CN102629341A - Web service QoS (Quality of Service) on-line prediction method based on geographic position information of user - Google Patents

Abstract

The invention discloses a Web service QoS (Quality of Service) on-line prediction method based on geographic position information of a user. The Web service QoS on-line prediction method comprises the following steps of: (11) collecting QoS historical data and IP (Internet Protocol) information provided by the user; (12) generating a geometric geographical position coordinate of the user by the collected IP information, calculating relative distance of the geographic position of the user according to the geometric geographical position coordinate, and generating a user relative distance information matrix; (13) receiving a QoS query request of a target user and requesting the target user to customize a neighbor threshold value theta; and (14) judging the QoS query request received by the step (13), retransmitting the QoS information fed back previously to the target user if the target user has called the QoS query request, and performing QoS prediction if the QoS query request has not been called. According to the Web service QoS on-line prediction method, the accuracy for prediction is effectively improved by using a matrix decomposition algorithm combining geographic characteristic; and in addition, individual QoS query requests of a plurality of users can be responded in real time by using an optimized matrix decomposition algorithm.

Description

A kind of Web service QoS on-line prediction method based on user's geographical location information

Technical field

The invention belongs to the web service field, relate in particular to a kind of Web service QoS on-line prediction method based on user's geographical location information.

Background technology

Along with the continuous development of Web 2.0 Time Technology revolutions, huge variation is just taking place in main form, the method for operation, the mode of production and the use-pattern of software approach under the internet environment.Based on the Web service dynamic aggregation, the distributed software method of Automatic Combined and elastic telescopic becomes the important trend of future network application and development.These Web service technical application are all launched on QoS research basis.In recent years, the QoS of Web service research becomes the emphasis of industry member and academia's concern.

Research supposes that all Web services all are known to all QoS of target customer's end subscriber, solve this problem through mathematical tool then about Web service QoS at present.Yet under truth, above-mentioned hypothesis is unpractiaca, and reason is following: the Web service framework of (1) Issues and Crucial Practices of Contemporary Enterprises tissue is complicated.For the final user, need the time cost of both expensive just can call all Web services and obtain QoS.(2) contemporary internet topological structure is complicated, and causes users more often can't obtained Web service QoS accurately.Therefore, in real application scenarios, existing a large amount of Web services is unknown to targeted customer's QoS.The basis of former service compute area research has been shaken in the existence of these unknown QoS.Therefore, the QoS to the unknown predicts it is the important prerequisite of Web service research.

In the prior art, to the QoS of the unknown predict main use come between the computing client end subscriber based on Pearson Correlation Coefficient (PCC) method or Web service between similarity.Yet, some deficiency below this computing method exist:

The study 1.PCC method need take statistics to the QoS in the historical record depends critically upon the accuracy and the completeness of data.Yet because the complicacy of contemporary internet environment, the QoS record might not all be accurately, causes PCC method similarity under the service compute scene to calculate accuracy rate and descends.

2. traditional P CC method is widely used in the recommend method field.Yet the application scenarios of recommend method and service compute exists the difference of essence.In the QoS historical record, each all is by user's actual use network environment decision.The characteristics of this data objectivity have directly reduced the precision that the PCC similarity is calculated.

3. traditional P CC algorithm need travel through the QoS record repeatedly and could produce the customer relationship matrix.Yet under the practical application scene, the PCC algorithm can't be made on-line prediction to user's historical record of magnanimity, therefore can only predict through offline mode, can't respond user real time QoS query requests.

Summary of the invention

To above-mentioned technological deficiency, the present invention proposes a kind of Web service QoS on-line prediction method based on user's geographical location information.

In order to solve the problems of the technologies described above, technical scheme of the present invention is following:

A kind of Web service QoS on-line prediction method based on user's geographical location information comprises the steps:

11) collect QoS historical data and the IP information that the user provides;

12) the IP information generating user's who collects according to step 11) geometry geographical position coordinates is calculated the relative distance in user geographic position according to said how much geographical position coordinates, generation user relative distance information matrix;

13) accept targeted customer QoS query requests, and request target User Defined neighbours threshold value θ;

14) the QoS query requests of step 13) being accepted is judged, once calls this QoS query requests like the targeted customer, then resends the QoS information of feeding back last time to the targeted customer; If this QoS query requests is not call, then carry out the QoS prediction;

Said QoS prediction comprises the steps:

141) receive the self-defined neighbours' threshold value of targeted customer θ according to step 13), for the targeted customer selects suitable neighbours; The suitable neighbours' selection strategy of said targeted customer is following:

G(i)＝{j|dist(i，j)≤θ，i≠j} (a)

Wherein (i j) is the relative distance in user geographic position to dist, and the user j that satisfies above-mentioned relation can be defined as the suitable neighbours of targeted customer i;

142) set up constraint condition based on user's geographical location information:

$\min {\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(b\right)$

According to the constraint condition of formula (b), set up the optimization equation that satisfies Web service prediction scene:

$J=\frac{1}{2}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}{(R_{\mathrm{ij}}-U_i^T{S}_j)}^2+\frac{{\mathrm{\λ}}_1}{2}{{\left|\right|U\left|\right|}_F^2+\frac{{\mathrm{\λ}}_2}{2}{\left|\right|S\left|\right|}_F^2+\frac{\mathrm{\α}}{2}\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(c\right)$

Said R is the QoS matrix of user-service; Rij is the QoS operating position of user i to service j.

Iij is that the designator of original matrix promptly exists QoS when record Iij=1, Iij=0 when there is not the QoS record in Rij as Rij; U and S are respectively user's the implicit features matrix and the implicit features matrix of service, and said Ui is the implicit features vector of targeted customer i, λ ₁And λ ₂Be rule factor, α is the factor of control formula (b) geographic rules item degree of participation;

143) adopt the gradient descent method to find the solution to formula (c);

144) find the solution after, for the eigenmatrix U and the S that satisfy end condition, through all QoS information of matrix inner product mode reconstruct original matrix R, i.e. R ≈ U ^TS, targeted customer i is Rij to the QoS predicted value of Web service j;

Said end condition is J-J '≤ε;

Said ε is an iteration threshold; The new J value of said J ' for drawing after the iteration;

15) predicted value is fed back to the targeted customer.

Further, gradient descent method said step 143) is found the solution and is comprised the steps:

21) to formula (c) respectively matrix variables U and S the partial derivative equation solution is obtained:

$\frac{\∂J}{{\∂U}_i}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)\left({-S}_j\right)+{\mathrm{\λ}}_1{U}_i+{\mathrm{\α}}_1(U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g),---\left(d\right)$

$\frac{\∂J}{{\∂S}_j}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)(-U_i)+{\mathrm{\λ}}_2{S}_j,---\left(e\right)$

Then, the gradient descent method has got into the iterative process of matrix variables:

${U_i}^{\′}=U_i-\mathrm{\η}\frac{\∂J}{{\∂U}_i},---\left(f\right)$

${S_J}^{\′}=S_J-\mathrm{\η}\frac{\∂J}{{\∂S}_j},---\left(g\right)$

Wherein η

iteration factor is used for controlling the gradient fall off rate;

22) for the matrix variables U and the S of iteration generation each time, with obtaining J ' in its substitution formula (c), stopping criterion for iteration is for being J-J '≤ε; Said ε is an iteration threshold, ε=0.001.

Further, the QoS information of forecasting Rij of said feedback is packaged into html page formatting, represents the result to the targeted customer through the front end display engine.

Further, the self-defined neighbours' threshold value of said targeted customer θ makes with limit: if θ＜θ ^*, accept θ; If θ>=θ ^*, accept θ ^*Said θ * is the upper limit of threshold value, and this value is set to 1000.

Beneficial effect of the present invention is: the geographical location information Dynamic Selection request user's through using the user similar users crowd.Simultaneously, combine the matrix decomposition algorithm of geographic entity to improve prediction accuracy effectively through having used.In addition, but the personalized QoS query requests through the matrix decomposition algorithm real-time response multi-user that use to optimize.

Description of drawings

Fig. 1 is based on the Web service QoS on-line prediction method flow diagram of user's geographical location information;

Fig. 2 is the internal process figure of QoS on-line prediction engine LBR.

Embodiment

To combine accompanying drawing and specific embodiment that the present invention is done further explanation below.

As shown in Figure 1, overall flow figure of the present invention comprises with the lower part:

Flow process 1: collect QoS historical data and IP information that the user provides.Suppose a total m user and a n service, so: (1) uses historical data to produce the QoS matrix R of user-service of a m*n: wherein each Rij is the QoS operating position of user i to service j.(2) use user's IP information can produce user's geometry geographical position coordinates.Corresponding tuple (alt (i), lat (i)) of each user wherein, the longitude station of alt (i) expression user i, the Position Latitude of lat (i) expression user i.

Flow process 2: according to the relative distance between the geographical first set of calculated user of geometry.Range formula is following:

$\mathrm{dist}(i,j)=\sqrt{{(\mathrm{alt}\left(i\right)-\mathrm{alt}\left(j\right))}^2+{(\mathrm{lat}\left(i\right)-\mathrm{lat}\left(j\right))}^2}\×c,---\left(1\right)$

Wherein c is the constant that becomes unit rice from the longitude and latitude unit conversion.Suppose that the earth is spherical, c is approximately equal to 111261 so.

All users are calculated relative distance, produce user's relative distance information matrix D of a m*m: each dist (i, j) relative distance information of expression user i and user j wherein.

Flow process 3: portal page is accepted targeted customer QoS query requests, and needs the self-defined neighbours' threshold value of targeted customer θ.

Flow process 4: the query requests that flow 3 is gathered performs an analysis:

(1) if former this Qos that called of targeted customer asks service, will there be corresponding QoS record in historical record so, so just can resend original QoS information the front end display engine of flow process 6, generates back page as a result.

(2) if the user did not call this request service in the past, need carry out the QoS prediction of flow process 5 so.

Flow process 5: on-line prediction LBR algorithm engine is the execution entity of QoS prediction.As shown in Figure 2, the sub-process of LBR algorithm engine comprises following a few part:

1) receives the self-defined neighbours' threshold value of targeted customer θ according to flow process 3, for the targeted customer selects suitable neighbours.

The suitable neighbours' selection strategy of targeted customer is following:

Wherein (i, j) by formula (1) definition, the user j that satisfies above-mentioned relation can be defined as the suitable neighbours of targeted customer i to dist.

Experiment shows that the value of θ can exert an influence to final precision of prediction.Take place for fear of the excessive situation that causes estimated performance to descend of θ occurring, the present invention makes with limit for user-defined threshold value θ:

● if θ＜θ ^*, the present invention accepts θ.

● if θ>=θ ^*, the present invention accepts θ ^*

θ * is the upper limit of threshold value.Through experiment show, the present invention θ * is set to 1000.Under this restriction, both can guarantee that engine can find suitable neighbours for the targeted customer, also filter those incoherent neighbours simultaneously, thereby promote precision of prediction.

2) for implementation algorithm can the online targeted customer that calculates to a nicety qos value, the present invention has at first introduced classical SVD forecasting techniques as algorithm template:

$\min \frac{1}{2}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}{(R_{\mathrm{ij}}-U_i^T{S}_j)}^2+\frac{{\mathrm{\λ}}_1}{2}{\left|\right|U\left|\right|}_F^2+\frac{{\mathrm{\λ}}_2}{2}{\left|\right|S\left|\right|}_F^2---\left(3\right)$

Wherein Iij is the designator (there is the QoS record in Iij=1 as Rij, otherwise Iij=0) of original matrix.According to the definition of SVD, U and S are respectively user's the implicit features matrix and the implicit features matrix of service.Two of backs are regularization term, avoid U and S over-fitting in the primal algorithm model.|| U|| _FBe not Luo Beini crow this norm (Frobenius norm) of matrix U, be defined as:

${\left|\right|U\left|\right|}_F=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\left|u_{\mathrm{ij}}\right|}^2}---\left(4\right)$

U wherein _IjElement for the capable j row of the i of matrix U.Right || S|| _FIn like manner, in formula (3), λ ₁And λ ₂Be rule factor, the match speed of gating matrix U and S.The SVD forecasting techniques produces U and the S that satisfies condition through minimizing formula (3), and the inner product of use matrix U and S is all QoS information of original matrix R also.

For flow process 1) obtain the suitable neighbours of targeted customer, the present invention proposes following hypothesis: " the common called Web service of targeted customer and neighbours, the service experience that they obtain should be similar." this hypothesis meets direct feel: because the targeted customer with neighbours owing to be in areal, they will use identical/similar IT infrastructure (network bandwidth and network topology structure etc.) jointly.Also just because of service experience is closely-related with IT infrastructure, so targeted customer and neighbours' implicit features should be similar.Experimental result also shows that this hypothesis is reasonable.

Based on above-mentioned hypothesis, the constraint condition based on user's geographical location information has been proposed:

$\min {\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(5\right)$

Wherein G (i) is by formula (2) definition, and Ui is the implicit features vector of targeted customer i.After obtaining above-mentioned constraint condition, targeted customer and neighbours' geographical relation and traditional SVD forecasting techniques are merged in the present invention, produce the optimization equation that satisfies Web service prediction scene:

$J=\frac{1}{2}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}{(R_{\mathrm{ij}}-U_i^T{S}_j)}^2+\frac{{\mathrm{\λ}}_1}{2}{{\left|\right|U\left|\right|}_F^2+\frac{{\mathrm{\λ}}_2}{2}{\left|\right|S\left|\right|}_F^2+\frac{\mathrm{\α}}{2}\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(6\right)$

λ wherein ₁And λ ₂Be rule factor, α is the factor of control formula (4) geographic rules item degree of participation.

3) for formula (6), the present invention need use the gradient descent method to find the solution.

The gradient descent method is that formula (6) carries out the partial derivative equation solution to the target loss function at first, and there be the matrix variables U and the S of 2 the unknowns in formula (6), so need ask local derviation respectively:

$\frac{\∂J}{{\∂U}_i}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)\left({-S}_j\right)+{\mathrm{\λ}}_1{U}_i+{\mathrm{\α}}_1(U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g)---\left(7\right)$

$\frac{\∂J}{{\∂S}_j}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)(-U_i)+{\mathrm{\λ}}_2{S}_j,---\left(8\right)$

Then, the gradient descent method has got into the iterative process of matrix variables:

${U_i}^{\′}=U_i-\mathrm{\η}\frac{\∂J}{{\∂U}_i},---\left(9\right)$

${S_J}^{\′}=S_J-\mathrm{\η}\frac{\∂J}{{\∂S}_j},---\left(10\right)$

Wherein η

iteration factor is used for controlling the gradient fall off rate.

4) the matrix variables U and the S that produce for iteration each time in the matrix variables U and S substitution formula (6) that newly produce, calculate J ' thereby the result of new formula (6) more.The end condition of iteration is:

J-J′≤ε (11)

ε is an iteration threshold, usually ε=0.001.

If loss function meets above-mentioned end condition, then iterative process stops.If do not satisfy, then return sub-process (3) gradient descent method and continue iteration, till satisfying condition.

5) according to the definition of classical SVD, for the eigenmatrix U and the S that satisfy condition, the present invention passes through all QoS information of formula (12) matrix inner product mode reconstruct original matrix R:

So, targeted customer i is Rij to the QoS value of information of Web service j.

The on-line prediction algorithm engine is the core of response user QoS inquiry.Under truth, the engine demand side is to numerous users' real-time query requests, and this requires algorithm must when improving precision of prediction, reduce complexity computing time.Algorithm time complexity of the present invention mainly is formula (7) and formula (8).Mathematical justification; The iteration time complexity is

each time wherein

be the density of original QoS matrix; D is a constant, is the dimension in implicit features space.The density of time complexity and original matrix that can find out iteration each time is linear.Usually, original matrix is very sparse, so the time complexity of single iteration is very low.Simultaneously, experiment proof prediction algorithm of the present invention can meet pre-conditioned in 15 left and right sides iteration usually.In sum, but prediction algorithm real-time response multi-user's of the present invention online QoS query requests.

Flow process 6: be responsible for receiving QoS information Rij and being packaged into html page formatting, and represent the result to the user through the front end display engine.

Test result:

QoS Forecasting Methodology that proposes among the present invention for representing of quantizing and the quality between traditional Forecasting Methodology use MAE (Mean Absolute Error) to come the accuracy of evaluation prediction.In order to explain NMAE better, first simple defining is MAE once:

$\mathrm{MAE}=\frac{{\mathrm{\Σ}}_{u,s}|r_{u,s}-{\hat{r}}_{u,s}|}{N}---\left(13\right)$

Wherein, r _{U, s}The disappearance QoS predicted value of expression client user u called Web service s,

The true qos value of expression client user u called Web service s, N representes that the disappearance QoS that predicts is total, MAE is all mean values of error between the actual value separately that predict the outcome together

The data centralization that experiment is used has comprised the detailed recalls information of 339 client users to 5825 Web services, therefore uses the client user-Web service matrix of a 339*5825 size to store.In experiment, this matrix has been divided into two parts: training sample and test sample book.In order to simulate actual environment for use truly as far as possible, the matrix samples that extracts some density randomly is as training sample, and is remaining as test sample book.

Simultaneously, θ=100, α=0.001, d=10 are set.

	5％	10％	15％	20％
					UMEAN	0.8813	0.8794	0.8787	0.8784
IMEAN	0.7888	0.7334	0.6810	0.6255
					UPCC	0.8129	0.7412	0.7060	0.6834
IPCC	0.7916	0.7311	0.6910	0.6310
					LBR	0.5389	0.5292	0.5180	0.4941

With present method UMEAN, IMEAN, UPCC compares with IPCC, and the MAE value of the method LBR among the present invention is littler, and it is more accurate promptly to predict the outcome.

The above only is a preferred implementation of the present invention; Should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the present invention's design; Can also make some improvement and retouching, these improvement and retouching also should be regarded as in protection scope of the present invention.

Claims (4)

1. the Web service QoS on-line prediction method based on user's geographical location information is characterized in that, comprises the steps:

11) collect QoS historical data and the IP information that the user provides;

12) the IP information generating user's who collects according to step 11) geometry geographical position coordinates is calculated the relative distance in user geographic position according to said how much geographical position coordinates, generation user relative distance information matrix;

13) accept targeted customer QoS query requests, and request target User Defined neighbours threshold value θ;

14) the QoS query requests of step 13) being accepted is judged, once calls this QoS query requests like the targeted customer, then resends the QoS information of feeding back last time to the targeted customer; If this QoS query requests is not call, then carry out the QoS prediction;

Said QoS prediction comprises the steps:

141) receive the self-defined neighbours' threshold value of targeted customer θ according to step 13), for the targeted customer selects suitable neighbours; The suitable neighbours' selection strategy of said targeted customer is following:

G(i)＝{j?|dist(i，j)≤θ，i≠j} (a)

Wherein (i j) is the relative distance in user geographic position to dist, and the user j that satisfies above-mentioned relation can be defined as the suitable neighbours of targeted customer i;

142) set up constraint condition based on user's geographical location information:

$\min {\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(b\right)$

According to the constraint condition of formula (b), set up the optimization equation that satisfies Web service prediction scene:

$J=\frac{1}{2}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}{(R_{\mathrm{ij}}-U_i^T{S}_j)}^2+\frac{{\mathrm{\λ}}_1}{2}{{\left|\right|U\left|\right|}_F^2+\frac{{\mathrm{\λ}}_2}{2}{\left|\right|S\left|\right|}_F^2+\frac{\mathrm{\α}}{2}\left|\right|U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g\left|\right|}_F^2---\left(c\right)$

Said R is the QoS matrix of user-service; Rij is the QoS operating position of user i to service j;

Iij is that the designator of original matrix promptly exists QoS when record Iij=1, Iij=0 when there is not the QoS record in Rij as Rij; U and S are respectively user's the implicit features matrix and the implicit features matrix of service, and said Ui is the implicit features vector of targeted customer i, λ ₁And λ ₂Be rule factor, α is the factor of control formula (b) geographic rules item degree of participation;

143) adopt the gradient descent method to find the solution to formula (c);

144) find the solution after, for the eigenmatrix U and the S that satisfy end condition, through all QoS information of matrix inner product mode reconstruct original matrix R, i.e. R ≈ U ^TS, targeted customer i is Rij to the QoS predicted value of Web service j;

Said end condition is J-J '≤ε;

Said ε is an iteration threshold; The new J value of said J ' for drawing after the iteration;

15) predicted value is fed back to the targeted customer.

2. a kind of Web service QoS on-line prediction method based on user's geographical location information according to claim 1 is characterized in that said step 143) the gradient descent method find the solution and comprise the steps:

21) to formula (c) respectively matrix variables U and S the partial derivative equation solution is obtained:

$\frac{\∂J}{{\∂U}_i}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)\left({-S}_j\right)+{\mathrm{\λ}}_1{U}_i+{\mathrm{\α}}_1(U_i-\frac{1}{\left|G\left(i\right)\right|}\underset{g\∈G\left(i\right)}{\mathrm{\Σ}}{U}_g),---\left(d\right)$

$\frac{\∂J}{{\∂S}_j}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{I}_{\mathrm{ij}}(R_{\mathrm{ij}}-U_i^T{S}_j)(-U_i)+{\mathrm{\λ}}_2{S}_j,---\left(e\right)$

Then, the gradient descent method has got into the iterative process of matrix variables:

${U_i}^{\′}=U_i-\mathrm{\η}\frac{\∂J}{{\∂U}_i},---\left(f\right)$

${S_J}^{\′}=S_J-\mathrm{\η}\frac{\∂J}{{\∂S}_j},---\left(g\right)$

Wherein η

iteration factor is used for controlling the gradient fall off rate;

22) for the matrix variables U and the S of iteration generation each time, with obtaining J ' in its substitution formula (c), stopping criterion for iteration is for being J-J '≤ε; Said ε is an iteration threshold, ε=0.001.

3. a kind of Web service QoS on-line prediction method according to claim 1 based on user's geographical location information; It is characterized in that; The QoS information of forecasting Rij of said feedback is packaged into html page formatting, represents the result to the targeted customer through the front end display engine.

4. a kind of Web service QoS on-line prediction method based on user's geographical location information according to claim 1 is characterized in that the self-defined neighbours' threshold value of said targeted customer θ makes with limit:

If θ＜θ ^*, accept θ; If θ>=θ ^*, accept θ ^*Said θ * is the upper limit of threshold value, and this value is set to 1000.

Images