CN106780073A - A kind of community network maximizing influence start node choosing method for considering user behavior and emotion - Google Patents
A kind of community network maximizing influence start node choosing method for considering user behavior and emotion Download PDFInfo
- Publication number
- CN106780073A CN106780073A CN201710019177.XA CN201710019177A CN106780073A CN 106780073 A CN106780073 A CN 106780073A CN 201710019177 A CN201710019177 A CN 201710019177A CN 106780073 A CN106780073 A CN 106780073A
- Authority
- CN
- China
- Prior art keywords
- emotion
- node
- word
- comment
- participle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000008451 emotion Effects 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000011156 evaluation Methods 0.000 claims abstract description 30
- 230000002996 emotional effect Effects 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000004458 analytical method Methods 0.000 claims abstract description 9
- 230000006399 behavior Effects 0.000 claims description 64
- 210000003813 thumb Anatomy 0.000 claims description 9
- 230000007306 turnover Effects 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 6
- 241000270322 Lepidosauria Species 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000001902 propagating effect Effects 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims 1
- 230000001934 delay Effects 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 11
- 230000000694 effects Effects 0.000 abstract description 10
- 102100035345 Cerebral dopamine neurotrophic factor Human genes 0.000 description 10
- 101000737775 Homo sapiens Cerebral dopamine neurotrophic factor Proteins 0.000 description 10
- 238000004891 communication Methods 0.000 description 7
- 238000013480 data collection Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of community network maximizing influence start node choosing method for considering user behavior and emotion, use the evaluation data set with tag along sort, build emotional words training pattern and extend existing sentiment dictionary, behavior disposition analysis and Sentiment orientation analysis are carried out to community network node respectively, as the judging basis of node influence power, the influence power propagation model BSIS for considering user behavior and emotion is built, and maximum effect power marginal benefit node is solved with reference to greedy algorithm and be added in start node set.The present invention consider user behavior tendency and Sentiment orientation, it is more efficient, accurate, truly excavate maximizing influence start node.
Description
Technical field
The invention belongs to field of computer technology, it is related to the community network influence power of a kind of consideration user behavior and emotion most
Bigization start node choosing method.
Background technology
As community network is fast-developing, more and more people are by societies such as microblogging, circle of friends, Facebook, Twitter
Net list up to oneself to other people, product, social event etc. personal views and view, exchange is shared with other people, by oneself
In work, life, affect display to community network.In community network, the information of the exchange and interdynamic between user illustrates use
Sentiment orientation between family, the emotion between usual user is directly connected to the factor of influence between user.According to community network
The emotion influence relation of middle user, finds the maximum user group of influence power so that information obtains virus-type with minimum cost
Propagate, influence power is spread in the way of " public praise effect ".Therefore sentiment analysis are carried out to community network user for influence power most
Bigization research is significant.
Innovatory algorithm and model quilt of many based on IC models (independent cascade model) and LT models (linear threshold model)
Propose to find maximizing influence start node.When influence power is propagated, user feeling was analyzed and based on the time in community network
The user behavior tendency of delay is seldom considered.
Therefore, it is necessary to provide a kind of community network maximizing influence start node choosing for considering user behavior and emotion
Take method.
The content of the invention
Technical problem solved by the invention is in view of the shortcomings of the prior art, to propose a kind of consideration user behavior and feelings
The community network maximizing influence start node choosing method of sense, based on community network user behavior tendency and Sentiment orientation,
Maximum marginal benefit node is solved as maximizing influence start node, it is as a result more accurate and reliable.
The technical scheme is that：
A kind of community network maximizing influence start node choosing method for considering user behavior and emotion, including it is following
Step：
Step 1：Using the evaluation data set with tag along sort, emotional words training pattern is built, in existing emotion
The emotional words vocabulary with emotion weight for obtaining being applied to network comment information sentiment analysis on the basis of dictionary extends
To new sentiment dictionary；
Because the evaluation information of community network is all without label, therefore using information on commodity comment, root in electric business network
Emotional semantic classification is carried out to evaluation information according to star is evaluated, 5 stars are evaluated for front, 1 star is unfavorable ratings；Herein, utilize
Reptile crawls Jingdone district information on commodity comment, extracts commodity and evaluates positive negative sample, and keeps positive and negative sample size balanced.Evaluation information
It is as follows that emotion vocabulary builds model step：
Commodity evaluation text is carried out Chinese word segmentation by step 1.1 using ANSJ, obtains indicating the word segmentation result of word part of speech,
Statistics part of speech is the number of times that adjectival participle occurs in forward direction evaluates sample and negative sense evaluates sample respectively；
Step 1.2 counts positive and evaluates sample and negative sense evaluation sample size summation N, sets participle frequency threshold value r, maximum
Emotion weight max；
Step 1.3 is adjectival participle word for part of speech, counts its occurrence number N in forward direction evaluates samplePWith
Negative sense evaluates the times N occurred in sampleNIf,Perform step 1.4,1.5,1.6；
Step 1.4 calculates the positive probability P of word wordP:
Wherein PP∈(0,1)；
Step 1.5 is according to its emotion weight of the positive probability calculation of participle word w1And w2：
Wherein PP∈ (0,1), then w1∈(-1,1)；
Wherein w2∈(-1,1)；
Step 1.6 gives adjective participle word, and its comprehensive emotion weight is w：
Step 2：Data prediction and data pick-up are carried out to community network data set, user behavior record is extracted, calculated
Consider the user behavior influence power of user behavior time delay；
In community network, if user node u and v are adjacent node, and the message that v is issued to u is by a series of times
Postpone to produce a series of behaviors, wherein behavior is including thumb up, comment and forwards, then consider user's row of user behavior time delay
For influence power B_Inf (u, v) is：
Wherein, λ ∈ (0,1), M (u) represents that u performed the massage set of behavior, and M (v) represents that user v performed behavior
Massage set, | M (u) | represents the message bar number that u performed behavior, and | M (u) ∩ M (v) | represents that v performs row to the message that u is issued
For bar number, t represent v pairs and u issue message process performing average time delay, T represent v pairs with its all neighbor node
The average time delay of the message process performing of issue.
Step 3：According to expanded sentiment dictionary, sentiment analysis are carried out to comment information between community network user,
Obtain the user feeling influence power based on user feeling tendency；
Two neighboring users node u and v are given, if u issues first message, v is evaluated it, and wherein one is evaluated
Content commentjCarry out participle and obtain a series of participle set { word1,word2,...,wordn, go search each participle exist
Corresponding participle emotion weight { w in sentiment dictionary1,w2,...,wn, ifSentiment dictionary, wi=0, then in this evaluation
The emotion score value of appearance：
What deserves to be explained is, in comment information often occur with turnover meaning word, for example Chinese in " no
Be, no " and English in the adverbial word such as " not ", this kind of word combined the emotion participle expression that it can be caused to modify with other emotion participles
The meaning it is reverse；Accordingly, it is considered to arrive the semantic modified adverbial word of turnover, we analyze the context of this kind of word, will be with its phase
Adjacent emotion participle is extracted and obtains turnover participle set { reverse_w1,reverse_w2,...,reverse_wm, go
Search their emotion weight { r_w respectively in sentiment dictionary1,r_w2,...,r_wm, to commentjEmotion score value enter
The emotion score value that row is corrected：
Therefore, if evaluation content commentjIn the positive emotion word that includes it is more, participle emotion weight accumulated value is higher
And it is denseer to the positive emotion of u just, to represent v, know that can to derive u bigger to the influence power of v therefrom.If evaluation content
commentjIn the negative sense emotion word that includes it is more, negative more of participle emotion weight accumulated value and be negative represent v to u negative senses
Emotion is denseer, thus derives u bigger to the influence power of v.Therefore, two neighboring user node u and v, v first commenting to u are given
By commentj, based on commentjEmotion score value derive u to the single emotion influence power of v：
Wherein ο ∈ (0,1)；
Therefore comment information { comment of two neighboring user node u and v and a plurality of v to u is given1,
comment2,...,commentn, the node emotion influence power being calculated is commented on according to every, average and draw u to v's
Emotion influence power：
Step 4：Consider the influence power of the Sentiment orientation based on user behavior tendency, obtain the total influence power of user
TotalInfu,v：
TotalInfu,v=β B_Infu,v+(1-β)SentiInfu,v；
Step 5：The total influence power and the propagation path of total influence power being calculated according to step 4, consider social network comprehensively
The topological relation of network node, therefore u is φ to the propagating influence summation of vu,v(v)：
There is the node set of behavior and emotion influence power in wherein I (v) expression network topology structures to v；
Step 6：The total influence power marginal benefits of calculate node u are σMarginal_Revenue(u)：
Wherein, S represents start node set, and A represents the set of all user nodes in community network；
Community network interior joint marginal benefit is calculated simultaneously to sort by size, choose the maximum preceding nk node of marginal benefit according to
Secondary to be inserted into queue Q, marginal benefit maximum node is inserted into S in ejection Q；
Step 7：Statistics S interior joint numbers, if | S | ＜ k, all node limits are received in recalculating queue Q according to step 6
Benefit and more new sort, marginal benefit maximum node is inserted into S in ejection Q, if | S |=k, S are final start node collection
Close；
Beneficial effect：
The present invention proposes a kind of community network maximizing influence start node selection for considering user behavior and emotion
Method.Based on user behavior and Sentiment orientation in online community network, proposition is a kind of to consider two kinds of influence powers of factor
Propagation model, is named as BSIS (Influence Spread Based on Users ' Behaviors and Sentiment)
Model.The model is divided into three phases, and the first stage builds emotional words training pattern expanding sentiment dictionary, and second stage builds
User behavior influence power, the phase III builds user feeling influence power.With reference to greedy algorithm and the maximum limit of BSIS model solutions
Income node is used as influence power maximum start node.Commodity evaluating data is crawled from Jingdone district, is divided into evaluation according to star is evaluated
Front is evaluated and the class of unfavorable ratings two, based on the existing sentiment dictionary of emotional words training pattern extension.Using Flickr and microblogging
Social network data collection, realizes model proposed by the present invention and algorithm on Hadoop and Spark platforms.Test result indicate that,
Influence power propagation model proposed by the present invention and algorithm compared to the method based on traditional IC, LT and CDNF models, with more preferable
Influence power communication effect, the start node of wider array of influence power spread scope and Geng Gao chooses quality.
Brief description of the drawings
Fig. 1 is a kind of community network maximizing influence start node for considering user behavior and emotion proposed by the present invention
The flow chart of choosing method；
Fig. 2 is to use in embodiment 1 Flickr data sets to be based on BSIS, and the different methods of this 4 kinds of CDNF, IC and LT are chosen
Start node influence power communication effect comparison diagram；
Fig. 3 is to use in embodiment 1 microblog data collection to be based on the influence power that the start node that BSIS chooses produces to propagate effect
Fruit is schemed；
Fig. 4 is to use in embodiment 1 Flickr data sets to be based on BSIS, and the different methods of this 4 kinds of CDNF, IC and LT are chosen
The thumb up of reception of start node, forwarding, comment behavior perform number of times comparison diagram；
Fig. 5 be use in embodiment 1 microblog data collection be based on the thumb up of the reception of the start node that BSIS chooses, forwarding,
Comment behavior performs number of times figure；
Specific embodiment
For a kind of more detailed description community network influence power for considering user behavior and emotion proposed by the present invention
Start node choosing method is maximized, the present invention is further illustrated with example below in conjunction with the accompanying drawings.
The present invention proposes a kind of community network maximizing influence start node selection for considering user behavior and emotion
Method.Based on user behavior and Sentiment orientation in online community network, a kind of influence power for considering both factors is proposed
Propagation model, is named as BSIS (Influence Spread Based on Users ' Behaviors and Sentiment)
Model.The model is divided into three phases, and the first stage builds emotional words training pattern expanding sentiment dictionary, and second stage builds
User behavior influence power, the phase III builds user feeling influence power.With reference to greedy algorithm and the maximum limit of BSIS model solutions
Income node is used as influence power maximum start node.Commodity evaluating data is crawled from Jingdone district website, will be evaluated according to star is evaluated
It is divided into front to evaluate and the class of unfavorable ratings two, based on the existing sentiment dictionary of emotional words training pattern extension.Using Flickr and
Microblogging social network data collection, realizes model proposed by the present invention and algorithm on Hadoop and Spark platforms.Experimental result table
It is bright, influence power propagation model proposed by the present invention and algorithm compared to the method on traditional IC, LT and CDNF models, with more preferable
Influence power communication effect, the start node of wider array of influence power spread scope and Geng Gao chooses quality.
Fig. 1 is a kind of community network maximizing influence start node for considering user behavior and emotion proposed by the present invention
Choosing method flow chart, specific implementation step is as follows：
Step 1：Using the evaluation data set with tag along sort, emotional words training pattern is built, in existing emotion
The emotional words vocabulary with emotion weight for obtaining being applied to network comment information sentiment analysis on the basis of dictionary extends
To new sentiment dictionary；
Because the evaluation information of community network is all without label, therefore using information on commodity comment, root in electric business network
Emotional semantic classification is carried out to evaluation information according to star is evaluated, 5 stars are evaluated for front, 1 star is unfavorable ratings；Herein, utilize
Reptile crawls Jingdone district information on commodity comment, extracts commodity and evaluates positive negative sample, and keeps positive and negative sample size balanced.Evaluation information
It is as follows that emotion vocabulary builds model step：
Commodity evaluation text is carried out Chinese word segmentation by step 1.1 using ANSJ, obtains indicating the word segmentation result of word part of speech,
Statistics part of speech is the number of times that adjectival participle occurs in forward direction evaluates sample and negative sense evaluates sample respectively；
Step 1.2 counts positive and evaluates sample and negative sense evaluation sample size summation N, sets participle frequency threshold value r, maximum
Emotion weight max；
Step 1.3 is adjectival participle word for part of speech, counts its occurrence number N in forward direction evaluates samplePWith
Negative sense evaluates the times N occurred in sampleNIf,Perform step 1.4,1.5,1.6；
Step 1.4 calculates the positive probability P of word wordP:
Wherein PP∈(0,1)；
Step 1.5 is according to its emotion weight of the positive probability calculation of participle word w1And w2：
Wherein PP∈ (0,1), then w1∈(-1,1)；
Wherein w2∈(-1,1)；
Step 1.6 gives adjective participle word, and its comprehensive emotion weight is w：
Step 2：Data prediction and data pick-up are carried out to community network data set, user behavior record is extracted, calculated
Consider the user behavior influence power of user behavior time delay；
In community network, if user node u and v are adjacent node, and the message that v is issued to u is by a series of times
Postpone to produce a series of behaviors, wherein behavior is including thumb up, comment and forwards, then consider user's row of user behavior time delay
For influence power B_Inf (u, v) is：
Wherein, λ ∈ (0,1), M (u) represents that u performed the massage set of behavior, and M (v) represents that user v performed behavior
Massage set, | M (u) | represents the message bar number that u performed behavior, and | M (u) ∩ M (v) | represents that v performs row to the message that u is issued
For bar number, t represent v pairs and u issue message process performing average time delay, T represent v pairs with its all neighbor node
The average time delay of the message process performing of issue.
Step 3：According to expanded sentiment dictionary, sentiment analysis are carried out to comment information between community network user,
Obtain the user feeling influence power based on user feeling tendency；
Two neighboring users node u and v are given, if u issues first message, v is evaluated it, and wherein one is evaluated
Content commentjCarry out participle and obtain a series of participle set { word1,word2,...,wordn, go search each participle exist
Corresponding participle emotion weight { w in sentiment dictionary1,w2,...,wn, ifSentiment dictionary, wi=0, then in this evaluation
The emotion score value of appearance：
What deserves to be explained is, in comment information often occur with turnover meaning word, for example Chinese in " no
Be, no " and English in the adverbial word such as " not ", this kind of word combined the emotion participle expression that it can be caused to modify with other emotion participles
The meaning it is reverse；Accordingly, it is considered to arrive the semantic modified adverbial word of turnover, we analyze the context of this kind of word, will be with its phase
Adjacent emotion participle is extracted and obtains turnover participle set { reverse_w1,reverse_w2,...,reverse_wm, go
Search their emotion weight { r_w respectively in sentiment dictionary1,r_w2,...,r_wm, to commentjEmotion score value enter
The emotion score value that row is corrected：
Therefore, if evaluation content commentjIn the positive emotion word that includes it is more, participle emotion weight accumulated value is higher
And it is denseer to the positive emotion of u just, to represent v, know that can to derive u bigger to the influence power of v therefrom.If evaluation content
commentjIn the negative sense emotion word that includes it is more, negative more of participle emotion weight accumulated value and be negative represent v to u negative senses
Emotion is denseer, thus derives u bigger to the influence power of v.Therefore, two neighboring user node u and v, v first commenting to u are given
By commentj, based on commentjEmotion score value derive u to the single emotion influence power of v：
Wherein ο ∈ (0,1)；
Therefore comment information { comment of two neighboring user node u and v and a plurality of v to u is given1,
comment2,...,commentn, the node emotion influence power being calculated is commented on according to every, take
Average value draws emotion influence powers of the u to v：
Step 4：Consider the influence power of the Sentiment orientation based on user behavior tendency, obtain the total influence power of user
TotalInfu,v：
TotalInfu,v=β B_Infu,v+(1-β)SentiInfu,v；
Step 5：The total influence power and the propagation path of total influence power being calculated according to step 4, consider social network comprehensively
The topological relation of network node, therefore u is φ to the propagating influence summation of vu,v(v)：
There is the node set of behavior and emotion influence power in wherein I (v) expression network topology structures to v；
Step 6：The total influence power marginal benefits of calculate node u are σMarginal_Revenue(u)：
Wherein, S represents start node set, and A represents the set of all user nodes in community network；
Community network interior joint marginal benefit is calculated simultaneously to sort by size, choose the maximum preceding nk node of marginal benefit according to
Secondary to be inserted into queue Q, marginal benefit maximum node is inserted into S in ejection Q；
Step 7：Statistics S interior joint numbers, if | S | ＜ k, all node limits are received in recalculating queue Q according to step 6
Benefit and more new sort, marginal benefit maximum node is inserted into S in ejection Q, if | S |=k, S are final start node collection
Close；
Embodiment 1：
In this embodiment, crawl Jingdone district using reptile and do shopping network user comment data as emotional words training pattern
Training set and test set, this data set evaluates and 14937 1 star unfavorable ratings comprising 17052 5 star fronts.The present embodiment
Emotional words training pattern is built based on step 1, is gathered using 80% as training, 20% gathers as test, final training
Obtain 468 emotional words with weight extend existing Taiwan Univ. Chinese emotion vocabulary NTUSD with
SentiWordNet English emotion vocabularys.
In this embodiment, using Flickr (one, U.S. picture shares social network sites) social network data collection, the number
Recorded comprising 40808 user nodes and 75269 user behaviors according to collection.Microblogging community network data set, the number are used simultaneously
Recorded comprising 412952 user nodes and 500977 user behaviors according to collection.
The embodiment is real on Hadoop2.5.2 and Spark2.0.2 (12Core, 120GB Memory, 6Workers)
It is existing, it is effectively quick BSIS influence powers propagation model and start node derivation algorithm to be realized, it is calculated maximizing influence
Start node set.Using Liang Ge Fiel meeting Network data set, from influence power communication effect and user's true impact power scope
Two aspect comparison of design experiments, by our result and traditional IC model (independent cascade model), LT models (linear threshold mould
Type) and CDNF models (the credit distributed model based on nodal properties) contrast relatively come the correctness and validity of verification method.
Side activation probability wherein in IC models between consecutive points is learnt by EM algorithms (EM algorithm), and LT models
The side of adjacent node v and u is activated Probability p p (u, v) and is calculated using formula pp (u, v)=1/N (u), and wherein N (u) represents section
The neighbor node number of point u.
As shown in Figure 2, left subgraph represented in Flickr community networks, is asked based on tetra- kinds of models of SBIS, CDNF, IC, LT
Total influence power distribution map for being calculated of preceding 50 start nodes of solution, right subgraph is represented based on tetra- kinds of SBIS, CDNF, IC, LT
The distribution map that total influence power summation of model solution start node increases with start node number.From the figure 3, it may be seen that left and right sub-chart
Show in microblogging community network, total influence distribution map and total influence power of preceding 50 start nodes based on SBIS model solutions are total
And growth chart.Represented by the checkings of Fig. 2 and 3, the behavial factor of user feeling, the start node tool of selection are considered due to SBIS
There is bigger influence power communication effect.
In order to verify the real influence power that the start node that SBIS models are chosen is produced in true community network, this reality
Thumb up, forwarding, the comment behavior number of times of the other users execution that example is received by the use of user are applied as true impact power criterion,
Statistics chooses the real behavior execution time that start node set is received based on tetra- influence power propagation models of SBIS, CDNF, IC, LT
Number.As shown in Figure 4, using Flickr data sets, left subgraph represents the first 50 initial sections chosen based on SBIS, CDNF, IC, LT
Point receive thumb up, forwarding, comment behavior perform number of times distribution map, right subgraph represent use Flickr data sets, based on SBIS,
CDNF, IC, LT choose thumb up, forwarding, the comment behavior execution number of times summation of start node reception with the growth of start node number
Growth chart.As shown in Figure 5, left and right subgraph represents use microblog data collection, first 50 initial sections of BSIS model solutions respectively
Thumb up, forwarding, comment behavior execution number of times distribution map and execution number of times summation growth chart that point is received.As seen from the figure, it is based on
It is higher than other models that the behavior that the start node that SBIS influence powers propagation model is chosen is received performs number of times, therefore compared to other
Model, SBIS can effectively excavate popular user, so as to ensure that the selection quality of start node, be produced in true community network
Wider array of influence power spread scope, more accurately reflects influence power communication effect.
Knowable to being tested more than, at the beginning of the inventive method is based on user behavior and Sentiment orientation factor excavation maximizing influence
Beginning node, can more truly reflect user force, and experiment shows that the inventive method chooses quality, influence in start node
Power communication effect and influence power spread scope aspect have accuracy and reliability.
Claims (1)
1. a kind of community network maximizing influence start node choosing method for considering user behavior and emotion, its feature exists
In comprising the following steps：
Step 1：Using the evaluation data set with tag along sort, emotional words training pattern is built, in existing sentiment dictionary
On the basis of obtain being applied to the extension of the emotional words vocabulary with emotion weight of network comment information sentiment analysis and obtain new
Sentiment dictionary；
Because the evaluation information of community network is all without label, therefore using information on commodity comment in electric business network, according to commenting
Valency star carries out emotional semantic classification to evaluation information, and 5 stars are evaluated for front, and 1 star is unfavorable ratings；Herein, using reptile
Jingdone district information on commodity comment is crawled, commodity is extracted and is evaluated positive negative sample, and keep positive and negative sample size balanced；Evaluation information emotion
It is as follows that vocabulary builds model step：
Commodity evaluation text is carried out Chinese word segmentation by step 1.1 using ANSJ, obtains indicating the word segmentation result of word part of speech, is counted
Part of speech is the number of times that adjectival participle occurs in forward direction evaluates sample and negative sense evaluates sample respectively；
Step 1.2 counts positive and evaluates sample and negative sense evaluation sample size summation N, sets participle frequency threshold value r, maximum emotion
Weight max；
Step 1.3 is adjectival participle word for part of speech, counts its occurrence number N in forward direction evaluates samplePCommented with negative sense
The times N occurred in valency sampleNIf,Perform step 1.4,1.5,1.6；
Step 1.4 calculates the positive probability P of word wordP:
Wherein PP∈(0,1)；
Step 1.5 is according to its emotion weight of the positive probability calculation of participle word w1And w2：
Wherein PP∈ (0,1), then w1∈(-1,1)；
Wherein w2∈(-1,1)；
Step 1.6 gives adjective participle word, and its comprehensive emotion weight is w：
Step 2：Data prediction and data pick-up are carried out to community network data set, user behavior record is extracted, is calculated and is considered
The user behavior influence power of user behavior time delay；
In community network, if user node u and v are adjacent node, and the message that v is issued to u is by a series of time delays
Producing a series of behaviors, wherein behavior includes thumb up, comment and forwards, then consider the user behavior shadow of user behavior time delay
Ringing power B_Inf (u, v) is：
Wherein, λ ∈ (0,1), M (u) represents that u performed the massage set of behavior, and M (v) represents that user v performed the message of behavior
Set, | M (u) | represents the message bar number that u performed behavior, and | M (u) ∩ M (v) | represents the message process performing that v is issued to u
Bar number, t represents the v pairs of average time delay of the message process performing issued with u, and T represents v pairs with its all neighbor node issue
Message process performing average time delay；
Step 3：According to expanded sentiment dictionary, sentiment analysis are carried out to comment information between community network user, obtained
User feeling influence power based on user feeling tendency；
Two neighboring users node u and v are given, if u issues first message, v is evaluated it, to wherein one evaluation content
commentjCarry out participle and obtain a series of participle set { word1,word2,...,wordn, remove to search each participle in emotion
Corresponding participle emotion weight { w in dictionary1,w2,...,wn, ifwi=0, then feelings of this evaluation content
Sense score value：
What deserves to be explained is, the word with turnover meaning often occurs in comment information, such as in Chinese " it is not,
Adverbial word such as " not " or not and in English, this kind of word is combined the meaning that the emotion participle that it can be caused to modify is expressed with other emotion participles
Think reverse；Accordingly, it is considered to arrive the semantic modified adverbial word of turnover, we analyze the context of this kind of word, will be adjacent thereto
Emotion participle is extracted and obtains turnover participle set { reverse_w1,reverse_w2,...,reverse_wm, go to search
Their emotion weight { r_w respectively in sentiment dictionary1,r_w2,...,r_wm, to commentjEmotion score value repaiied
The emotion score value just corrected：
Therefore, if evaluation content commentjIn the positive emotion word that includes it is more, participle emotion weight accumulated value is higher and is
Just, the positive emotion for representing v to u is denseer, knows that can to derive u bigger to the influence power of v therefrom；If evaluation content commentj
In the negative sense emotion word that includes it is more, negative more of participle emotion weight accumulated value and be negative, it is denseer to u negative sense emotions to represent v,
Thus u is derived bigger to the influence power of v；Therefore, the first comment of two neighboring user node u and v, v to u is given
commentj, based on commentjEmotion score value derive u to the single emotion influence power of v：
Wherein ο ∈ (0,1)；
Therefore comment information { comment of two neighboring user node u and v and a plurality of v to u is given1,comment2,...,
commentn, the node emotion influence power being calculated, the emotion influence power for drawing u to v of averaging are commented on according to every：
Step 4：Consider the influence power of the Sentiment orientation based on user behavior tendency, obtain the total influence power of user
TotalInfu,v：
TotalInfu,v=β β _ Infu,v+(1-β)·SentiInfu,v；
Step 5：The total influence power and the propagation path of total influence power being calculated according to step 4, consider community network section comprehensively
The topological relation of point, therefore u is φ to the propagating influence summation of vu,v(v)：
There is the node set of behavior and emotion influence power in wherein I (v) expression network topology structures to v；
Step 6：The total influence power marginal benefits of calculate node u are σMarginal_Revenue(u)：
Wherein, S represents start node set, and A represents the set of all user nodes in community network；
Calculate community network interior joint marginal benefit and sort by size, choose the maximum preceding nk node of marginal benefit and insert successively
Enter to queue Q, marginal benefit maximum node is inserted into S in ejection Q；
Step 7：Statistics S interior joint numbers, if | S | ＜ k, in recalculating queue Q according to step 6, all node marginal benefits are simultaneously
More new sort, marginal benefit maximum node is inserted into S in ejection Q, if | S |=k, S are final start node set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710019177.XA CN106780073B (en) | 2017-01-11 | 2017-01-11 | Social network influence maximization initial node selection method considering user behaviors and emotions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710019177.XA CN106780073B (en) | 2017-01-11 | 2017-01-11 | Social network influence maximization initial node selection method considering user behaviors and emotions |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106780073A true CN106780073A (en) | 2017-05-31 |
| CN106780073B CN106780073B (en) | 2021-05-25 |
Family
ID=58949265
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710019177.XA Active CN106780073B (en) | 2017-01-11 | 2017-01-11 | Social network influence maximization initial node selection method considering user behaviors and emotions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106780073B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460010A (en) * | 2018-01-17 | 2018-08-28 | 南京邮电大学 | A kind of comprehensive grade model implementation method based on sentiment analysis |
| CN108549632A (en) * | 2018-04-03 | 2018-09-18 | 重庆邮电大学 | A kind of social network influence power propagation model construction method based on sentiment analysis |
| CN108804524A (en) * | 2018-04-27 | 2018-11-13 | 成都信息工程大学 | Emotion based on stratification taxonomic hierarchies differentiates and importance division methods |
| CN110138619A (en) * | 2019-05-28 | 2019-08-16 | 湖南大学 | A kind of start node choosing method that realizing maximizing influence and system |
| CN110598960A (en) * | 2018-05-23 | 2019-12-20 | 北京国双科技有限公司 | Entity-level emotion assessment method and device |
| CN110727881A (en) * | 2019-10-23 | 2020-01-24 | 精硕科技（北京）股份有限公司 | Method and device for determining target public figure |
| CN110738421A (en) * | 2019-10-17 | 2020-01-31 | 西南大学 | method for measuring influence of multi-layer network users based on shortest propagation path |
| CN112052995A (en) * | 2020-08-31 | 2020-12-08 | 杭州电子科技大学 | Social network user influence prediction method based on fusion emotional tendency theme |
| CN112487304A (en) * | 2020-11-26 | 2021-03-12 | 杭州电子科技大学 | Method for establishing influence propagation model based on viewpoint vectorization |
| CN113688202A (en) * | 2021-07-30 | 2021-11-23 | 杭州网易云音乐科技有限公司 | Emotion polarity analysis method and device, electronic equipment and computer storage medium |
| CN116245555A (en) * | 2023-03-09 | 2023-06-09 | 清瑞网络科技(山东)有限责任公司 | User information collecting and analyzing system based on big data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101770487A (en) * | 2008-12-26 | 2010-07-07 | 聚友空间网络技术有限公司 | Method and system for calculating user influence in social network |
| CN103530360A (en) * | 2013-10-12 | 2014-01-22 | 广西师范学院 | Network society influence maximization algorithm based on microblog text affective computing |
-
2017
- 2017-01-11 CN CN201710019177.XA patent/CN106780073B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101770487A (en) * | 2008-12-26 | 2010-07-07 | 聚友空间网络技术有限公司 | Method and system for calculating user influence in social network |
| CN103530360A (en) * | 2013-10-12 | 2014-01-22 | 广西师范学院 | Network society influence maximization algorithm based on microblog text affective computing |
Non-Patent Citations (1)
| Title |
|---|
| XIAOHENG DENG等: "Credit Distribution and Influence Maximization in", 《2015 12TH INTERNATIONAL CONFERENCE ON FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460010A (en) * | 2018-01-17 | 2018-08-28 | 南京邮电大学 | A kind of comprehensive grade model implementation method based on sentiment analysis |
| CN108549632A (en) * | 2018-04-03 | 2018-09-18 | 重庆邮电大学 | A kind of social network influence power propagation model construction method based on sentiment analysis |
| CN108804524B (en) * | 2018-04-27 | 2020-03-27 | 成都信息工程大学 | Emotion distinguishing and importance dividing method based on hierarchical classification system |
| CN108804524A (en) * | 2018-04-27 | 2018-11-13 | 成都信息工程大学 | Emotion based on stratification taxonomic hierarchies differentiates and importance division methods |
| CN110598960B (en) * | 2018-05-23 | 2022-06-03 | 北京国双科技有限公司 | Entity-level emotion assessment method and device |
| CN110598960A (en) * | 2018-05-23 | 2019-12-20 | 北京国双科技有限公司 | Entity-level emotion assessment method and device |
| CN110138619B (en) * | 2019-05-28 | 2020-05-19 | 湖南大学 | Initial node selection method and system for realizing influence maximization |
| CN110138619A (en) * | 2019-05-28 | 2019-08-16 | 湖南大学 | A kind of start node choosing method that realizing maximizing influence and system |
| CN110738421A (en) * | 2019-10-17 | 2020-01-31 | 西南大学 | method for measuring influence of multi-layer network users based on shortest propagation path |
| CN110738421B (en) * | 2019-10-17 | 2023-08-22 | 西南大学 | Multilayer network user influence measuring method based on shortest propagation path |
| CN110727881A (en) * | 2019-10-23 | 2020-01-24 | 精硕科技（北京）股份有限公司 | Method and device for determining target public figure |
| CN112052995A (en) * | 2020-08-31 | 2020-12-08 | 杭州电子科技大学 | Social network user influence prediction method based on fusion emotional tendency theme |
| CN112487304A (en) * | 2020-11-26 | 2021-03-12 | 杭州电子科技大学 | Method for establishing influence propagation model based on viewpoint vectorization |
| CN112487304B (en) * | 2020-11-26 | 2022-05-03 | 杭州电子科技大学 | Method for establishing influence propagation model based on viewpoint vectorization |
| CN113688202A (en) * | 2021-07-30 | 2021-11-23 | 杭州网易云音乐科技有限公司 | Emotion polarity analysis method and device, electronic equipment and computer storage medium |
| CN113688202B (en) * | 2021-07-30 | 2024-03-15 | 杭州网易云音乐科技有限公司 | Emotion polarity analysis method and device, electronic equipment and computer storage medium |
| CN116245555A (en) * | 2023-03-09 | 2023-06-09 | 清瑞网络科技(山东)有限责任公司 | User information collecting and analyzing system based on big data |
| CN116245555B (en) * | 2023-03-09 | 2023-12-08 | 张家口巧工匠科技服务有限公司 | User information collecting and analyzing system based on big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106780073B (en) | 2021-05-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106780073A (en) | A kind of community network maximizing influence start node choosing method for considering user behavior and emotion | |
| CN106980692B (en) | Influence calculation method based on microblog specific events | |
| Barbosa et al. | Robust sentiment detection on twitter from biased and noisy data | |
| CN108628833B (en) | Method and device for determining summary of original content and method and device for recommending original content | |
| CN103577989B (en) | A kind of information classification approach and information classifying system based on product identification | |
| CN110781668B (en) | Text information type identification method and device | |
| CN106940732A (en) | A kind of doubtful waterborne troops towards microblogging finds method | |
| Mhatre et al. | Dimensionality reduction for sentiment analysis using pre-processing techniques | |
| CN105183717B (en) | A kind of OSN user feeling analysis methods based on random forest and customer relationship | |
| CN104881458B (en) | A kind of mask method and device of Web page subject | |
| CN109815485B (en) | Method and device for identifying emotion polarity of microblog short text and storage medium | |
| CN105095419A (en) | Method for maximizing influence of information to specific type of weibo users | |
| WO2008022581A1 (en) | Method and device for obtaining the new words and input method system | |
| CN108960574A (en) | Quality determination method, device, server and the storage medium of question and answer | |
| CN113849598B (en) | Social media false information detection method and detection system based on deep learning | |
| CN105893484A (en) | Microblog Spammer recognition method based on text characteristics and behavior characteristics | |
| CN102789449A (en) | Method and device for evaluating comment text | |
| CN110909230A (en) | Network hotspot analysis method and system | |
| CN106156117A (en) | Hidden community core communication circle detection towards particular topic finds method and system | |
| Sadiq et al. | High dimensional latent space variational autoencoders for fake news detection | |
| Yao et al. | Online deception detection refueled by real world data collection | |
| De Zarate et al. | Vocabulary-Based Method for Quantifying Controversy in Social Media. | |
| Modi et al. | Sentiment analysis of Twitter feeds using flask environment: A superior application of data analysis | |
| CN106933798B (en) | Information analysis method and device | |
| Wu et al. | Reducing noisy labels in weakly labeled data for visual sentiment analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |