CN107807919A - A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated - Google Patents
A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated Download PDFInfo
- Publication number
- CN107807919A CN107807919A CN201711131318.3A CN201711131318A CN107807919A CN 107807919 A CN107807919 A CN 107807919A CN 201711131318 A CN201711131318 A CN 201711131318A CN 107807919 A CN107807919 A CN 107807919A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- user
- blog article
- microblogging
- 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.)
- Withdrawn
Links
- 230000002996 emotional effect Effects 0.000 title claims abstract description 81
- 238000005295 random walk Methods 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000006870 function Effects 0.000 claims description 41
- 238000013507 mapping Methods 0.000 claims description 32
- 239000011159 matrix material Substances 0.000 claims description 32
- 239000013598 vector Substances 0.000 claims description 29
- 230000008451 emotion Effects 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 19
- 230000015572 biosynthetic process Effects 0.000 claims description 10
- 238000003786 synthesis reaction Methods 0.000 claims description 10
- 238000012546 transfer Methods 0.000 claims description 8
- 230000007704 transition Effects 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 2
- 230000005012 migration Effects 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 241000894007 species Species 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010304 firing Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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
- 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)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of method for carrying out microblog emotional classification prediction using random walk network is circulated.Mainly comprise the following steps:1) be directed to one group of user and microblogging blog article data set, build between user and between user and microblogging blog article correlation network.And the network to be formed is directed to, user's microblog emotional classification anticipation function is formed using random walk network is circulated.2) for obtained user's microblog emotional classification anticipation function, the classification prediction for user's microblog emotional is produced.Compared in general user microblog emotional classification solution, the present invention can utilize the social networks between the information of microblogging blog article and user simultaneously.Present invention effect acquired in microblog emotional classifies forecasting problem is more preferable compared to traditional method.
Description
Technical field
The present invention relates to microblog emotional classification to predict, more particularly to one kind carries out microblogging feelings using random walk network is circulated
The method of sense classification prediction.
Background technology
For currently booming microblogging, the problem of prediction of user's microblog emotional is one important.This problem
Target be based on having been observed customer relationship in current network and the emotional semantic classification of microblogging that user is sent out is for user
The emotional semantic classification of following microblogging is predicted.
Existing technology, which mainly classifies microblog emotional, to be done as a kind of text emotion classification task, only for
Family microblogging sent out in the past is trained, and obtains the sentiment classification model of its user's microblogging, so as to predict that the following user is sent out
The emotional semantic classification of microblogging, this method lock into the difficulty that validity expression is carried out for microblogging.
The present invention will carry out the prediction of user's microblog emotional classification, the network using a kind of heterogeneous microblog emotional sorter network
The social networks between the deep semantic expression of user's content of microblog and user can be extracted simultaneously, and it is random that the network is accompanied with one
Migration layer learns heterogeneous microblogging semantic classification network mapping, and the network just can be learnt end to end from the beginning.
The content of the invention
It is an object of the invention to solve the problems of the prior art, in order to overcome lack in the prior art it is micro- for user
The problem of effective expression of rich blog article, the present invention provide one kind and carry out microblog emotional classification prediction using random walk network is circulated
Method.Concrete technical scheme of the present invention is:
Solve the problems, such as microblog emotional classification prediction using random walk network is circulated, comprise the following steps:
1st, one group of social network user and its microblogging blog article are directed to, structure synthesis includes social networks and use between user
The network of correlation between family and microblogging blog article.
2nd, using circulating, random walk Network Capture user is following to send out the intersection entropy loss item and certain that microblog emotional classifies
The uniformity of sending out microblogging blog article loss item of the user under the influence of other users, and both addition is obtained into final damage
Lose item.By study, this final loss item is minimized, to train to obtain final user's microblog emotional classification prediction letter
Number.
3rd, the user's microblog emotional classification anticipation function obtained using study obtains the prediction emotional semantic classification of user's microblogging.
Above-mentioned steps can be specifically using being implemented as described below mode:
1st, the microblogging sent out for given user and user, according to social networks between the user of real data concentration
And user and the issue relation of microblogging blog article form heterogeneous microblog emotional sorter network, are designated as MSC networks.
2nd, the MSC networks completed for structure, for given microblogging blog article, its word is passed through into the good list of training in advance
Word mapping network obtains the mapping of its word.For the microblogging blog article x being made up of a word sequenceiIf its t-th of word leads to
Cross the word that the good word mapping network of training in advance obtains and be mapped as xit, then by sequence (xi1,xi2,...,xik) it is used as microblogging
Blog article xiWord mapping table reach, afterwards, by blog article xiIt is divided into some sections, and using each section of word sequence of mapping as LSTM nets
The input of network, reached using the output of last hidden layer of LSTM networks as the mapping table of this section of blog article, afterwards by each section
Output inputs a maximum pond layer simultaneously, by the output t of pond layeri∈RdAs microblogging blog article xiMapping table reach, tiFor one
Individual d dimensional vectors.
3rd, using softmax functions come designing user individualized emotion disaggregated model, the mapping of given j-th strip microblogging blog article
Express tj, then it is as follows for the personalized semantic function of i-th of user:
Wherein, c is the species number of all emotional semantic classifications, vectorialFor the prediction emotion point of j-th strip microblogging blog article
Class vector,To be learnt the emotion anticipation function of certain microblogging for i-th of user, u0∈Rd*cFor for
The overall Semantic mapping matrix of all users, ui∈Rd*cFor the certain semantic mapping matrix for i-th of user, softmax
() is softmax functions.
Then for above formula final gained vectorEvery one-dimensional fu,k(tj), it is calculated by equation below:
Wherein u0,kWith ui,kRespectively u0With uiThe vector of corresponding kth dimension.
4th, with reference to the emotion predicted vector of user's microblogging blog article obtained in the previous stepIt is micro- with the user in real training set
Rich blog article emotional semantic classification vector y, the following intersection entropy loss item for sending out microblog emotional classification of user is obtained using equation below:
Wherein, set AiThe set formed for all microblogging blog articles of i-th of user,For the prediction feelings of j-th strip microblogging
Sense classification, yjClassify for the real feelings of j-th strip microblogging, m is overall user number, for yjVector, it only corresponds to correct feelings
The dimension values of sense classification are 1, and the value of remaining dimension is 0.
5th, with reference to resulting MSC networks, the correlation matrix S ∈ R between user are obtainedm*m, m is overall user
Number, if i-th of user is paying close attention to j-th of user, sij=1, otherwise, sij=0.Obtain the pass between microblogging blog article and user
It is matrix A ∈ Rn*m, wherein n is the number of overall microblogging blog article, if i-th microblogging blog article is sent by j-th of user,
aij=1, otherwise, aij=0.
One-dimensional microblogging blog article relational matrix B=ASA is obtained by s-matrix and A matrixesTIf then i-th microblogging blog article and jth
Bar microblogging blog article is to be sent by same user or sent by two users of only 1 hop distance between in MSC networks, then bij
=1, otherwise, bij=0.Specify | Ai| it is microblogging blog article bar number related to i-th microblogging blog article in B matrixes, then can obtain
Diagonal matrix D=diag (| A1|,|A2|,...,|An|), then obtain single order microblogging blog article transfer matrix W=D-1B.Then in bij
On the premise of=1, the transition probability for being directed to i-th and j-th strip microblogging blog article isIf i-th and j-th strip microblogging
There is no correlation between blog article, then wij=0.
6th, single order microblogging blog article transfer matrix W, the initial predicted emotional semantic classification vector of j-th strip microblogging blog article are givenThis
Invention carries out successive ignition with reference to the thinking of random walk, obtains prediction emotional semantic classification of the j-th strip microblogging blog article in the step of kth+1
VectorWherein,Represent the prediction emotional semantic classification vector that j-th strip microblogging blog article walks in kth, W(k)
Represent W k power.Then the semantic consistency of j-th strip microblogging blog article retains emotion prediction and can obtained by equation below:
Wherein,For W(k)Element in matrix, represent i-th microblogging blog article and won with j-th strip microblogging blog article in k rank microbloggings
Literary transfer matrix W(k)Correlation.
Then shown in the loss item equation below of the kth rank semantic consistency of i-th microblogging blog article:
Wherein,Represent 2 rank frobenius norms.
7th, then synthetic user future sends out the semantic consistency for intersecting entropy loss item and microblogging blog article of microblog emotional classification
Loss item, it is as follows to can obtain final loss function:
Wherein, α is the balance parameter for intersecting the loss item of entropy loss item and semantic consistency, and k is random walk layer
The number of plies.
8th, for the circulation random walk network constructed by step 2 to step 7, parameter sets all in the network are set
For θ, the loss function obtained by being combined with step 7, the final goal function of circulation random walk network model is obtained such as
Under:
Wherein, θ is all parameters in model, and λ is the balance parameter between training penalty values and regular terms.
For the final object function in step 8, the present invention carrys out undated parameter using the method for stochastic gradient descent, and
And using the renewal of all parameters in Adagrad learning rate update method progress network, obtain final all users'
Microblogging blog article emotional semantic classification anticipation function
9th, the order standard anticipation function formed using step 8For a certain
The mapping table for the microblogging text that user is sent reaches, and tries to achieve the emotional semantic classification predicted value of the microblogging blog article, will have maximum probability
Emotional category as prediction the microblogging emotional semantic classification.
Brief description of the drawings
Fig. 1 be it is used in the present invention using existing social networks between user and user and microblogging blog article directly by phase
The overall schematic of the MSC networks of mutual relation structure.Fig. 2 is the circulation of progress microblog emotional classification prediction used in the present invention
The schematic diagram of random walk network learning model.
Embodiment
The present invention is further elaborated and illustrated with reference to the accompanying drawings and detailed description.
A kind of as shown in figure 1, method bag for carrying out microblog emotional classification using random walk network is circulated and predicting of the present invention
Include following steps:
1) be directed to one group of user and microblogging, structure is comprehensive include between user social networks and user and microblogging blog article it
Between correlation network;
2) for the synthesis obtained by step 1) include user between social networks and user and microblogging blog article mutually
The network of relation, using circulating, random walk Network Capture user is following to send out the intersection entropy loss item and certain that microblog emotional classifies
The semantic consistency of sending out microblogging blog article loss item of the user under the influence of other users, and both addition is obtained finally
Loss item;By study, this final loss item is minimized, to train to obtain final user's microblog emotional classification in advance
Survey function;
3) the user's microblog emotional classification anticipation function obtained using step 2) study obtains the prediction emotion of user's microblogging
Classification.
Described step 2) the user microblog emotional classification anticipation function final using random walk Network Capture is circulated, its
Concretely comprise the following steps:
2.1) for step 1) formed synthesis include user between social networks and user and microblogging blog article mutually
The network of relation, obtain user using the word mapping network, LSTM networks and softmax functions of pre-training and send out microblogging in future
The intersection entropy loss item of emotional semantic classification;
2.2) for step 1) formed synthesis include user between social networks and user and microblogging blog article mutually
The network of relation, utilize the semantic congruence of sending out microblogging blog article of the random walk Network Capture user under the influence of other users
Property loss item, and combine that user that step 2.1) obtains is following to be sent out the intersection entropy loss item that microblog emotional is classified and obtain finally
Object function;
2.3) the final goal function found out using step 2.2), learn the microblog emotional point of all users by training
Class anticipation function.
Described step 2.1) is specially:
Being directed to the synthesis of step 1) acquisition includes mutually closing between social networks and user and microblogging blog article between user
The network of system, for given microblogging blog article, its word is obtained into its word by the good word mapping network of training in advance and reflected
Penetrate.For the microblogging blog article x being made up of a word sequenceiIf its t-th of word maps net by the good word of training in advance
The word that network obtains is mapped as xit, then by sequence (xi1,xi2,...,xik) it is used as microblogging blog article xiWord mapping table reach, it
Afterwards, by blog article xiIt is divided into some sections, and the input using each section of word sequence of mapping as LSTM networks, with LSTM networks most
The output of the latter hidden layer is reached as the mapping table of this section of blog article, and each section of output is inputted into a maximum pond simultaneously afterwards
Layer, by the output t of pond layeri∈RdAs microblogging blog article xiMapping table reach, tiFor a d dimensional vector.
Reflected afterwards using softmax functions come designing user individualized emotion disaggregated model, given j-th strip microblogging blog article
Firing table reaches tj, then it is as follows for the personalized semantic function of i-th of user:
Wherein, c is the species number of all emotional semantic classifications, vectorialFor the prediction emotion point of j-th strip microblogging blog article
Class vector,To be learnt the emotion anticipation function of certain microblogging for i-th of user, u0∈Rd*cFor for
The overall Semantic mapping matrix of all users, ui∈Rd*cFor the certain semantic mapping matrix for i-th of user, softmax
() is softmax functions.
Then for above formula final gained vectorEvery one-dimensional fu,k(tj), it is calculated by equation below:
Wherein u0,kWith ui,kRespectively u0With uiThe vector of corresponding kth dimension.
In conjunction with the emotion predicted vector of user's microblogging blog article obtained in the previous stepIt is micro- with the user in real training set
Rich blog article emotional semantic classification vector y, the following intersection entropy loss item for sending out microblog emotional classification of user is obtained using equation below:
Wherein, set AiThe set formed for all microblogging blog articles of i-th of user,For the prediction feelings of j-th strip microblogging
Sense classification, yjClassify for the real feelings of j-th strip microblogging, m is overall user number, for yjVector, it only corresponds to correct feelings
The dimension values of sense classification are 1, and the value of remaining dimension is 0.
Described step 2.2) is specially:
Being directed to the synthesis that step 1) obtained is included between user between social networks and user and microblogging blog article mutually
The network of relation, obtain the correlation matrix S ∈ R between userm*m, m is overall number of users, if i-th of user is paying close attention to
J-th of user, then sij=1, otherwise, sij=0.Obtain the relational matrix A ∈ R between microblogging blog article and usern*m, wherein n is
The number of overall microblogging blog article, if i-th microblogging blog article is sent by j-th of user, aij=1, otherwise, aij=0.
One-dimensional microblogging blog article relational matrix B=ASA is obtained by s-matrix and A matrixesTIf then i-th microblogging blog article and jth
Bar microblogging blog article is to be sent by same user or sent by two users of only 1 hop distance between in MSC networks, then bij
=1, otherwise, bij=0.Specify | Ai| it is microblogging blog article bar number related to i-th microblogging blog article in B matrixes, then can obtain
Diagonal matrix D=diag (| A1|,|A2|,...,|An|), then obtain single order microblogging blog article transfer matrix W=D-1B.Then in bij
On the premise of=1, the transition probability for being directed to i-th and jth bar microblogging blog article isIf i-th micro- with j-th strip
There is no correlation between rich blog article, then wij=0.
Then give single order microblogging blog article transfer matrix W, the initial predicted emotional semantic classification vector of j-th strip microblogging blog articleKnot
The thinking for closing random walk carries out successive ignition, obtains prediction emotional semantic classification vector of the j-th strip microblogging blog article in the step of kth+1Wherein,Represent the prediction emotional semantic classification vector that j-th strip microblogging blog article walks in kth, W(k)Generation
Table W k power.Then the semantic consistency of j-th strip microblogging blog article retains emotion prediction and can obtained by equation below:
Wherein,For W(k)Element in matrix, i-th microblogging blog article is represented with j-th strip microblogging blog article in k rank microbloggings
Blog article transfer matrix W(k)Correlation.
Then shown in the loss item equation below of the kth rank semantic consistency of i-th microblogging blog article:
Wherein,Represent 2 rank frobenius norms.
The then following intersection entropy loss item for sending out microblog emotional classification of synthetic user and the semantic consistency of microblogging blog article
Item is lost, it is as follows to can obtain final loss function:
Wherein, α is the balance parameter for intersecting the loss item of entropy loss item and semantic consistency, and k is random walk layer
The number of plies.
Step 2.3) is specially:
For the circulation random walk network constructed by step 2), parameter sets all in the network are set to θ, with reference to
The final loss function obtained using step 2), using the target letter that equation below is overall as circulation random walk network model
Numerical value:
Wherein, θ is all parameters in model, and λ is the balance parameter between training penalty values and regular terms.
Afterwards, carry out undated parameter using the method for stochastic gradient descent, and use Adagrad learning rate update method
The renewal of all parameters in network is carried out, obtains the microblogging blog article emotional semantic classification anticipation function of final all users
Described step 3) is specially:
The microblogging blog article emotional semantic classification anticipation function of all users formed using step 2)
The mapping table of the microblogging text sent for a certain user reaches, and tries to achieve the emotional semantic classification predicted value of the microblogging blog article, will have
Emotional semantic classification of the emotional category of maximum probability as the microblogging of prediction.
The above method is applied in the following example below, it is specific in embodiment with the technique effect of the embodiment present invention
Step repeats no more.
Embodiment
The present invention is on Stanford Twitter Sentiment data sets and Obama-McCain Debate data sets
Face carries out experimental verification respectively.Include the micro- of 22262 tape labels in Stanford Twitter Sentiment data sets altogether
Rich blog article, wherein 11959 microblogging blog articles are marked as positive emotion, 10303 microblogging blog articles are marked as negative sense emotion,
The microblogging blog article number of average each user is 2.63 in Stanford Twitter Sentiment data sets.Obama-McCain
Include the microblogging blog article of 1827 tape labels in Debate data sets altogether, wherein 747 microblogging blog articles are marked as positive emotion,
1080 microblogging blog articles are marked as negative sense emotion, the microblogging for each user that is averaged in Obama-McCain Debate data sets
Blog article number is 2.49.
In order to objectively evaluate the performance of the algorithm of the present invention, the present invention uses in selected test set
Accuracy come for the present invention effect evaluate.The step of according to described in embodiment, the experiment knot of gained
Fruit as shown in table 1 and table 2, the present invention used in method be designated as RRWNL, and be directed to respectively all training sets 10%, 25%,
50%th, 100% training data obtains experimental result as final training set:
The present invention of table 1 is directed to the test result of Stanford Twitter Sentiment data sets
The present invention of table 2 is directed to the test result of Obama-McCain Debate data sets.
Claims (6)
- A kind of 1. method for carrying out microblog emotional classification prediction using random walk network is circulated, it is characterised in that including following step Suddenly:1) one group of user and microblogging are directed to, structure is comprehensive to include between user phase between social networks and user and microblogging blog article The network of mutual relation;2) for the social networks obtained by step 1), using circulating, random walk Network Capture user is following to send out microblog emotional The semantic consistency of sending out microblogging blog article loss item of the intersection entropy loss item of classification with certain user under the influence of other users, And both addition is obtained into final loss item, by study, this final loss item is minimized, to train to obtain most Whole user's microblog emotional classification anticipation function;3) the user's microblog emotional classification anticipation function obtained using step 2) study obtains the prediction emotional semantic classification of user's microblogging.
- 2. the method for carrying out microblog emotional classification prediction using random walk network is circulated according to claim 1, its feature Being described step 2), it is concretely comprised the following steps:2.1) correlation between social networks and user and microblogging blog article between user is included for the synthesis that step 1) is formed Network, obtain that user is following to send out microblog emotional using the word mapping network, LSTM networks and softmax functions of pre-training The intersection entropy loss item of classification;2.2) synthesis formed using step 1) includes correlation between social networks and user and microblogging blog article between user Network, using the semantic consistency of sending out microblogging blog article of the random walk Network Capture user under the influence of other users Loss item, and combine that user that step 2.1) obtains is following to be sent out the intersection entropy loss item that microblog emotional is classified and obtain final mesh Scalar functions;2.3) the final goal function found out using step 2.2), the microblog emotional for learning all users by training are classified in advance Survey function.
- 3. the method for carrying out microblog emotional classification prediction using random walk network is circulated according to claim 2, its feature It is that described step 2.1) is specially:Being directed to the synthesis of step 1) acquisition includes correlation between social networks and user and microblogging blog article between user Network, for given microblogging blog article, its word is obtained into its word by the good word mapping network of training in advance and mapped;It is right In the microblogging blog article x being made up of a word sequenceiIf its t-th of word is obtained by the good word mapping network of training in advance The word taken is mapped as xit, then by sequence (xi1,xi2,...,xik) it is used as microblogging blog article xiWord mapping table reach, afterwards, will Blog article xiIt is divided into some sections, and the input using each section of word sequence of mapping as LSTM networks, with last of LSTM networks The output of individual hidden layer is reached as the mapping table of this section of blog article, and each section of output is inputted into a maximum pond layer simultaneously afterwards, By the output t of pond layeri∈RdAs microblogging blog article xiMapping table reach, tiFor a d dimensional vector;Afterwards using softmax functions come designing user individualized emotion disaggregated model, the mapping table of given j-th strip microblogging blog article Up to tj, then it is as follows for the personalized semantic function of i-th of user:<mrow> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> <mo>=</mo> <msub> <mi>f</mi> <msub> <mi>u</mi> <mi>i</mi> </msub> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mi>s</mi> <mi>o</mi> <mi>f</mi> <mi>t</mi> <mi> </mi> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mrow> <mo>(</mo> <msup> <mrow> <mo>&lsqb;</mo> <msub> <mi>u</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>u</mi> <mi>i</mi> </msub> <mo>&rsqb;</mo> </mrow> <mi>T</mi> </msup> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "(" close = ")"> <mtable> <mtr> <mtd> <mrow> <msub> <mi>f</mi> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <msub> <mi>f</mi> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>...</mn> </mtd> </mtr> <mtr> <mtd> <mrow> <msub> <mi>f</mi> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>c</mi> </mrow> </msub> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>Wherein, c is the species number of all emotional semantic classifications, vectorialFor j-th strip microblogging blog article prediction emotional semantic classification to Amount, fui() is to be learnt certain microblogging of the emotion anticipation function to(for) i-th of user, u0∈Rd*cFor for all The overall Semantic mapping matrix of user, ui∈Rd*cFor the certain semantic mapping matrix for i-th of user, softmax () is Softmax functions;Then for above formula final gained vectorEvery one-dimensional fu,k(tj), it is calculated by equation below:<mrow> <msub> <mi>f</mi> <mrow> <mi>u</mi> <mo>,</mo> <mi>k</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>exp</mi> <mrow> <mo>(</mo> <msup> <mrow> <mo>&lsqb;</mo> <msub> <mi>u</mi> <mrow> <mn>0</mn> <mo>,</mo> <mi>k</mi> </mrow> </msub> <mo>+</mo> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>k</mi> </mrow> </msub> <mo>&rsqb;</mo> </mrow> <mi>T</mi> </msup> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>p</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>c</mi> </msubsup> <mi>exp</mi> <mrow> <mo>(</mo> <msup> <mrow> <mo>&lsqb;</mo> <msub> <mi>u</mi> <mrow> <mn>0</mn> <mo>,</mo> <mi>k</mi> </mrow> </msub> <mo>+</mo> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>p</mi> </mrow> </msub> <mo>&rsqb;</mo> </mrow> <mi>T</mi> </msup> <msub> <mi>t</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>Wherein u0,kWith ui,kRespectively u0With uiThe vector of corresponding kth dimension;In conjunction with the emotion predicted vector of user's microblogging blog article obtained in the previous stepWon with user's microblogging in real training set Literary emotional semantic classification vector y, the following intersection entropy loss item for sending out microblog emotional classification of user is obtained using equation below:<mrow> <msub> <mi>L</mi> <mi>c</mi> </msub> <mo>=</mo> <mo>-</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <munder> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>A</mi> <mi>i</mi> </msub> </mrow> </munder> <msubsup> <mi>y</mi> <mi>j</mi> <mi>T</mi> </msubsup> <mi>l</mi> <mi>o</mi> <mi>g</mi> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> </mrow>Wherein, set AiThe set formed for all microblogging blog articles of i-th of user,For the prediction emotion point of j-th strip microblogging Class, yjClassify for the real feelings of j-th strip microblogging, m is overall user number, for yjVector, it only corresponds to correct emotion point The dimension values of class are 1, and the value of remaining dimension is 0.
- 4. the method for carrying out microblog emotional classification prediction using random walk network is circulated according to claim 2, its feature It is that described step 2.2) is specially:Being directed to the synthesis that step 1) obtained includes between user correlation between social networks and user and microblogging blog article Network, obtain the correlation matrix S ∈ R between userm*m, m is overall number of users, if i-th of user is in concern jth Individual user, then sij=1, otherwise, sij=0;Obtain the relational matrix A ∈ R between microblogging blog article and usern*m, wherein n is overall Microblogging blog article number, if i-th microblogging blog article is sent by j-th of user, aij=1, otherwise, aij=0;One-dimensional microblogging blog article relational matrix B=ASA is obtained by s-matrix and A matrixesTIf then i-th microblogging blog article and j-th strip microblogging Blog article is to be sent by same user or sent by two users of only 1 hop distance between in MSC networks, then bij=1, it is no Then, bij=0;Specify | Ai| it is microblogging blog article bar number related to i-th microblogging blog article in B matrixes, then can obtains to angular moment Battle array D=diag (| A1|,|A2|,...,|An|), then obtain single order microblogging blog article transfer matrix W=D-1B;Then in bij=1 Under the premise of, the transition probability for being directed to i-th and j-th strip microblogging blog article isIf i-th and j-th strip microblogging blog article Between there is no correlation, then wij=0;Then give single order microblogging blog article transfer matrix W, the initial predicted emotional semantic classification vector of j-th strip microblogging blog articleWith reference to The thinking of machine migration carries out successive ignition, obtains prediction emotional semantic classification vector of the j-th strip microblogging blog article in the step of kth+1Wherein,Represent the prediction emotional semantic classification vector that j-th strip microblogging blog article walks in kth, W(k)Represent W k power;Then the semantic consistency of j-th strip microblogging blog article retains emotion prediction and can obtained by equation below:<mrow> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>i</mi> </msub> <mo>&ap;</mo> <msub> <mi>&Sigma;</mi> <mrow> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>></mo> <mn>0</mn> <mo>,</mo> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>&Element;</mo> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msup> </mrow> </msub> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> </mrow>Wherein,For W(k)Element in matrix, represent i-th microblogging blog article and turn with j-th strip microblogging blog article in k rank microbloggings blog article Move matrix W(k)Correlation;Then shown in the loss item equation below of the kth rank semantic consistency of i-th microblogging blog article:<mrow> <mo>|</mo> <mo>|</mo> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>&Sigma;</mi> <mrow> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>></mo> <mn>0</mn> <mo>,</mo> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>&Element;</mo> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msup> </mrow> </msub> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> <mo>|</mo> <msubsup> <mo>|</mo> <mi>F</mi> <mn>2</mn> </msubsup> </mrow>Wherein,Represent 2 rank frobenius norms;The then following loss for intersecting entropy loss item and the semantic consistency of microblogging blog article for sending out microblog emotional classification of synthetic user , it is as follows to can obtain final loss function:<mrow> <mi>L</mi> <mo>=</mo> <msub> <mi>L</mi> <mi>c</mi> </msub> <mo>+</mo> <mi>&alpha;</mi> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mo>|</mo> <mo>|</mo> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>&Sigma;</mi> <mrow> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>></mo> <mn>0</mn> <mo>,</mo> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <mo>&Element;</mo> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msup> </mrow> </msub> <msubsup> <mi>w</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </msubsup> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> <mo>|</mo> <msubsup> <mo>|</mo> <mi>F</mi> <mn>2</mn> </msubsup> </mrow><mrow> <msub> <mi>L</mi> <mi>c</mi> </msub> <mo>=</mo> <mo>-</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <munder> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>&Element;</mo> <msub> <mi>A</mi> <mi>i</mi> </msub> </mrow> </munder> <msubsup> <mi>y</mi> <mi>j</mi> <mi>T</mi> </msubsup> <mi>l</mi> <mi>o</mi> <mi>g</mi> <msub> <mover> <mi>y</mi> <mo>^</mo> </mover> <mi>j</mi> </msub> </mrow>Wherein, α is the balance parameter for intersecting the loss item of entropy loss item and semantic consistency, and k is the layer of random walk layer Number.
- 5. the method for carrying out microblog emotional classification prediction using random walk network is circulated according to claim 2, its feature It is that described step 2.3) is specially:For the circulation random walk network constructed by step 2), parameter sets all in the network are set to θ, are combined with The final loss function that step 2) obtains, using the object function that equation below is overall as circulation random walk network model Value:<mrow> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mi>&theta;</mi> </munder> <mi>L</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&Sigma;</mo> <mrow> <mi>t</mi> <mo>&Element;</mo> <mi>T</mi> </mrow> </munder> <mi>L</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>+</mo> <mi>&lambda;</mi> <mo>|</mo> <mo>|</mo> <mi>&theta;</mi> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow>Wherein, θ is all parameters in model, and λ is the balance parameter between training penalty values and regular terms;Afterwards, using with The method that machine gradient declines carrys out undated parameter, and uses all ginsengs in Adagrad learning rate update method progress network Several renewals, obtain the microblogging blog article emotional semantic classification anticipation function of final all users
- 6. the method for carrying out microblog emotional classification prediction using random walk network is circulated according to claim 1, its feature It is that described step 3) is specially:The microblogging blog article emotional semantic classification anticipation function of all users formed using step 2) The mapping table of the microblogging text sent for a certain user reaches, and tries to achieve the emotional semantic classification predicted value of the microblogging blog article, will have Emotional semantic classification of the emotional category of maximum probability as the microblogging of prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711131318.3A CN107807919A (en) | 2017-11-15 | 2017-11-15 | A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711131318.3A CN107807919A (en) | 2017-11-15 | 2017-11-15 | A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107807919A true CN107807919A (en) | 2018-03-16 |
Family
ID=61580117
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711131318.3A Withdrawn CN107807919A (en) | 2017-11-15 | 2017-11-15 | A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107807919A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108717587A (en) * | 2018-05-25 | 2018-10-30 | 杭州知智能科技有限公司 | A method of text prediction forwarding task is pushed away based on the solution of multi-panel sorting network |
CN108804689A (en) * | 2018-06-14 | 2018-11-13 | 合肥工业大学 | The label recommendation method of the fusion hidden connection relation of user towards answer platform |
CN109213831A (en) * | 2018-08-14 | 2019-01-15 | 阿里巴巴集团控股有限公司 | Event detecting method and device calculate equipment and storage medium |
CN110647804A (en) * | 2019-08-09 | 2020-01-03 | 中国传媒大学 | Violent video identification method, computer system and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101714135A (en) * | 2009-12-11 | 2010-05-26 | 中国科学院计算技术研究所 | Emotional orientation analytical method of cross-domain texts |
CN104268230A (en) * | 2014-09-28 | 2015-01-07 | 福州大学 | Method for detecting objective points of Chinese micro-blogs based on heterogeneous graph random walk |
CN105260356A (en) * | 2015-10-10 | 2016-01-20 | 西安交通大学 | Chinese interactive text emotion and topic identification method based on multitask learning |
CN105740349A (en) * | 2016-01-25 | 2016-07-06 | 重庆邮电大学 | Sentiment classification method capable of combining Doc2vce with convolutional neural network |
US20160283462A1 (en) * | 2015-03-24 | 2016-09-29 | Xerox Corporation | Language identification on social media |
CN107341270A (en) * | 2017-07-28 | 2017-11-10 | 东北大学 | Towards the user feeling influence power analysis method of social platform |
-
2017
- 2017-11-15 CN CN201711131318.3A patent/CN107807919A/en not_active Withdrawn
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101714135A (en) * | 2009-12-11 | 2010-05-26 | 中国科学院计算技术研究所 | Emotional orientation analytical method of cross-domain texts |
CN104268230A (en) * | 2014-09-28 | 2015-01-07 | 福州大学 | Method for detecting objective points of Chinese micro-blogs based on heterogeneous graph random walk |
US20160283462A1 (en) * | 2015-03-24 | 2016-09-29 | Xerox Corporation | Language identification on social media |
CN105260356A (en) * | 2015-10-10 | 2016-01-20 | 西安交通大学 | Chinese interactive text emotion and topic identification method based on multitask learning |
CN105740349A (en) * | 2016-01-25 | 2016-07-06 | 重庆邮电大学 | Sentiment classification method capable of combining Doc2vce with convolutional neural network |
CN107341270A (en) * | 2017-07-28 | 2017-11-10 | 东北大学 | Towards the user feeling influence power analysis method of social platform |
Non-Patent Citations (1)
Title |
---|
ZHOU ZHAO、HANQING LU、DENG CAI、XIAOFEI HE、YUETING ZHUANG: "Microblog Sentiment Classification via Recurrent Random Walk Network Learning", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/28371331》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108717587A (en) * | 2018-05-25 | 2018-10-30 | 杭州知智能科技有限公司 | A method of text prediction forwarding task is pushed away based on the solution of multi-panel sorting network |
CN108717587B (en) * | 2018-05-25 | 2022-03-15 | 杭州一知智能科技有限公司 | Method for solving tweet prediction forwarding task based on multi-face sequencing network |
CN108804689A (en) * | 2018-06-14 | 2018-11-13 | 合肥工业大学 | The label recommendation method of the fusion hidden connection relation of user towards answer platform |
CN108804689B (en) * | 2018-06-14 | 2020-10-16 | 合肥工业大学 | Question-answering platform-oriented label recommendation method integrating user hidden connection relation |
CN109213831A (en) * | 2018-08-14 | 2019-01-15 | 阿里巴巴集团控股有限公司 | Event detecting method and device calculate equipment and storage medium |
CN110647804A (en) * | 2019-08-09 | 2020-01-03 | 中国传媒大学 | Violent video identification method, computer system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bang et al. | Explaining a black-box by using a deep variational information bottleneck approach | |
CN108009285B (en) | Forest Ecology man-machine interaction method based on natural language processing | |
CN103345656B (en) | A kind of data identification method based on multitask deep neural network and device | |
CN104598611B (en) | The method and system being ranked up to search entry | |
CN107153642A (en) | A kind of analysis method based on neural network recognization text comments Sentiment orientation | |
CN109299262A (en) | A kind of text implication relation recognition methods for merging more granular informations | |
CN108364028A (en) | A kind of internet site automatic classification method based on deep learning | |
CN107807919A (en) | A kind of method for carrying out microblog emotional classification prediction using random walk network is circulated | |
CN109325231A (en) | A kind of method that multi task model generates term vector | |
CN109919316A (en) | The method, apparatus and equipment and storage medium of acquisition network representation study vector | |
CN108829763A (en) | A kind of attribute forecast method of the film review website user based on deep neural network | |
CN107358293A (en) | A kind of neural network training method and device | |
CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
CN103324954B (en) | Image classification method based on tree structure and system using same | |
CN106529503A (en) | Method for recognizing face emotion by using integrated convolutional neural network | |
CN107341145A (en) | A kind of user feeling analysis method based on deep learning | |
CN105447510B (en) | Fluctuating wind speed prediction technique based on artificial bee colony optimization LSSVM | |
CN105868773A (en) | Hierarchical random forest based multi-tag classification method | |
CN103399932B (en) | A kind of situation identification method based on semantic community network ontological analysis technology | |
CN105260746B (en) | A kind of integrated Multi-label learning system of expansible multilayer | |
CN106934071A (en) | Recommendation method and device based on Heterogeneous Information network and Bayes's personalized ordering | |
CN110263979A (en) | Method and device based on intensified learning model prediction sample label | |
CN104794367A (en) | Hospitalizing resource scoring and recommending method based on latent factor model | |
CN107679225A (en) | A kind of reply generation method based on keyword | |
CN106202377A (en) | A kind of online collaborative sort method based on stochastic gradient descent |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180316 |
|
WW01 | Invention patent application withdrawn after publication |