CN102779126A - Internet virtual space user credibility evaluation method based on game theory - Google Patents

Abstract

The invention relates to an internet virtual space user credibility evaluation method based on a game theory and belongs to the field of network security. The evaluation method mainly includes establishing a random game model, constructing a virtual social network, and introducing influencing factors of an interactive relationship between users to obtain a user utility function; then calculating a gain function of the two interactive users, and performing derivation to determine a Nash equilibrium; and finally using a cross multiplication method to obtain an initial credibility of a node, wherein the influencing factors include connection relation coefficients and semantic relation coefficients. According to the internet virtual space user credibility evaluation method based on the game theory, data information is fully acquired, and by means of the high accuracies and the accurate logical judgment capabilities of computer technologies, the time-consuming and labor-consuming brainwork of persons can be replaced so that software can automatically grade the credibility of the users according to mutual information and semantic information on the basis of information contents.

Description

Internet virtual space user's reliability evaluation method based on theory of games

Technical field

The present invention relates to a kind of internet virtual space user's reliability evaluation method, belong to network safety filed based on theory of games.

Background technology

BBS is a public place; The user is not familiar with each other; Therefore can freely talk animatedly; Because BBS is easy to and is used all over the world easily, and can utilize the communication background of magnanimity for the offender shielding to be provided, and therefore becomes the most attractive exchange way of crime or terroristic organization.Utilize information extraction technique can identify criminal organization, and the relation between individuality, tissue and the incident.Therefore we can use methods of social network, from the BBS virtual society, extract potential criminal network.

Some new technology have been applied to be analyzed in the criminal network.For example: COPLINK ^[1-3,10-13]Technology has been applied to the corporations that identify in the criminal network.Also have, some disperse or continuous Markov model is used to following crime dramas or the process of prediction in statistical study ^[4,5]., social network analysis ^[6,7]In methods such as cluster, centrad metering method, multidimensional scaling and block models be used to study criminal organization's network based on the crime dramas data construct ^[8,9]Some visual, automatic methods have presented the evolutionary process of criminal network ^[8]Dynamic society's network analysis method is by Daning Hu, and Siddharth Kaza and Hsinchun Chen in are used for investigating some suspicious information transmission persons in the narcotics trade network of real society ^[11]In this piece paper, we combine community network and spatial analysis technology, the transmission range between the research corporations ^[12]..In recent years, represented to the research of transportation of multiple drugs and other running aspect, market potential that potential network had certain dirigibility, to the investigation of law enforcement agency with intervene and also have certain elasticity ^[14]

Because BBS is a huge virtual society; Each user is a potential criminal; Interactive relation between the user is through posting and reply foundation, so the criminal investigation personnel hope to obtain through these clues each member's potential crime probability, and the relation between the user.As if this sound it being an impossible task, because the investigator must spend great amount of time and energy search database, reads the crime report, seeks the clue of crime corporations, artificial then crime probability of inferring other members in the potential criminal network.These tasks all are not only time-consuming but also effort.

Yet, also do not have a technology to extract interactive relation at present, and automatically make up potential criminal network from BBS.Correlative study before ^[13]By the partial information of having examined and combine the conviction propagation algorithm to infer other members' crime probability.But these potential crime probability nets all are manual foundation.Simultaneously, the local evidence of node i, and the compatible function between node i and the node j all produces at random.Though experimental result has confirmed our hypothesis, it still is difficult promoting this method.

Summary of the invention

Technical matters to be solved by this invention is the deficiency to the above-mentioned background technology; A kind of internet virtual space user's reliability evaluation method based on theory of games is provided, has made software give each user's confidence level marking automatically according to interconnected user on the network's interactive information and semantic information.

The present invention adopts following technical scheme for realizing the foregoing invention purpose:

Internet virtual space user's reliability evaluation method based on theory of games comprises the steps:

Step 1: set up a betting model at random;

Step 2: make up the virtual society network;

Step 3: two influence factors introducing the interactive relation between user and the user: annexation coefficient and semantic relation coefficient;

Step 4: obtain user's utility function;

Step 5: calculate mutual both sides' gain function:

Step 6: Nash Equilibrium is confirmed in differentiate to both sides' gain function;

Step 7: estimate the user credit grade, obtain the initial credit degree of this node with cross multiplication.

Said internet virtual space user's reliability evaluation method based on theory of games, the foundation of the betting model at random in the step 1 is following:

Definition tlv triple G=[i, { A _i, R _i], wherein player or user collect i={1 ..., K}, the available set of strategies A of user i _i={ C representes the offender for C, L}, and L representes legal person, the benefit function of user i

$R_i=\{U_i^{\mathrm{cc}},U_i^{\mathrm{cl}},U_i^{\mathrm{lc}},U_i^{\mathrm{ll}}\},$

Wherein, K is arbitrarily greater than 1 integer; The benefit function of user i when

expression all is the offender as user i with the mutual user of user i; The benefit function of user i when

expression is offender and legal person respectively as user i with the mutual user of user i; The benefit function of user i when

expression is legal person and offender respectively as user i with the mutual user of user i, the benefit function of user i when

expression all is legal person as user i with the mutual user of user i.

Said internet virtual space user's reliability evaluation method based on theory of games, the virtual society network in the step 2 comprises: node and limit,

Said node is represented the user;

Said limit representes that two points that it connects have interactive relation, and promptly the user of two some representatives has interactive relation.

Said internet virtual space user's reliability evaluation method based on theory of games, the definition of annexation coefficient and semantic relation coefficient is following in the step 3:

M＝(t+1)·(e+1) (1)

N＝(s+1)·λ (2)

Wherein, M representes the annexation coefficient between the user, and N representes the semantic relation coefficient between the user; T representes the common answer degree of association between the user; E representes the mutual answer degree of association between the user, and s representes semantic association degree between the user, and λ representes to be used for the balance factor of balanced M and N.

Said internet virtual space user's reliability evaluation method based on theory of games, the definition of utility function is following in the step 4:

h _i＝log ₁₀(p _i·γ+r _i·τ)?(3)

Wherein, σ representes the mutual beneficial coefficient between the offender, and δ representes the mutual beneficial coefficient between the validated user; P representes offender's semantic coefficient, and q representes the semantic coefficient of validated user, and α representes the relative advantage coefficient between the offender; β representes the relative advantage coefficient between the validated user, h _iThe liveness of expression user i, p _iThe number of posting of expression user i, r _iThe answer number of expression user i, γ representes p _iWeights, τ representes r _iWeights.

Said internet virtual space user's reliability evaluation method based on theory of games, the definition of mutual both sides' gain function is following in the step 5:

Suppose variable x, y be respectively user i, with the mutual user's of user i potential crime probability, the gain function of user i is following:

$R_i(x,y)=U_i^{\mathrm{cc}}\·x\·y+U_i^{\mathrm{cl}}\·x\·(1-y)+U_i^{\mathrm{lc}}\·(1-x)\·y+U_i^{\mathrm{ll}}\·(1-x)\·(1-y)---\left(4\right)$

Said internet virtual space user's reliability evaluation method based on theory of games, the confirming as follows of Nash Equilibrium in the step 6: the gain function of user i is to representing the variable differentiate of the potential crime probability of user i,

k∈N(i) (5)

Wherein, all of its neighbor node of N (i) expression node i,

Expression user i is e on the limit _IkOn distribution probability.

Said internet virtual space user's confidence level evaluation system based on theory of games, node goes out confirming as follows of initial trusted degree in the step 7:

${\mathrm{\φ}}_i({\mathrm{\ϵ}}_i^c,{\mathrm{\ϵ}}_i^l)=(\frac{\underset{k\∈N\left(i\right)}{\mathrm{\Π}}{\mathrm{\η}}_{x_{\mathrm{ik}}}}{\underset{k\∈N\left(i\right)}{\mathrm{\Π}}{\mathrm{\η}}_{x_{\mathrm{ik}}}+\underset{k\∈N\left(i\right)}{\mathrm{\Π}}(1-{\mathrm{\η}}_{x_{\mathrm{ik}}})},1-\frac{\underset{k\∈N\left(i\right)}{\mathrm{\Π}}{\mathrm{\η}}_{x_{\mathrm{ik}}}}{\underset{k\∈N\left(i\right)}{\mathrm{\Π}}{\mathrm{\η}}_{x_{\mathrm{ik}}}+\underset{k\∈N\left(i\right)}{\mathrm{\Π}}(1-{\mathrm{\η}}_{x_{\mathrm{ik}}})})---\left(6\right)$

Wherein, The all of its neighbor node of N (i) expression node i;

expression user i is offender's a probability, and expression user i is the probability of validated user.

The present invention adopts technique scheme, has following beneficial effect: abundant image data information; The pinpoint accuracy of computer technology and accurate logic determines ability can replace some brainwork of wasting time and energy of people; Software can be given a mark according to interactive information and automatic each the user's confidence level of must giving of semantic information based on the information content.

Description of drawings

Fig. 1 is the synoptic diagram of game at random between two users.

Fig. 2 is must the meet accident synoptic diagram of probability of node multiplication cross.

Fig. 3 (a) is the constructed original potential network OPN of the present invention.

Fig. 3 (b) is the constructed common answer potential network TPN of the present invention.

Fig. 3 (c) is the constructed common answer potential network EPN of the present invention.

Fig. 3 (d) is the constructed semantic potential network SPN of the present invention.

Fig. 3 (e) is the constructed original potential network k-OPN of k degree of the present invention.

Fig. 4 (a) is a Nash Equilibrium for every limit.

Fig. 4 (b) is the compatible function of node 1.

Fig. 5 is the curve map of the initial probability of node among the k-OPN.

Fig. 6 (a) is KPN in 2003, and threshold value k is 20.

Fig. 6 (b) is KPN in 2003, and threshold value k is 10.

Fig. 7 (a) is the local evidence of 2003 annual data samples one.

Fig. 7 (b) is the local evidence of 2003 annual data samples two.

Label declaration among the figure:

Among Fig. 1; Strategies representes set of strategies; Criminal representes the offender; Legal representes legal person; The benefit function of

expression user A when user A and user B are the offender; The benefit function of

expression user A when user A and user B are offender and legal person respectively; The benefit function of

expression user A when user A and user B are legal person and offender respectively; The benefit function of

expression user A when user A and user B are legal person; The benefit function of

expression user B when user A and user B are the offender;

benefit function of user B when user A and user B are offender and legal person respectively; The benefit function of

expression user B when user A and user B are legal person and offender respectively,

are represented the benefit function of user B when user A and user B are legal person.

Node is represented the user on the BBS among Fig. 2, and the relation of posting-replying between two nodes is represented on the limit, the frequency of interaction between two nodes of numeric representation among Fig. 2 (a) and (b), (c) on every limit; Among Fig. 2 (d), the numeric representation semantic information value on every limit; Among Fig. 2 (e), every limit marked is that numerical value is the value of annexation coefficient M and semantic relation coefficient N, the initial motivation of this node of data representation on each node limit.

Embodiment

Be elaborated below in conjunction with the technical scheme of accompanying drawing to invention:

The international observation plate of forum is gathered the data in year June in June, 2003 to 2010 from the ends of the earth, sets up database with this, comprising 324666 users, 99735 themes, 4712859 answers.User in the database posts or replys the share member at the international observation plate in the past 8 years.We utilize the web crawlers program to download the source code of webpage, and the information that collects comprises user's name, theme id, and subject content, reply content is posted date and answer date.We utilize regular expression from the webpage source code, to extract key message.

Realization based on game theoretic internet virtual space user's reliability evaluation method is following:

Step 1: introduce as betting model as shown in Figure 1, two players are exactly last two users with interactive relation of BBS.They be not familiar with each other, all surfing on the net.They are defined as the offender to be defined as legal person exactly.A just _i={ C, L}.

expression when user A and user B are the offender, the benefit function of user A.Other can be by that analogy.Benefit function to user k can represent that

each user knows; If they are selected to validated user, everyone can obtain higher gain (benefit of value 100) so.If two people are selected to the disabled user, two people capitals obtain gain (value is 80 benefit) so.If they wherein a choose become the disabled user; And select another to be selected to validated user; The disabled user can obtain a gain (value is 20) and another validated user can obtain higher gain (value is 50) so. because therefore two users are difficult to and can cooperate respectively from the different local of the world and be not familiar with each other.In this example, from the selfish strategy of maximization payoff function, each user can be selected to validated user.Being easy to be confirmed, (L L) be Nash Equilibrium point unique in this game, and this equilibrium is not a Pareto optimality to point, and (C C) will produce bigger benefit for two players, yet this needs the cooperation between two criminals because select.Therefore, the rational conflict that exists between the social welfare very much clearly of individual early in this example.

As shown in Figure 2: because the node in potential network has a lot of abutment points, its local evidence can be influenced by a plurality of adjacent node, so the distribution probability of this node on certain bar limit can not be as local evidence.Here we obtain the initial probability of this node with multiplication cross, may the extend one's service potential crime probability of E of multiplication cross, and this allows.Because the purpose of our research is to help the investigator from mass data, to find suspicious person as much as possible.Although there is error, this method is seen still effectively on the whole.

Step 2: extract partial information from database and make up the virtual society network, comprise original potential network (OPN) as shown in Figure 1, jointly reply potential network (TPN), reply potential network (EPN), semantic potential network (SPN) jointly,

Because the relation between the node is easy to set up, therefore the limit of low weights does not have reference value in OPN, has therefore introduced threshold value, has removed the limit of weights less than k, if has all removed on all limits of certain node, this node also will be removed so.According to this method, made up that the k degree is original and post-reply network (k-OPN), shown in Fig. 1 (e), wherein k=1. connection type and node behavioural information can be from TPN, extract among EPN and the SPN;

Step 3: confirm annexation coefficient M and semantic relation coefficient N: with limit edge _1,6Be example, limit edge _1,6In, t _1,6=0, e _1,6=3, combine formula (1) to get M so _1,6=(t _1,6+ 1) (e _1,6+ 1)=4.Because

So N=s _1,61.3 ≈ 9, other limits are similar,

Wherein,

What calculate is balance factor, and avgt represents the average common recovery factor of whole network; Avg _eRepresent the mutual recovery factor of whole network; Avg _{Abs (s)}Represent the mean of the semantic coefficient absolute value of whole network.

Step 4: confirm utility function: the number of posting of node 1 is 20, and replying number is 245, obtains h in conjunction with formula (3) ₁=log ₁₀(p ₁γ+r ₁κ)=2.6484.Definition γ is 10, and κ is 1, and according to identical method, we can obtain h ₆=2.1903, other limits are similar.

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{cc}}=\mathrm{\&sigma;}\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">h</figure-callout>}}_1\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}+p\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="N"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="N"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">N</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=1\&times;2.6484\&times;4+0.5\&times;9=15.0936\\ {\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{cl}}=\mathrm{\&alpha;}\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">h</figure-callout>}}_1\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.2\&times;2.6484\&times;4=2.1187\\ {\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{lc}}=\mathrm{\&beta;}\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">h</figure-callout>}}_1\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.3\&times;2.6484\&times;4=3.1781\\ {\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{ll}}=\mathrm{\&delta;}\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">h</figure-callout>}}_1\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}+q\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="N"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="N"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">N</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.7\&times;2.6484\&times;4+1\&times;9=16.4155\end{array}\right.$

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{cc}}=\mathrm{\&sigma;}\&CenterDot;{\mathrm{<figure-callout\; id="6"\; label="h"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">h</figure-callout>}}_6\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}+p\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="N"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="N"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">N</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=1\&times;2.1903\&times;4+0.5\&times;9=13.2612\\ {\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{cl}}=\mathrm{\&alpha;}\&CenterDot;{\mathrm{<figure-callout\; id="6"\; label="h"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">h</figure-callout>}}_6\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.2\&times;2.1903\&times;4=1.7522\\ {\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{lc}}=\mathrm{\&beta;}\&CenterDot;{\mathrm{<figure-callout\; id="6"\; label="h"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">h</figure-callout>}}_6\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.3\&times;2.1903\&times;4=2.6284\\ {\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{ll}}=\mathrm{\&delta;}\&CenterDot;{\mathrm{<figure-callout\; id="6"\; label="h"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">h</figure-callout>}}_6\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="M"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">M</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}+q\&CenterDot;{\mathrm{<figure-callout\; id="1"\; label="N"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">\; <figure-callout\; id="6"\; label="N"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">N</figure-callout>\; </figure-callout>}}_{\mathrm{1,6}}=0.7\&times;2.1903\&times;4+1\&times;9=15.1328\end{array}\right.$

Among the k-OPN node to enliven coefficient as shown in table 1, wherein, post-num representes the number of posting, reply-num representes to reply number, liveness representes to enliven coefficient.

Node	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
																Post?num	20	2	2	4	5	8	?...	7	4	8	10	10	30	3	15
Reply-num	245	35	65	45	33	75	?...	55	25	85	145	100	95	65	145
																liveness	2.65	1.74	193	193	1.92	2.19	?...	2.10	181	2.22	2.39	2.30	2.60	198	2.47

Node enlivens coefficient among the table 1k-OPN

Step 5: calculate mutual both sides' gain function, suppose variable x, y be respectively user 1, with the mutual user's of user 6 potential crime probability, combine formula (4) to obtain user 1 and user's 6 gain function:

$R_1(x,y)={\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{cc}}\&CenterDot;x\&CenterDot;y+{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{cl}}\&CenterDot;x\&CenterDot;(1-y)+{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{lc}}\&CenterDot;(1-x)\&CenterDot;y+{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{ll}}\&CenterDot;(1-x)\&CenterDot;(1-y)$

$R_6(x,y)={\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{cc}}\&CenterDot;x\&CenterDot;y+{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{cl}}\&CenterDot;y\&CenterDot;(1-x)+{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{lc}}\&CenterDot;(1-y)\&CenterDot;x+{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{ll}}\&CenterDot;(1-x)\&CenterDot;(1-y)$

Step 6: Nash Equilibrium is confirmed in differentiate to gain function, obtains in conjunction with formula (5):

${\mathrm{\&eta;}}_{x_{\mathrm{1,6}}}=\frac{{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{cc}}-{\mathrm{<figure-callout\; id="6"\; label="U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">U</figure-callout>}}_6^{\mathrm{<figure-callout\; id="6"\; label="lc\; U"\; filenames="HDA0000060388210000031.png"\; state="\{\{state\}\}">lc</figure-callout>}}}{U_{}^{}}=0.443,$ ${\mathrm{\&eta;}}_{x_{\mathrm{6,1}}}=\frac{{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{cc}}-{\mathrm{<figure-callout\; id="1"\; label="U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">U</figure-callout>}}_1^{\mathrm{<figure-callout\; id="1"\; label="lc\; U"\; filenames="HDA0000060388210000013.png,HDA0000060388210000041.png"\; state="\{\{state\}\}">lc</figure-callout>}}}{U_{}^{}}=0.436$

In like manner can get the Nash Equilibrium on other limit.The Nash Equilibrium on every limit is shown in Fig. 3 (a), and the compatible function to node 1 of same method is shown in Fig. 3 (b).

Step 7: the initial probability of calculating each node.

Bring each value of step 4 gained into formula (6)

${\mathrm{\&phi;}}_i({\mathrm{\&epsiv;}}_i^c,{\mathrm{\&epsiv;}}_i^l)=(\frac{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}}}{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}}+\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}(1-{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}})},1-\frac{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}}}{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}}+\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}(1-{\mathrm{\&eta;}}_{x_{\mathrm{Ik}}})}),$ Obtain the initial probability of node among the k-OPN.

At last, do an emulation relatively.From database, extract data in 2003, extract two samples as research.105 users are arranged in the sample one, 156 limits, threshold value k is 20.Comprised 321 users in the sample two, 576 limits, based on the purpose of sensibility analysis, we are made as 10 to threshold value k.We have obtained two K-OPN then.Like Fig. 5 (a), under the less relatively situation of threshold value k, building network is a dense graph.According to the game theory principle, we can obtain each node local evidence separately, and the local evidence of sample one and sample two is as shown in Figure 6.

For the relatively output of these two samples, we utilize the Chi-square Test of conspicuousness α=0.05 that the result is carried out the degree of fitting test, confirm with whether there being significant difference between the sample.When the crime probability of node greater than 0.75 the time, we think that this user is the offender.According to this criterion, we have found 6 offenders in sample one, in sample two, have found 41 offenders.The experimental result of two samples is presented in the table 3.According to the result, we can obtain χ ²=4.02, v=(2-1) (2-1).The inquiry chi-square distribution table, we obtain working as H ₀=0.05 o'clock, 0.025＜P＜0.05.Therefore we think that when the scale of potential probability net was different, their crime probability also existed significant difference.The Chi-square Test result is as shown in table 2:

Table 2 Chi-square Test result

For raw data is assessed, we have invited the relevant expert in police office, Nanjing to our laboratory, and he had worked last 20 year in the law enforcement post.He has assessed the crime probability and the consistent matrix of each node in sample one and the sample two by hand.Suppose that our uses algorithm has found N different individuality, their crime probability all greater than 0.75. we also please this expert according to his experience, select the individuality of equal number.In order to detect the validity of our algorithm, we define an accurate rate formula:

$\mathrm{precision}=\frac{{\mathrm{\&Phi;}}_{\mathrm{expert}}\&cap;{\mathrm{\&Phi;}}_{\mathrm{algorithm}}}{{\mathrm{\&Phi;}}_{\mathrm{algorithm}}}\&times;100\%$

Φ wherein _{Exp ert}By the offender that the expert chooses, Φ _{Alg orithm}Be the offender who chooses according to our algorithm.The result shows Φ _{Exp ert}Almost with Φ _{Alg orithm}Number is the same, and the accuracy rate of our algorithm has reached 72.03% and 63.87% respectively in sample one and sample two.

Claims (8)

1. based on internet virtual space user's reliability evaluation method of theory of games, it is characterized in that: comprise the steps:

Step 1: set up a betting model at random;

Step 2: make up the virtual society network;

Step 3: two influence factors introducing the interactive relation between user and the user: annexation coefficient and semantic relation coefficient;

Step 4: obtain user's utility function;

Step 5: calculate mutual both sides' gain function:

Step 6: Nash Equilibrium is confirmed in differentiate to both sides' gain function;

Step 7: estimate the user credit grade, obtain the initial credit degree of this node with cross multiplication.

2. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1, it is characterized in that: the foundation of the described betting model at random of step 1 is following:

Definition tlv triple G=[i, { A _i, R _i], wherein player or user collect i={1 ..., K}, the available set of strategies A of user i _i={ C representes the offender for C, L}, and L representes legal person, the benefit function of user i

$R_i=\{U_i^{\mathrm{cc}},U_i^{\mathrm{cl}},U_i^{\mathrm{lc}},U_i^{\mathrm{ll}}\},$

Wherein, K is arbitrarily greater than 1 integer; The benefit function of user i when

expression all is the offender as user i with the mutual user of user i; The benefit function of user i when

expression is offender and legal person respectively as user i with the mutual user of user i; The benefit function of user i when expression is legal person and offender respectively as user i with the mutual user of user i, the benefit function of user i when

expression all is legal person as user i with the mutual user of user i.

3. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1 is characterized in that: the described virtual society network of step 2 comprises: node and limit,

Said node is represented the user;

Said limit representes that two points that it connects have interactive relation, and promptly the user of two some representatives has interactive relation.

4. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1 is characterized in that: the definition of said annexation coefficient of step 3 and semantic relation coefficient is following:

M＝(t+1)·(e+1) (1)

N＝(s+1)·λ (2)

Wherein, M representes the annexation coefficient between the user, and N representes the semantic relation coefficient between the user; T representes the common answer degree of association between the user; E representes the mutual answer degree of association between the user, and s representes semantic association degree between the user, and λ representes to be used for the balance factor of balanced M and N.

5. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1, it is characterized in that: the definition of the said utility function of step 4 is following:

$\left\{\begin{array}{c}{U}_i^{\mathrm{cc}}=\mathrm{\&sigma;}\&CenterDot;h_i\&CenterDot;M+p.N\\ U_i^{\mathrm{cl}}=\mathrm{\&alpha;}\&CenterDot;h_i\&CenterDot;M\\ U_i^{\mathrm{lc}}=\mathrm{\&beta;}\&CenterDot;h_i\&CenterDot;M\\ U_i^{\mathrm{ll}}=\mathrm{\&delta;}\&CenterDot;h_i\&CenterDot;M+q\&CenterDot;N\end{array}\right.,$ h _i＝log ₁₀(p _i·γ+r _i·τ) (3)

Wherein, σ representes the mutual beneficial coefficient between the offender, and δ representes the mutual beneficial coefficient between the validated user; P representes offender's semantic coefficient, and q representes the semantic coefficient of validated user, and α representes the relative advantage coefficient between the offender; β representes the relative advantage coefficient between the validated user, h _iThe liveness of expression user i, p _iThe number of posting of expression user i, r _iThe answer number of expression user i, γ representes p _iWeights, τ representes r _iWeights.

6. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1 is characterized in that: the definition of the mutual both sides' gain function of step 5 is following:

Suppose variable x, y be respectively user i, with the mutual user's of user i potential crime probability, the gain function of user i is following:

$R_i(x,y)=U_i^{\mathrm{cc}}\&CenterDot;x\&CenterDot;y+U_i^{\mathrm{cl}}\&CenterDot;x\&CenterDot;(1-y)+U_i^{\mathrm{lc}}\&CenterDot;(1-x)\&CenterDot;y+U_i^{\mathrm{ll}}\&CenterDot;(1-x)\&CenterDot;(1-y)---\left(4\right)$

7. the internet virtual space user's reliability evaluation method based on theory of games as claimed in claim 1 is characterized in that: the confirming as follows of said Nash Equilibrium of step 6: the gain function of user i is to representing the variable differentiate of the potential crime probability of user i,

K ∈ N (i) (5) wherein, all of its neighbor node of N (i) expression node i,

Expression user i is e on the limit _IkOn distribution probability.

8. the internet virtual space user's confidence level evaluation system based on theory of games as claimed in claim 1 is characterized in that: the described node of step 7 goes out confirming as follows of initial trusted degree:

${\mathrm{\&phi;}}_i({\mathrm{\&epsiv;}}_i^c,{\mathrm{\&epsiv;}}_i^l)=(\frac{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{ik}}}}{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{ik}}}+\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}(1-{\mathrm{\&eta;}}_{x_{\mathrm{ik}}})},1-\frac{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{ik}}}}{\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}{\mathrm{\&eta;}}_{x_{\mathrm{ik}}}+\underset{k\&Element;N\left(i\right)}{\mathrm{\&Pi;}}(1-{\mathrm{\&eta;}}_{x_{\mathrm{ik}}})})---\left(6\right)$

Wherein, The all of its neighbor node of N (i) expression node i;

expression user i is offender's a probability, and expression user i is the probability of validated user.

Images