogging:
For text, vision, three modal characteristics of emoticon, all calculated using Euclidean distance, obtained all any two
Three kinds of distances of mode, are expressed as Dist with distance matrix respectively between bar microbloggingbotw,DistbovwAnd Distboew。
4. as claimed in claim 1 it is a kind of based on multi-modal hypergraph study microblog emotional Forecasting Methodology, it is characterised in that in step
In rapid 3, the specific method for building multi-modal hypergraph model is as follows:
Similarity under step 3.1 calculating different modalities between sample, specific method is as follows:Three be calculated using Euclidean distance
Emotion distance under kind mode between any two microbloggings i and j, and then calculate its similarity s (i, j);
Step 3.2 calculates the incidence matrix of hypergraph, and hypergraph can be expressed asWhereinAll node sets are represented,
All super line sets are represented, w represents the weight set on super side;Build hypergraph incidence matrix H:If node viIn super side ejIt is interior, then H
(vi,ej)=s (i, j);Conversely, H (vi,ej)=0;
Step 3.3 calculates the number of degrees on the node number of degrees and super side:
The node v number of degrees are calculated as follows:For all super side e comprising node v, by corresponding weights w (e) of super side e with associate
Node v relating value h (v, e) corresponding with super side e are multiplied and add up in matrix H, obtain the node v number of degrees;
The super side e number of degrees are calculated as follows:Super side e relating value h (v, e) corresponding with all node v that it is included are added up, is obtained
To the number of degrees of super side e;
Step 3.4 calculates hypergraph Laplacian Matrix Δ and Laplace regularization Ω:
Ω (f)=fTΔf
Wherein, Dv、De, W and I represent respectively node number of degrees diagonal matrix, super edge degree number diagonal matrix, super side right weight diagonal matrix
And unit matrix, f represents the emotional category vector of prediction, and hypergraph Laplacian Matrix reflects the association situation of each node, draws
This regularization term of pula then reflects the association situation of the node with different classes of label;
Step 3.5 builds object function, that is, minimize by Laplace regularization Ω, expected loss Remp(f) and on
The loss function of W regularization terms composition, tries to achieve optimal f and W;
Wherein Remp(f)=| | f-y | |2, the expected loss between prediction categorization vector f and label vector y is represented,Represent L2
Positive regular terms, wiRepresent i-th on W diagonal, neThe number on super side is represented, λ and μ is respectively loss term coefficient and regular terms
Coefficient.
5. as claimed in claim 1 it is a kind of based on multi-modal hypergraph study microblog emotional Forecasting Methodology, it is characterised in that in step
In rapid 4, the specific method of the hypergraph study is as follows:
Object function is optimized using alternating iteration optimization method, i.e., each iteration, first fix W, make object function seek f
Partial derivative, tries to achieve the optimal f of current iteration, and then the f of fixed optimization, makes object function seek partial derivative to W, tries to achieve current iteration most
Excellent W;The value of the f optimized after each iteration and W is initialized f and W in next iteration, so iterate optimization, until
Loss function value restrains, and obtains final product the f and W of final optimal, and wherein f includes the emotion prediction to ameleia label microblogging.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611128388.9A CN106776554A (en) | 2016-12-09 | 2016-12-09 | A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611128388.9A CN106776554A (en) | 2016-12-09 | 2016-12-09 | A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106776554A true CN106776554A (en) | 2017-05-31 |
Family
ID=58879430
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611128388.9A Pending CN106776554A (en) | 2016-12-09 | 2016-12-09 | A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106776554A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107507195A (en) * | 2017-08-14 | 2017-12-22 | 四川大学 | The multi-modal nasopharyngeal carcinoma image partition methods of PET CT based on hypergraph model |
CN107832663A (en) * | 2017-09-30 | 2018-03-23 | 天津大学 | A kind of multi-modal sentiment analysis method based on quantum theory |
CN108595515A (en) * | 2018-03-25 | 2018-09-28 | 哈尔滨工程大学 | A kind of microblog emotional analysis method of the weak relationship of combination microblogging |
CN109376239A (en) * | 2018-09-29 | 2019-02-22 | 山西大学 | A kind of generation method of the particular emotion dictionary for the classification of Chinese microblog emotional |
CN109711366A (en) * | 2018-12-29 | 2019-05-03 | 浙江大学 | A kind of recognition methods again of the pedestrian based on group information loss function |
CN110097112A (en) * | 2019-04-26 | 2019-08-06 | 大连理工大学 | A kind of figure learning model based on reconstruct image |
CN110321918A (en) * | 2019-04-28 | 2019-10-11 | 厦门大学 | The method of public opinion robot system sentiment analysis and image labeling based on microblogging |
CN110502638A (en) * | 2019-08-30 | 2019-11-26 | 重庆誉存大数据科技有限公司 | A kind of Company News classification of risks method based on target entity |
CN110895700A (en) * | 2018-09-12 | 2020-03-20 | 北京京东尚科信息技术有限公司 | Image recognition method and system |
CN111046136A (en) * | 2019-11-13 | 2020-04-21 | 天津大学 | Method for calculating multi-dimensional emotion intensity value by fusing emoticons and short text |
CN111046137A (en) * | 2019-11-13 | 2020-04-21 | 天津大学 | Multidimensional emotion tendency analysis method |
CN111221962A (en) * | 2019-11-18 | 2020-06-02 | 重庆邮电大学 | Text emotion analysis method based on new word expansion and complex sentence pattern expansion |
CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
CN113129267A (en) * | 2021-03-22 | 2021-07-16 | 杭州电子科技大学 | OCT image detection method and system based on retina hierarchical data |
WO2023024017A1 (en) * | 2021-08-26 | 2023-03-02 | Ebay Inc. | Multi-modal hypergraph-based click prediction |
CN117892237A (en) * | 2024-03-15 | 2024-04-16 | 南京信息工程大学 | Multi-modal dialogue emotion recognition method and system based on hypergraph neural network |
CN117892237B (en) * | 2024-03-15 | 2024-06-07 | 南京信息工程大学 | Multi-modal dialogue emotion recognition method and system based on hypergraph neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014048479A1 (en) * | 2012-09-27 | 2014-04-03 | Qatar Foundation | A system and method for the automatic creation or augmentation of an electronically rendered publication document |
CN104217026A (en) * | 2014-09-28 | 2014-12-17 | 福州大学 | Chinese microblog tendency retrieving method based on graph model |
-
2016
- 2016-12-09 CN CN201611128388.9A patent/CN106776554A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014048479A1 (en) * | 2012-09-27 | 2014-04-03 | Qatar Foundation | A system and method for the automatic creation or augmentation of an electronically rendered publication document |
CN104217026A (en) * | 2014-09-28 | 2014-12-17 | 福州大学 | Chinese microblog tendency retrieving method based on graph model |
Non-Patent Citations (2)
Title |
---|
DAMIAN BORTH ET AL.: "Large-scale Visual Sentiment Ontology and Detectors Using Adjective Noun Pairs", 《PROCEEDINGS OF THE 21ST ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
FUHAI CHEN ET AL.: "Multimodal Hypergraph Learning for Microblog Sentiment Prediction", 《2015 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107507195A (en) * | 2017-08-14 | 2017-12-22 | 四川大学 | The multi-modal nasopharyngeal carcinoma image partition methods of PET CT based on hypergraph model |
CN107507195B (en) * | 2017-08-14 | 2019-11-15 | 四川大学 | The multi-modal nasopharyngeal carcinoma image partition method of PET-CT based on hypergraph model |
CN107832663B (en) * | 2017-09-30 | 2020-03-06 | 天津大学 | Multi-modal emotion analysis method based on quantum theory |
CN107832663A (en) * | 2017-09-30 | 2018-03-23 | 天津大学 | A kind of multi-modal sentiment analysis method based on quantum theory |
CN108595515A (en) * | 2018-03-25 | 2018-09-28 | 哈尔滨工程大学 | A kind of microblog emotional analysis method of the weak relationship of combination microblogging |
CN110895700A (en) * | 2018-09-12 | 2020-03-20 | 北京京东尚科信息技术有限公司 | Image recognition method and system |
CN109376239A (en) * | 2018-09-29 | 2019-02-22 | 山西大学 | A kind of generation method of the particular emotion dictionary for the classification of Chinese microblog emotional |
CN109376239B (en) * | 2018-09-29 | 2021-07-30 | 山西大学 | Specific emotion dictionary generation method for Chinese microblog emotion classification |
CN109711366A (en) * | 2018-12-29 | 2019-05-03 | 浙江大学 | A kind of recognition methods again of the pedestrian based on group information loss function |
CN110097112A (en) * | 2019-04-26 | 2019-08-06 | 大连理工大学 | A kind of figure learning model based on reconstruct image |
CN110321918A (en) * | 2019-04-28 | 2019-10-11 | 厦门大学 | The method of public opinion robot system sentiment analysis and image labeling based on microblogging |
CN110502638A (en) * | 2019-08-30 | 2019-11-26 | 重庆誉存大数据科技有限公司 | A kind of Company News classification of risks method based on target entity |
CN111046136A (en) * | 2019-11-13 | 2020-04-21 | 天津大学 | Method for calculating multi-dimensional emotion intensity value by fusing emoticons and short text |
CN111046137A (en) * | 2019-11-13 | 2020-04-21 | 天津大学 | Multidimensional emotion tendency analysis method |
CN111221962A (en) * | 2019-11-18 | 2020-06-02 | 重庆邮电大学 | Text emotion analysis method based on new word expansion and complex sentence pattern expansion |
CN111221962B (en) * | 2019-11-18 | 2023-05-26 | 重庆邮电大学 | Text emotion analysis method based on new word expansion and complex sentence pattern expansion |
CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
CN113129267A (en) * | 2021-03-22 | 2021-07-16 | 杭州电子科技大学 | OCT image detection method and system based on retina hierarchical data |
WO2023024017A1 (en) * | 2021-08-26 | 2023-03-02 | Ebay Inc. | Multi-modal hypergraph-based click prediction |
CN117892237A (en) * | 2024-03-15 | 2024-04-16 | 南京信息工程大学 | Multi-modal dialogue emotion recognition method and system based on hypergraph neural network |
CN117892237B (en) * | 2024-03-15 | 2024-06-07 | 南京信息工程大学 | Multi-modal dialogue emotion recognition method and system based on hypergraph neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106776554A (en) | A kind of microblog emotional Forecasting Methodology based on the study of multi-modal hypergraph | |
Ma et al. | Effects of user-provided photos on hotel review helpfulness: An analytical approach with deep leaning | |
Iwana et al. | Judging a book by its cover | |
Novendri et al. | Sentiment analysis of YouTube movie trailer comments using naïve bayes | |
Li et al. | Weakly supervised user profile extraction from twitter | |
Wang et al. | Sentiment analysis for social media images | |
Wang et al. | Microblog sentiment analysis based on cross-media bag-of-words model | |
CN107025284A (en) | The recognition methods of network comment text emotion tendency and convolutional neural networks model | |
CN107038609A (en) | A kind of Method of Commodity Recommendation and system based on deep learning | |
CN108108849A (en) | A kind of microblog emotional Forecasting Methodology based on Weakly supervised multi-modal deep learning | |
CN108763216A (en) | A kind of text emotion analysis method based on Chinese data collection | |
CN106980648A (en) | It is a kind of that the personalized recommendation method for combining similarity is decomposed based on probability matrix | |
CN110750656A (en) | Multimedia detection method based on knowledge graph | |
CN110196945B (en) | Microblog user age prediction method based on LSTM and LeNet fusion | |
CN110569920A (en) | prediction method for multi-task machine learning | |
Ibrahim et al. | Twitter sentiment classification using Naive Bayes based on trainer perception | |
CN111814453A (en) | Fine-grained emotion analysis method based on BiLSTM-TextCNN | |
CN112115712B (en) | Topic-based group emotion analysis method | |
Rochmawati et al. | Opinion analysis on Rohingya using Twitter data | |
CN106777040A (en) | A kind of across media microblogging the analysis of public opinion methods based on feeling polarities perception algorithm | |
Sunarya et al. | Comparison of accuracy between convolutional neural networks and Naïve Bayes Classifiers in sentiment analysis on Twitter | |
Claypo et al. | Opinion mining for Thai restaurant reviews using neural networks and mRMR feature selection | |
CN112132633A (en) | Consumption intention identification and prediction method based on consumption affair map | |
CN104572915B (en) | One kind is based on the enhanced customer incident relatedness computation method of content environment | |
CN114443846A (en) | Classification method and device based on multi-level text abnormal composition and electronic equipment |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |