CN103218477B - Network viewpoint propagation and Forecasting Methodology - Google Patents

Abstract

The invention discloses a kind of network viewpoint propagation and Forecasting Methodology, the method includes: network topological information is digitized modeling, carries out digital quantization for each individual viewpoint state；According to the quantitative information obtained, set up forecast model based on discrete individual interbehavior；This method can be used to predict the integral viewpoint trend of the network user, at the environment giving network user's state and network topology structure, the method of the present invention can the tendency of certain viewpoint in more efficiently prediction network, and then the Mutual Influence Law between the prediction of network public opinion, the coping strategy of accident and the network user and viewpoint tendency is made summary.

Description

Network viewpoint propagation and Forecasting Methodology

Technical field

The present invention relates to network viewpoint propagation and Forecasting Methodology, be based particularly on Modified Discrete The network viewpoint propagation of behavior and continuous viewpoint AC characteristic and Forecasting Methodology, belong to network letter Breath technical field.

Background technology

In recent years, the impact of public opinion is constantly put by mass media such as such as the Internets Greatly.By Internet technology, the On-line funchon such as chat software, network forum is the most popular, Personal comminication between the network user with user communicates and becomes simple.Public opinion is never The individual directly exchange of user and the interaction that communicates is depended on as today.The phase of complex network Close theory and similar behavior can be made description and research.

By the method for statistical physics and theoretical (such as public opinion kinetics, multinode modeling Method and Monte Carlo emulation mode), the problem of many complex networks can be by successfully Solve.In viewpoint kinetics, the interaction between node with exchange, promote network integral viewpoint Develop, be a kind of existing effective ways.In present existing viewpoint model, most models Assume a series of individuality or the node ultimate unit as network, the most a certain between node Topic or information are launched mutually exchange and are discussed, and by this behavior, update that between node This viewpoint, and try hard to advise other nodes to admit oneself.

The dynamic (dynamical) model of public opinion viewpoint can be divided into two big classes: individual viewpoint value for from Dissipate the model (discrete viewpoint model) of numerical value, and the viewpoint value of individuality is serial number Model (continuous viewpoint model).And CODA model (Continuous Opinions and Discrete Actions) way of binary is introduced viewpoint model, it is assumed that in network Individuality, can be provided simultaneously with discrete viewpoint and select and (or inherent in viewpoint in continuous print Tendentiousness).

In diffusion of innovation theory, the process of Innovation Diffusion, equally regard a kind of sight as Point carries out restraining and the process of viewpoint polymerization in a network.This viewpoint or tendentiousness, Can be individual to the view of some new thought, preference degree to a certain new product, Yi Jiqi The problem that the two-value that he is similar to selects, either-or problem etc..In viewpoint evolution phenomenon In research process, Many researchers have noticed the phenomenon of limited trust.That is: hold Between the individuality of similar views, it is easier to trust each other and occur exchange behavior is discussed.

CODA model and limited trust model are that the research work of network public opinion proposes effectively Method, and there is good versatility, can with the character of self explain many other The phenomenon of model.But, in actual network, along with the development of Internet technology, opinion Popularizing of the new applications such as altar, community, news follow-up, individual releasing object in a network It is not limited in a complex network (or regular network) around oneself present position Limited neighbours are individual.Individual may it is observed that the individual viewpoint of multiple neighbours, even with Apart from far individual viewpoint.

Achievement in research before shows, opinion leader or the structure of network, be to affect network The key factor of overall public opinion tendency.While it is true, recent studies suggest that, system is held in inside The viewpoint colony of one viewpoint, the formation to public opinion plays an important role equally.Many models are neglected Having omited the effect of viewpoint colony, immediate cause is because them and have ignored individuality and can be subject to simultaneously To multiple neighbours affected the fact.

According to network public opinion research before, we learn, the trickle difference of viewpoint evolution rule Different, frequently can lead to distinct result.If for practical situation, model is made Above-mentioned change, the subsequent affect brought is the most unknowable, but certainly, for reality The crucial change that border situation is made, can make the conclusion of method closer to reality.

In sum, the network public opinion analysis means that present stage possesses can not be effective with method Reality rule is made and being reacted accurately by ground, exists the tightest in prediction, deduction function aspects The hysteresis quality of weight, research and development are a kind of actual, accurately, network point prediction side fast and efficiently Method, is the most necessary.

Summary of the invention

In order to overcome the deficiency of art methods, it is an object of the invention to provide based on changing Enter the discrete behavior of type and the network viewpoint propagation of continuous viewpoint AC characteristic and Forecasting Methodology, energy Enough the prediction of network public opinion, the coping strategy of accident are made reference.

The technical solution adopted for the present invention to solve the technical problems is:

Network viewpoint propagation and Forecasting Methodology, it is characterised in that comprise the steps of

1) initialization step: network topological information is digitized modeling, to individual each Individual viewpoint state carries out digital quantization, obtains predicting required network structure and original state；

2) set up forecast model step: according to the init state obtained, determine corresponding pre- Survey rule of inference, it is thus achieved that initialize network and individual data, set up and predict mould accordingly Type；

3) simulation and prediction step, according to built formwork erection type, it was predicted that the viewpoint of the network user develops Tendency.

Described step 1) farther includes:

Initialize network topology structure step, find algorithm according to network topology structure, obtain The physical arrangement of network；

Initialize individual information step, excavate viewpoint state individual in network and to described net Individuality in network carries out digital quantization.

Described step 2) inner according to the data initializing network and individuality obtained, foundation Forecast model is multinode Undirected networks model, if join together or different joining depends on network topology Structure.

Described set up in forecast model step, the foundation step of described multinode Undirected networks model Suddenly include:

Multinode Undirected networks establishment step: according to initialization data, establishes network topology, Set individual nodes number and connection state；

Data substitute into step: set the network information in described model, the data of individual viewpoint；

Model set-up procedure: trained by data, adjusts individual viewpoint value, network size Isoparametric value；

Evolution rule establishment step: definition individual with exchange rule between individuality, set up Kind exchange program.

Described simulation and prediction step includes:

Simulation and prediction step: by set up forecast model step determines data, network topology, Individual parameter brings model into；According to the evolution rule set up, carry out Computer Simulation evolution；

Result statistic procedure: collect prediction conclusion, for follow-up study, analyzes and decision-making offer Support.

In terms of existing technologies, the invention have the advantages that the public opinion before comparing Forecasting Methodology, it is possible to react the composite factor such as group structure, opinion leader more accurately to whole The impact of volume grid public opinion, it is possible to deduce out the development of network public opinion in the regular period accurately Trend.Can be the relevant Decision offer well support of network public opinion, the false speech of reduction, The baneful influence that passive and malice public opinion causes.

Accompanying drawing explanation

Fig. 1 is an illustration of online social networks involved in the present invention, such as list of references 1(list of references 1: the infection phenomenon of social networks, " science and technology China " entry. http://www.techcn.com.cn/index.php?Doc-view-155847.html)；

Fig. 2 is in the present invention, and under the influence of different alpha parameter values, individuality observes neighbours' Behavior, on the impact produced in viewpoint tendentiousness in self next portion individual.Abscissa is individual Body currently inherent tendentiousness, vertical coordinate is the inherent tendentiousness that subsequent time is individual.S=1 table Show that individuality observes that neighbor choice is A, corresponding, S=-1 represents that individuality observes that neighbours select It is selected as B；

Fig. 3 to Fig. 4 is the prediction and statistical result carried out according to this method；

Fig. 4 is statistical result schematic diagram of the present invention.

Detailed description of the invention

When considered in conjunction with the accompanying drawings, by referring to detailed description below, it is possible to more completely more Understand well the present invention and easily learn the advantage that many of which is adjoint, but described herein Accompanying drawing be used for providing a further understanding of the present invention, constitute the part of the present invention.

Understandable for enabling the above-mentioned purpose of the present invention, feature to become apparent from, below in conjunction with attached The present invention is further detailed explanation with detailed description of the invention for figure.

Initialization procedure is divided into following two step:

1, initialize network topology structure step, according to community discovery algorithm, obtain network and open up Flutter structure；The method of concrete acquisition network topology can use prior art, such as with reference to literary composition Offer 2(list of references 2: community structure based on rough set finds algorithm, and red legend lies prostrate by force jade Treasure. " computer engineering " .2011.14) in technology.

2, individual information step is initialized, at complex network and the dynamic (dynamical) research process of viewpoint In, we will often find that: considers many factors meticulously, through further Emulation and experiment, acquired results and do not consider these factors but enter with specific random fashion Row is identical.The effect that this explanation multiple (at random) factor produces is cancelled out each other.This is also Microcosmic (considering many random factors) contacts to the one between macroscopic view (not considering these random factors) Mode.It is therefore possible to appear that when the original state thinking over each individuality Time, it has been found that result and with certain random fashion select original state be identical.Individual letter Breath acquisition methods can use prior art, such as list of references 3(list of references 3: based on The research of the Chinese network comment viewpoint abstracting method of NLP technology, Lou Decheng. Shanghai traffic is big Learn master thesis .2007, TP391.1) in technology.

The result of initialization step gained is exactly individual inclination parameter, network topology and network The initial information of scale or statistical information.The result that initialization procedure obtains is individual inherent sight Point p, 1 ＞ p ＞ 0, represent that the heart of a certain viewpoint is inclined to degree by this individuality.Assume individual i When selecting in the face of the viewpoint of an i.e. B of non-A, may not have and clearly and clearly select, and It is to be inclined to viewpoint A with Probability p, then i is 1-p to the tendency probability of viewpoint B.? In the basic assumption of CODA model, if Probability p ＞ 0.5, illustrate that i is more likely to viewpoint A, Then when i makes viewpoint selection, otherwise A(then i is i.e. selected to select viewpoint B).But individual i Neighbors j can only observe i viewpoint select A, it is impossible to observe the inherent tendentiousness of i p.Colony's AC characteristic based on Modified Discrete behavior Yu continuous conduit, uses this result Predict with viewpoint in network user's modeling.

On the basis of initialization step, the individual networks data obtained according to initialization procedure, Set up forecast model, launch emulation and prediction.

Set up the process of forecast model and be divided into following five steps:

1, complex network establishment step.The structure of typical complex network as it is shown in figure 1, forum, The network of the aspects such as news analysis can present obvious community structure or implicit expression corporations knot Structure, and the structure of microblogging, online social networks (such as Facebook) can be tightr, has Significantly worldlet is with distribution network characteristic.The foundation of complex network, can answer according to existing With software, obtain the topological structure of known network, or use programming software to realize.

2, data substitute into step.Determine the value of individual inclination in described network.According to front Literary composition is analyzed and is understood, and is clustered and semantic analysis technology by Chinese word segmentation, topic, is advised greatly The data of lay wire network individuality and viewpoint distribution proportion, the regularity of distribution, then according to this result pair Individual or node carries out random assignment, is the simplification individuality viewpoint assignment through practical proof The effective means of complexity.

3, model set-up procedure.Adjust topology of networks and individual viewpoint distribution and value, It is allowed to more meet actual environment.

4, evolution rule establishment step, definition is individual and rules of interaction between individuality thus The evolution rule of definition overall network public opinion.Assume the individual process delivering viewpoint in a network It is: observe the selection that neighbours are individual, change the inherent tendentiousness of oneself, and then make viewpoint Select.Such as list of references 4(list of references 4:An opinion dynamics model for the diffusion of innovations,Physica A,A.C.R.Martins,2009vol388No4, Pp.3225 3232) in algorithm.In the method, by Modified Discrete behavior and company Continuous viewpoint algorithm, the observation-development between individuality, the direct friendship being reduced between individuality Flow through journey.

For the ease of describing, the individuality in network is numbered with i, and other parameters numbering is as following Shown in:

p_i: represent the inherent tendentiousness of individual i；

S_i: the external viewpoint representing individual i selects.

O_i: in original discrete behavior continuous viewpoint model, the viewpoint logarithm representing individuality is general Rate, is expressed as:

$O_i\left(n\right)=\log \frac{p_i\left(n\right)}{1-p_i\left(n\right)}---\left(1\right)$

Then individual viewpoint selects S_iI.e. can be expressed as:

S_i(n)=sign (O_i) (2)

In original algorithm, the rules of interaction between definition individuality is as follows:

1), assume that n is that mutual step number occurs, define that all individualities are the most updated is once Complete a step, initial n=0.

2), p_iIt is not the fixed value of individual i, when individual i observes certain neighbours individuality j Viewpoint in the step number n moment selects S_jN ()=1(i.e. selects viewpoint A), then individual i will Change the inherent tendency p of oneself_i, this variation causes the viewpoint of i to change, the most external selection Change.

3) if viewpoint A is optimum selection, then α=P (OA | A) is defined permissible for individual i Observe the probability being chosen as A of its neighbours individuality j.Corresponding, if viewpoint B is Optimum selection, definition β=P (OB | B) be individuality i it is observed that its neighbours individuality j It is chosen as the probability of B.

The log probability rule change of definition individuality is as follows:

$O_i(n+1)=\left\{\begin{array}{ccc}{O}_i\left(n\right)+a,& \mathrm{if}& S_j\left(n\right)=+1\\ O_i\left(n\right)-b,& \mathrm{if}& S_j\left(n\right)=-1\end{array}\right.---\left(3\right)$

Wherein, parameter a, b are defined as:

$a=\log \left(\frac{\mathrm{\α}}{1-\mathrm{\β}}\right)$

$b=\log \left(\frac{\mathrm{\β}}{1-\mathrm{\α}}\right)---\left(4\right)$

Mutual mode also has many restrictions.Proceed from the reality and consider, it can be assumed that be individual I can not observe directly the inherent tendency p of neighbours individuality j_j, i can only observe that j's is external Viewpoint selects S_j.In extreme circumstances, some individuality may be after once or twice be observed The viewpoint i.e. changing oneself selects, observing, some individuality may repeatedly oppose that viewpoint selects The viewpoint the most just changing oneself selects.

Although this algorithm has many good qualities, but viewpoint interaction is the most directly perceived.In reality Network in, the selection that the individual neighbours that the most not only can only observe oneself are individual, a lot Individual it is observed that the selection of other individualities multiple.Still original algorithm is made improvement As follows:

1), original formula, direct for individuality interactive relation derived by we, and it is right to omit Number probability O_iWith intermediate parameters a, the impact of b, the observation between individuality simplifies with interaction For following formula:

$p_i(n+1)=\left\{\begin{array}{ccc}\frac{p_i\left(n\right)\×\mathrm{\α}}{(1-p_i\left(n\right))\×(1-\mathrm{\β})+p_i\left(n\right)\×\mathrm{\α}},& \mathrm{if}& S_j\left(n\right)=+1\\ \frac{p_i\left(n\right)\×(1-\mathrm{\α})}{(1-p_i\left(n\right))\×\mathrm{\β}+p_i\left(n\right)\×(1-\mathrm{\α})},& \mathrm{if}& S_j\left(n\right)=-1\end{array}\right.---\left(5\right)$

2), according to diffusion of innovation theory, different according to initial tendentiousness, individuality is divided It is five grades, innovator, Early Adopter, masses, later stage adopter, delayed Person.

3), individual range of observation is expanded, in order to check the change expanding individual scope The impact caused under the model, first expands as two neighbours by individuality range of observation Body.In order to simplify emulation complexity, it is assumed that α=β, wherein, viewpoint is become by α value Fig. 2 is shown in the impact changed, it is known that α value affects the saltus step step-length of individual viewpoint, α equally Value is the biggest, the viewpoint of the easiest individual violent variation oneself.

5, detecting step, checks the effectiveness of described model.

Simulation and prediction process is divided into following two steps:

1, simulation and prediction step.

After said process, Matlab is used to launch the emulation of this algorithm.In this emulation, N initial value is 0, and inherent tendentiousness p of each individuality is substituted into data by system, the most individual Select to be calculated by inherent tendentiousness

2, result statistic procedure

From the point of view of physical significance, the value of α and β is between 1 to 0.5, two parameters The impact that viewpoint is developed by physical significance with it, has made discussion above.In this step In, the viewpoint tendency under the influence of different parameters value is made statistics and is discussed by us.

What Fig. 3 represented is in the case of A viewpoint reaches unification, under different parameters value, Individuality adopts the ratio change along with step-length of A viewpoint.Number of samples is 900, A viewpoint Final triumph.In figure, when false α is 0.6, observation the 25th, 50,75, to 300 In step, if the number of individuals holding A viewpoint numerical value is respectively 6,24,56,101,160, 236,326,430,552,690,809 and 899；When false α is 0.9, in observation The 25th, 50,75, in 300 steps, if the number of individuals holding A viewpoint numerical value is respectively 46,196,466,806,899,900,900,900,900,900,900 and 900.

What Fig. 4 represented is, and Early Adopter is initial distribution, and Early Adopter is for concentrating In the case of being distributed two kinds, each obtain the contrast of 100 A viewpoint triumph meansigma methodss.From figure In it can be seen that in the case of integrated distribution, viewpoint convergence rate is the most true with researcher before Fixed " the Early Adopter advantage of random distribution " is compared, and effect is close, in α value relatively Time big, (being expressed as radical, wavy network point state) even more has superiority.

The viewpoint distribution of initial individuals can be concluded that from the above, equally to viewpoint Development trend causes the most important impact, generally expands two neighbours at individual range of observation While occupying individuality, the initially viewpoint collective in integrated distribution situation, to follow-up viewpoint trend There is relatively important advantage.Its effect is also no less than " dividing at random involved by early-stage Study The advantage that viewpoint is distributed by the Early Adopter of cloth ".Assume that, when individual range of observation When expanding further, the advantage of the viewpoint colony of integrated distribution may proceed to expand.

First the present invention carries out pretreatment to the network information, for the individual and tool of network topology Body information carries out pattern quantization；According to the quantitative information obtained, initialize the knot of Web Community Structure and individual state, set up corresponding forecast model；Network public opinion is analyzed based on this model Development trend.Giving the environment of network user's state and network topology structure, this Bright method can the tendency of certain viewpoint in more efficiently prediction network, and then to network carriage Phase between the prediction of opinion, the coping strategy of accident and the network user and viewpoint tendency Affecting laws makes summary mutually.

Principle and the embodiment of the present invention are explained by specific embodiment used herein Stating, the explanation of above example is only intended to help to understand that the method for the present invention and core thereof are thought Think；Simultaneously for one of ordinary skill in the art, according to the thought of the present invention, at tool All will change on body embodiment and range of application.In sum, in this specification Hold and should not be construed as limitation of the present invention.

Claims (3)

1. network viewpoint propagation and Forecasting Methodology, it is characterised in that comprise the steps of

1) initialization step: network topological information is digitized modeling, to individual each Individual viewpoint state carries out digital quantization, obtains predicting required network structure and original state；

2) set up forecast model step: according to the init state obtained, determine corresponding pre- Survey rule of inference, it is thus achieved that initialize network and individual data, set up and predict mould accordingly Type, described forecast model is multinode Undirected networks model, if join together or different joining is depended on Network topology structure；The establishment step of described multinode Undirected networks model includes:

2.1) multinode Undirected networks establishment step: according to initialization data, establishes network and opens up Flutter, set individual nodes number and connection state；

2.2) data substitution step: the network information in setting model, the data of individual viewpoint；

2.3) model set-up procedure: trained by data, adjusts value and the net of individual viewpoint The parameter value of network scale；

2.4) evolution rule establishment step: the rules of interaction between definition individuality and individuality, from And define the evolution rule of overall network public opinion；Assume the individual mistake delivering viewpoint in a network Cheng Wei: observe the selection that neighbours are individual, change the inherent tendentiousness of oneself, and then make sight Point selection；By Modified Discrete behavior and continuous viewpoint algorithm, by the observation between individuality- The direct communication process that development is reduced between individuality；Specifically:

Individuality in network is numbered with i, and other parameters are numbered as follows:

p_i: represent the inherent tendentiousness of individual i；

S_i: the external viewpoint representing individual i selects；

O_i: in original discrete behavior continuous viewpoint model, the viewpoint logarithm representing individuality is general Rate, is expressed as:

$O_i\left(n\right)=log\frac{p_i\left(n\right)}{1-p_i\left(n\right)}---\left(1\right)$

Then individual viewpoint selects S_iI.e. can be expressed as:

S_i(n)=sign (O_i) (2)

2.4.1) first, the rules of interaction between the individuality defined in original algorithm is:

I) assume that n is that mutual step number occurs, define that all individualities are the most updated has been once Become a step, initial n=0；

ii)p_iIt is not the fixed value of individual i, when individual i observes certain neighbours individuality j Viewpoint in the step number n moment selects S_jN ()=1, i.e. selects viewpoint A, then individual i will change Become the inherent tendency p of oneself_i, this variation causes the viewpoint change of individual i, the most external choosing The change selected；

Iii) if viewpoint A is optimum selection, then α=P (OA | A) is defined permissible for individual i Observe the probability being chosen as A of its neighbours individuality j；Accordingly, observe as individual i S is selected to certain neighbours individuality j viewpoint in the step number n moment_jN ()=-1, i.e. viewpoint B are Optimum selection, then define β=P (OB | B) for individual i it is observed that its neighbours individuality j The probability being chosen as B；

The log probability rule change of definition individuality is as follows:

$O_i(n+1)=\left\{\begin{array}{ccc}{O}_i\left(n\right)+a,& if& S_j\left(n\right)=+1\\ O_i\left(n\right)-b,& if& S_j\left(n\right)=-1\end{array}\right.---\left(3\right)$

Wherein, intermediate parameters a, b are defined as:

$a=log\left(\frac{\mathrm{\α}}{1-\mathrm{\β}}\right)$

$b=log\left(\frac{\mathrm{\β}}{1-\mathrm{\α}}\right)---\left(4\right)$

Directly interactive relation individual in above-mentioned existing algorithm is derived, omits logarithm general Rate O_iWith intermediate parameters a, the impact of b, the observation between individuality and interaction are reduced to down Formula:

$p_i(n+1)=\left\{\begin{array}{ccc}\frac{p_i\left(n\right)\×\mathrm{\α}}{(1-p_i(n\left)\right)\×(1-\mathrm{\β})+p_i\left(n\right)\×\mathrm{\α}},& if& S_j\left(n\right)=+1\\ \frac{p_i\left(n\right)\×(1-\mathrm{\α})}{(1-p_i(n\left)\right)\×\mathrm{\β}+p_i\left(n\right)\×(1-\mathrm{\α})},& if& S_j\left(n\right)=-1\end{array}\right.---\left(5\right)$

2.4.2) according to diffusion of innovation theory, different according to initial tendentiousness, individuality is divided into Innovator, Early Adopter, masses, later stage adopter and five grades of stagnant the latter；

2.4.3) individual range of observation is expanded, in order to check the change expanding individual scope to exist The impact caused under this model, first expands as two neighbours' individualities by individuality range of observation；For Simplify emulation complexity, it is assumed that α=β；Wherein, the impact that viewpoint is changed by α value, α value is the biggest, the viewpoint of the easiest individual violent variation oneself；

3) simulation and prediction step: according to the multinode Undirected networks model set up, it was predicted that net The viewpoint evolution tendency of network user.

Network viewpoint propagation the most according to claim 1 and Forecasting Methodology, its feature exists In, described step 1) farther include:

Initialize network topology structure step: find algorithm according to network topology structure, obtain The physical arrangement of network；

Initialize individual information step: excavate viewpoint state individual in network and to described net Individuality in network carries out digital quantization.

Network viewpoint propagation the most according to claim 1 and Forecasting Methodology, its feature exists In, described step 3) including:

3.1) by set up forecast model step determines data, network topology, individual parameter Bring model into；

3.2) according to the evolution rule set up, Computer Simulation evolution is carried out；

3.3) collect prediction conclusion, for follow-up study, analyze to provide with decision-making and support.