CN103902690A - Method for improving accuracy of influence of user generate content (UGC) information of social network - Google Patents

Abstract

The invention discloses a method for improving accuracy of influence of user generate content (UGC) information of the social network. The UGC comprises M keywords and involves N users. The method includes steps of establishing a social network UGC member involvement mechanism, building a fan network unauthorized oriental map according to user fans relation of the UGC and dividing communities; setting up an interest network authorized non-oriental map according to user replay relation of the UGC and dividing communities; calculating social influence of a user X according to relevant degrees of various influence factors of the member involvement mechanism; calculating social influence of a keyword K published by the user X according to the formula, the m is propagation times of the keywords on the user X, if the m=0, SKX=0; calculating comprehensive social influence of the keyword K in the UGC according to the formula; calculating the sum of the comprehensive social influences of the M keywords in the UGC and obtaining information influence INF of the UGC.

Description

A kind of method that improves social networks user and produce content information influence power accuracy

Technical field

The present invention relates to information monitoring technology, particularly a kind of method that improves social networks user and produce content information influence power accuracy.

Background technology

Internet has entered into the web2.0 epoch, each user can freely communicate one's views, a lot of important contents or news are all to produce content (UGC by user, User Generate Content) first generate, then by social networks wide-scale distribution, finally in certain specific social circle, even entire society produces tremendous influence power.Therefore, the research of UGC influence power all has very important effect for information acquisition, monitoring, prediction etc.But because the quantity of UGC is too huge, the speed of increase is very fast, be difficult to all UGC to process, must filter out quality UGC good and that influence power is high and study and utilize.Thus, the research of the quality to UGC and informational influence force estimation more and more comes into one's own.

At present mainly apply influence power diffusion model (IDM about the research of informational influence power, Influence Diffusion Model) and improved model (as influence power spreading probability model IDPM, Influence Diffusion Probability Model etc.) analyze.Influence power diffusion model IDM based on text session utilizes the reply chain structure in session, and the similarity of calculating between text based on word frequency is calculated the influence power diffusivity in source, and the influence power sum of every reply diffusion is the influence power diffusivity of the text.After this model proposes, become the important foundation stone of informational influence power research, descendant is that this model is improved for the research of informational influence power mostly; The false influence power propagation problem that influence power spreading probability model IDPM solves the influence power transferring structure tomography problem in IDM model and pours water and cause by define single key words probability of spreading influence power in whole space of interest, by considering that the effective key words in sentence solves the influence power transferring content tomography problem in IDM model.

But these models exist some very significantly defects, as identical in the weights of each comment or reply, do not consider relation between user etc.Take a model on BBS as UGC as example, as shown in Figure 1:

User 1 is information publisher, user 2～user 5 is user 1 reply person, A, B, C, D, E, F are the keyword that this model comprises, heavy line represents the influence power propagation relation of model in user, the direction of heavy line is the influence power direction of propagation of model, dot-and-dash line represents bean vermicelli relation between user, and dotted line represents that user belongs to the relation of same community in interest network, and fine line represents that user belongs to the relation of same community in bean vermicelli network.

In Fig. 1, user 2～user 5 has replied user 1 model, but user 2 is beans vermicelli of user 1, user 3 and user 1 belong to same interest network social association, user 4 belongs to identical bean vermicelli network social association (but not being user 1 bean vermicelli) with user 1, user 5 is new users, before may almost it doesn't matter with user 1.

This shows, in IDPM model, the keyword of UGC is not carried out to the informational influence power deviation that weighting processing respectively can cause calculating UGC.

Summary of the invention

In view of this, the present invention proposes a kind of method that improves social networks user and produce content information influence power accuracy, efficiently solve in prior art not the defect of the keyword of UGC being carried out informational influence power that weighting processing respectively causes calculating UGC and existed deviation.The technical scheme that the present invention proposes is:

Improve the method that social networks user produces content information influence power accuracy, the method comprises:

A. set up social networks UGC member participation mechanism, determine the path coefficient between the each influence factor of described member's participation mechanism, described path coefficient is the degree of correlation between the each influence factor of described member's participation mechanism;

B. build bean vermicelli network according to user's bean vermicelli relation of described UGC and have no right digraph, described bean vermicelli network is had no right to digraph and carry out community division; Build the interest network non-directed graph of having the right according to user's reply relation of described UGC, the described interest network non-directed graph of having the right is carried out to community division;

C. calculate the social influence power U of user X according to the degree of correlation between the each influence factor of described member's participation mechanism _x;

D. basis

calculate the social influence power that user X issues keyword K, m is the propagation times of keyword K on user X, if m=0, S _kX=0;

E. according to formula

calculate the comprehensive social influence power of keyword K in described UGC;

F. calculate the comprehensive social influence power sum of a described M keyword in described UGC, obtain the informational influence power INF of described UGC.

In such scheme, described member's participation mechanism comprises information quality, group identification sense, is worth perception and participates in four influence factors, and the path coefficient of information quality and group identification sense is a ₁, the path coefficient of information quality and value perception is a ₂, the path coefficient that is worth perception and group identification sense is a ₃, the path coefficient of participation and group identification sense is a ₄.

In such scheme, described step C further comprises:

According to formula calculate the social influence power of user X in described UGC,

Wherein, b is the number of times that user X is directly replied in described UGC, if user does not exist direct reply person, U _x=0;

If user X reply person direct with it belongs to same interest network social association, C ₁=a ₁, otherwise, C ₁=1;

If user X belongs to identical bean vermicelli network social association, C with its direct reply person ₂=a ₂× a ₃, otherwise, C ₂=1;

If user X is described UGC information publisher's bean vermicelli, f=a ₂, otherwise f=1.

In such scheme, described step F further comprises:

According to formula

the informational influence power INF of described UGC.

In such scheme, the path coefficient a of information quality and group identification sense ₁=0.333, the path coefficient a of information quality and value perception ₂=0.824, be worth the path coefficient a of perception and group identification sense ₃=0.624, the path coefficient a of participation and group identification sense ₄=0.437.

In sum, the present invention proposes a kind of method that improves social networks user and produce content information influence power accuracy, application sociability is expanded influence power spreading probability model (S-IDPM, Sociability-based Influence Diffusion Probability Model) calculating UGC informational influence power, mainly utilize user social contact network (comprising bean vermicelli network and interest network) and reply chain structure the reply of different user is weighted, thereby the keyword of UGC is weighted respectively to processing, improve social networks user and produced the accuracy that content information influence power is calculated.

Accompanying drawing explanation

Fig. 1 is by between user and a graph of a relation of being posted.

Fig. 2 is bean vermicelli network chart.

Fig. 3 is interest network chart.

Fig. 4 is user member's participation mechanism.

Fig. 5 is the UGC structural drawing of embodiment of the method one.

Fig. 6 is the process flow diagram of embodiment of the method one.

Fig. 7 is the artificial mark fine work note accumulation containing ratio comparison diagram of embodiment of the method two.

Fig. 8 is class-eigenwert contrast figure of embodiment of the method two.

Fig. 9 is the machine mark fine work note accumulation containing ratio comparison diagram of embodiment of the method two.

Embodiment

Clearer for what the object, technical solutions and advantages of the present invention were expressed, below in conjunction with drawings and the specific embodiments, the present invention is further described in more detail.

The technical scheme of one embodiment of the invention is:

A. set up social networks UGC member participation mechanism, determine the path coefficient between the each influence factor of described member's participation mechanism, described path coefficient is the degree of correlation between the each influence factor of described member's participation mechanism;

B. build bean vermicelli network according to user's bean vermicelli relation of described UGC and have no right digraph, described bean vermicelli network is had no right to digraph and carry out community division; Build the interest network non-directed graph of having the right according to user's reply relation of described UGC, the described interest network non-directed graph of having the right is carried out to community division;

C. calculate the social influence power U of user X according to the degree of correlation between the each influence factor of described member's participation mechanism _x;

D. basis

calculate the social influence power that user X issues keyword K, m is the propagation times of keyword K on user X, if m=0, S _kX=0;

E. according to formula

calculate the comprehensive social influence power of keyword K in described UGC;

F. calculate the comprehensive social influence power sum of a described M keyword in described UGC, obtain the informational influence power INF of described UGC.

Technical solution of the present invention is added user's factor to user and is produced in the calculating of content information influence power, and all users of the social networks such as BBS, microblogging, Renren Network are divided into information publisher and information-reply person.Build bean vermicelli network according to the user's bean vermicelli relation that participates in UGC and have no right digraph, as shown in Figure 2, user 1 is user 2 bean vermicelli, between user 1 and user 2, has a limit of pointing to user 2 from user 1; Build the interest network non-directed graph of having the right according to the user's reply relation that participates in UGC, as shown in Figure 3, user 1 and user 2 have participated in 7 information discussion that information publisher issues jointly, between user 1 and user 2, have the nonoriented edge that weights are 7.

According to prior art, above-mentioned bean vermicelli network and interest network are carried out to community division, and the community of dividing is numbered, the community numbering that belongs to same community user is identical, and the user of identical community numbering (identical bean vermicelli network social association numbering or same interest network social association numbering) has similar values to a certain extent.User's community partitioning technology is prior art, be not described in detail in this, table 1 for Fig. 2 community divide after bean vermicelli network social association, table 2 for community divide after interest network social association example, as shown in Table 1, user 1 and user 3 are same bean vermicelli network social association, and as shown in Table 2, user 1 and user 2 are same interest network social association.

Table 1

User	Bean vermicelli network social association numbering f
		1	1
2	2
		3	1
4	3
		5	4
6	5

Table 2

User	Interest network social association numbering r
		1	3
2	3
		3	1
4	2
		5	4
6	5

Technical solution of the present invention has built user member's participation mechanism, as shown in Figure 4.

Information quality is the informational influence power of UGC in social networks, represents the quality of correctness, promptness, novelty and the service quality of stability, the information of described UGC.

Being worth perception utilizes the bean vermicelli relation between user in social networks to represent.For an information publisher U1, if the bean vermicelli user U2 of U1 has replied this information publisher, think and drive factor that bean vermicelli user U2 replys U1 except to the approval of U1, also with certain factor that maintains interpersonal relation, in this case, be not only that the informational influence issued of U1 has driven the participation of bean vermicelli user U2, the factor of the interpersonal relation between user of wherein also having adulterated, in social networks UGC informational influence power is calculated, the content that bean vermicelli user is replied will reduce corresponding weight.

Group identification sense represents that in social networks, the community between user is divided and replys the impact of evaluating UGC informational influence power.It is the community division of aforesaid bean vermicelli network and interest network that community is divided, if reply person belongs to identical community (identical bean vermicelli network social association and/or same interest network social association) with the person of being responded, alleviates respective weights, otherwise increases respective weights; If the content quality of a UGC is very high or possess the potentiality that produce certain influence power, user is ready to participate in such UGC mostly, and therefore replying to evaluate becomes the key factor that UGC influence power is calculated.In reply evaluation, the propagation times of keyword in UGC is more, illustrates that the influence power of this UGC is larger.

What the path coefficient in Fig. 4 user member participation mechanism was weighed is two degrees of correlation between variable, uses respectively a ₁, a ₂, a ₃, a ₄represent, and 0 < a ₁< 1,0 < a ₂< 1,0 < a ₃< 1,0 < a ₄< 1.

Calculate the social influence power U of the participating user X of social networks UGC according to the degree of correlation between the each influence factor of described member's participation mechanism _x:

$U_X=\{1+\ln [+\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}({\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">C</figure-callout>}}_1\&times;C_2)\left]\right\}\&times;f---\left(1\right)$

Wherein, b is the number of times that user X is directly replied in a UGC.If user X is not directly replied, U _x=0.

C ₁whether the reply person who represents user X belongs to identical interest network social association with user X, if so, and because information quality and group identification sense have a ₁degree of correlation, C ₁=a ₁; Otherwise, C ₁=1.C ₁=1 shows that influence power is stronger if can attract to reply to the user of different interest network social associations.

C ₂whether the reply person who represents user X belongs to identical bean vermicelli network social association with user X, if so, due to information quality be worth perception and have a ₂degree of correlation, being worth perception and group identification sense has a ₃degree of correlation, so C ₂=a ₂× a ₃; Otherwise, C ₂=1.C ₂=1 shows that influence power is stronger if can attract to reply to the user of different bean vermicelli network social associations.

F represents whether user X is the information publisher's of this UGC bean vermicelli, if so, and because information quality and value perception have a ₂degree of correlation, f=a ₂; Otherwise f=1.This shows if user X is information publisher's bean vermicelli, and the reply of user X is not only to the affirming of content, also with the reason that maintains social networks, therefore will reduce weight.

The social influence power U of user X in a UGC in social networks _xafter determining, the social influence power of the keyword K that this user X issues can further be determined:

$S_{\mathrm{KX}}=\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{U}_X---\left(2\right)$

Wherein, m is the propagation times of keyword K on user X,, in the direct reply person of user X, also replys direct reply person's quantity of keyword K, if m=0, S _kX=0.

The comprehensive social influence power of keyword K in whole UGC is the social influence power sum of all users of this UGC (comprising information publisher and reply person) issue keyword K, that is:

${\mathrm{Sum}}_K=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Ki}}---\left(3\right)$

Wherein, N is the number of users (being information publisher and reply person's quantity sum) that participates in this UGC.

UGC informational influence power is the comprehensive social influence power sum of all keywords in this UGC, that is:

$\mathrm{INF}=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\log [1+\frac{{\mathrm{Sun}}_i}{\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{Sum}}_j}]---\left(4\right)$

Wherein, M is all keyword quantity that this UGC comprises.

With embodiment, technical solution of the present invention is described further below.

Embodiment of the method one

Fig. 5 is the structural drawing of a UGC of the present embodiment, and as shown in Figure 5, this UGC comprises 4 users: user 1, user 2, user 3, user 4.User 1 is information publisher, and the keyword of issue is A, B, C; User 2 and user 3 have directly replied respectively user 1, and the keyword that user 2 issues is A, C, D, and the keyword that user 3 issues is B, F; User 4 has directly replied user 2, and the keyword of issue is C, F.Interest network social association numbering represents with r, r ₁=1, r ₂=1, r ₃=2, r ₄=3; Bean vermicelli network social association numbering represents with f, f ₁=1, f ₂=2, f ₃=1, f ₄=3.User 2 and user 4 are beans vermicelli of information publisher user 1.The present embodiment is that in member's participation mechanism, the path coefficient assignment between each factor is: a ₁=0.333, a ₂=0.824, a ₃=0.624, a ₄=0.437.Fig. 6 is the process flow diagram of the present embodiment, as shown in Figure 6, comprises the following steps:

Step 601: calculate respectively the social influence power of each user in this UGC.

According to formula

calculate respectively each user's social influence power.Describe as example take the social influence power of calculating user 1, user 2～user 4 therewith computing method is identical, repeats no more.

Direct reply person's quantity of user 1 is 2, i.e. b=2; User 2 from user 1 at different interest network social associations, in identical bean vermicelli network social association, therefore C ₁=a ₁=0.333, C ₂=1; User 3 with user 1 at same interest network social association, in different bean vermicelli network social associations, therefore C ₁=1, C ₂=a ₂× a ₃=0.514; User 1 is not self bean vermicelli, f=1, therefore

U ₁＝{1+ln[1+a ₁×1+1×a ₂×a ₃]}×1＝1.614

In like manner U ₂=1.395, U ₃=0, U ₄=0.

Step 602: the comprehensive social influence power of calculating respectively each keyword in this UGC.

Be calculated as example with the comprehensive social influence power of keyword C and describe, the comprehensive social influence power computing method of keyword A, B, D, F are identical, repeat no more.

The user who issues keyword C has user 1, user 2, user 4, and for user 1, keyword C has only propagated once (direct reply person user 2 of the user 1 has issued keyword C), S _c1=U ₁; For user 2, keyword C has only propagated once (direct reply person user 4 of the user 2 has issued keyword C), S _c2=U ₂; For user 3 and user 4, keyword C does not propagate (user 3 and user 4 all not direct reply person issue keyword C), S _c3=0, S _c4=0, therefore, the social influence power of keyword C is: ${\mathrm{Sum}}_C=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Ci}}={\mathrm{<figure-callout\; id="1"\; label="S\; C"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">S</figure-callout>}}_C+{\mathrm{<figure-callout\; id="2"\; label="S\; C"\; filenames="HDA0000482980390000012.png,HDA0000482980390000021.png"\; state="\{\{state\}\}">S</figure-callout>}}_C={\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1+{\mathrm{<figure-callout\; id="2"\; label="U"\; filenames="HDA0000482980390000012.png,HDA0000482980390000021.png"\; state="\{\{state\}\}">U</figure-callout>}}_2=3.009.$

In like manner, ${\mathrm{Sum}}_A=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Ai}}={\mathrm{<figure-callout\; id="1"\; label="S\; A"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">S</figure-callout>}}_A={\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1=1.614;{\mathrm{Sum}}_B=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Bi}}={\mathrm{<figure-callout\; id="1"\; label="S\; B"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">S</figure-callout>}}_B={\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1=1.614;$

${\mathrm{Sum}}_D=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Di}}=0;{\mathrm{Sum}}_F=\underset{i=1}{\overset{4}{\mathrm{\&Sigma;}}}{S}_{\mathrm{Fi}}=0.$

Step 603: the informational influence power of calculating UGC.

According to formula

calculate the informational influence power of this UGC.

$\mathrm{INF}=2\log (1+\frac{{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1}{{3\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1+U_2})+\log (1+\frac{{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1+{\mathrm{<figure-callout\; id="2"\; label="U"\; filenames="HDA0000482980390000012.png,HDA0000482980390000021.png"\; state="\{\{state\}\}">U</figure-callout>}}_2}{{3\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000482980390000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_1+U_2})=0.371$

Embodiment of the method two

How the UGC that embodiment of the method one participates in take less user calculates social networks UGC informational influence power as example to technical solution of the present invention is illustrated, and the present embodiment is described further technical solution of the present invention as an example of the user of ends of the earth forum tittle-tattle column upper 2012 year and model information example.

User profile comprises the ID of 181841 user ID, its bean vermicelli altogether, the model ID issuing at this column and the ID that replys model at this column; Model information comprises the ID of 43609 pieces of models, sequence number, publisher ID and the content thereof in this model Zhong Mei building altogether.By judging whether model contains the keeper's of forum fine work symbol, from model information, filter out the fine work model collection of 827 pieces of models as artificial mark, other models are as non-fine work model collection.Because data volume is huge, therefore concentrate the fine work model of randomly drawing 9173 pieces of models and 827 pieces of artificial marks to be mixed into the model sample of 10000 pieces from non-fine work model, and utilize this sample respectively to S-IDPM, IDM, IDPM contrasts, assessment and analysis.In addition, also utilize the clustering method of statistical information to carry out machine mark to model, obtained the fine work model collection under machine mark, and equally to S-IDPM, IDM, IDPM has carried out contrast and analysis.

In table 3, provide the contrast and experiment of above-mentioned three kinds of methods, because model quantity is very large, only provided the model of rank front 5 here.

Table 3

From table 3, can observe, the key distinction of IDM and S-IDPM is on model 2894103 and 2366245.By observing corresponding language material, find that model 2894103 is advertisement collection notes, he has issued corresponding advertisement masterplate, all users will reply according to set form, so according to IDM model, mainly utilize co-occurrence word to calculate influence power, therefore, this model influence power under IDM is very high.But, from title, can find out, that this model attracts is the user that a group is liked automobile, from the user interest network calculating, also can find out that a lot of users once replied some model jointly, therefore, they have identical interest network numbering in identical interest network, and this illustrates that this model only propagates in an inner circle of people.Therefore, in S-IDPM, its rank is not very high, does not enter first 5.And model 2366245 has caused and pays close attention to widely and reply, have 1476757th floors replies, in model, user does not have significantly large-scale bean vermicelli and interest network, and user's circle relatively disperses, and illustrates that this model has caused the extensive concern of the various customer groups in community.Therefore, its model influence power rank in S-IDPM is higher.

And between IDPM and S-IDPM the key distinction on model 2510082 and model 2713599.By observing corresponding language material, find user's " I am Japanese car owner " by name of the people that posts of model 2713599, very similar with the title of model, also find that by the User Page of observing her this user is without any bean vermicelli and concern, do not reply any model and only sent out this piece of model yet, these have absolutely proved that this user name is a vest name, do not have the relation of social activity or interest network with anyone.This piece of model causes the reply of totally 14944th floors, although and model 2510082 has caused the reply of 28211st floors, more than the reply number of model 2713599, the reply user's of model 2713599 bean vermicelli and interest network disperse more, therefore, the influence power rank in S-IDPM is more forward.

From above qualitative analysis, can find out, S-IDPM can solve some problems that IDM and IDPM do not consider to a certain extent.

Next, the effect of three kinds of methods of quantitative test in model influence power is calculated.

First, the present embodiment has contrasted fine work note manually to mark (keeper of forum mark) as fine work note standard, relatively IDM, and IDPM and S-IDPM fine work note accumulation containing ratio comparison diagram, as shown in Figure 7.As can be seen from Figure 7, S-IDPM fine work note accumulation containing ratio reaches 70% left and right the soonest at first 3000.Illustrate in the influence power sequence of S-IDPM, covered 70% fine work model in front 30%, and front 10%, 20% all higher than IDM and IDPM model.Therefore, illustrate that S-IDPM is better to the result of calculation of model influence power, more meets artificial annotation results.

Next, the leader of opinion's discovery algorithm based on cluster in prior art is used in model influence power analysis of the present invention, realize the algorithm that utilizes statistical information clustering method to find fine work note.

Select the building of model to count F, duration T, replys number P, and reply hourly building number is expressed as F/T, and the word number in average every building is W/F, and the difference D that non-building-owner replys number and building-owner replys number is as eigenwert, the number of members of N representation class.Utilize subclass quantity choosing method and clustering algorithm (being prior art) to obtain 8 classes, as shown in Figure 8.

Screening conditions in leader of opinion's discovery algorithm based on cluster are adjusted into: class members's number is less, the member in the larger class of class members's eigenwert average is as the fine work note of machine mark.Therefore, the member in No. 5 and No. 7 classes is as the fine work note of experiment next, and No. 5 and No. 7 classes have 1001 members, and it is identical that 827 fine work notes that wherein mark with the keeper of forum only have 291 pieces of models, so, this from shown in Fig. 7, test different.Next utilize these 1001 pieces of fine work notes to carry out comparison IDM, IDPM and S-IDPM fine work note accumulation containing ratio comparison diagram, as shown in Figure 9:

As can be seen from Figure 9, S-IDPM fine work note cumulative percentage curve is also always in middle reaches level, and in the time of first 2000 pieces, reach fine work note accumulative total rate more than 85%, illustrate that S-IDPM is still fine to the calculating effect of model influence power in the situation that utilizing statistical information machine mark fine work note.

Finally, three kinds of algorithms fine work note accuracy rate in the situation that manually mark and machine mark is contrasted, as shown in table 4, S-IDPM marks in two kinds of situations at artificial mark and machine, and fine work note accuracy rate is all higher than other two kinds of models.

Table 4

?	Pt 0	Pt 1
			IDM	28.1%	68.1%
IDPM	30.2%	67.3%
			S-IDPM	32.4%	68.4%

By utilizing fine work note accumulative total rate contrast experiment and the fine work note accuracy rate contrast experiment of artificial mark and machine mark, comprehensive above experimental result can find out that S-IDPM is more accurate to the result of model influence power calculating, is better than IDM and IDPM method.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any modifications of making within the spirit and principles in the present invention, be equal to replacement, improvement etc., within all should being included in the scope of protection of the invention.

Claims (5)

1. improve the method that social networks user produces content information influence power accuracy, be applied to social networks user and produce content UGC, described UGC comprises M keyword, and total N user participates in described UGC, it is characterized in that, the method comprises:

A. set up social networks UGC member participation mechanism, determine the path coefficient between the each influence factor of described member's participation mechanism, described path coefficient is the degree of correlation between the each influence factor of described member's participation mechanism;

B. build bean vermicelli network according to user's bean vermicelli relation of described UGC and have no right digraph, described bean vermicelli network is had no right to digraph and carry out community division; Build the interest network non-directed graph of having the right according to user's reply relation of described UGC, the described interest network non-directed graph of having the right is carried out to community division;

C. calculate the social influence power U of user X according to the degree of correlation between the each influence factor of described member's participation mechanism _x;

D. basis

calculate the social influence power that user X issues keyword K, m is the propagation times of keyword K on user X, if m=0, S _kX=0;

E. according to formula calculate the comprehensive social influence power of keyword K in described UGC;

F. calculate the comprehensive social influence power sum of a described M keyword in described UGC, obtain the informational influence power INF of described UGC.

2. method according to claim 1, is characterized in that, described member's participation mechanism comprises information quality, group identification sense, is worth perception and participates in four influence factors, and the path coefficient of information quality and group identification sense is a ₁, the path coefficient of information quality and value perception is a ₂, the path coefficient that is worth perception and group identification sense is a ₃, the path coefficient of participation and group identification sense is a ₄.

3. method according to claim 1 and 2, is characterized in that, described step C further comprises:

According to formula

calculate the social influence power of user X in described UGC,

Wherein, b is the number of times that user X is directly replied in described UGC, if user does not exist direct reply person, U _x=0;

If user X reply person direct with it belongs to same interest network social association, C ₁=a ₁, otherwise, C ₁=1;

If user X belongs to identical bean vermicelli network social association, C with its direct reply person ₂=a ₂× a ₃, otherwise, C ₂=1;

If user X is described UGC information publisher's bean vermicelli, f=a ₂, otherwise f=1.

4. method according to claim 1, is characterized in that, described step F further comprises:

According to formula

the informational influence power INF of described UGC.

5. method according to claim 2, is characterized in that, the path coefficient a of information quality and group identification sense ₁=0.333, the path coefficient a of information quality and value perception ₂=0.824, be worth the path coefficient a of perception and group identification sense ₃=0.624, the path coefficient a of participation and group identification sense ₄=0.437.

Images