CN105024886A - Rapid Web server QoS (Quality of Service) prediction method based on user metadata - Google Patents

Abstract

The invention discloses a rapid Web server QoS (Quality of Service) prediction method based on user metadata. According to the method, the field of a request user is accurately selected through use of the metadata (geographic position information and history QoS record) of a user, and meanwhile the accuracy of prediction is effectively increased through a matrix decomposition algorithm for fusing user extension intrinsic attributes. The solving efficiency is increased greatly through a matrix decomposition acceleration technology, so that real-time response to the individual QoS inquiry requests of a plurality of users is realized.

Description

A kind of Fast W eb service QoS Forecasting Methodology based on user metadata

Technical field

The invention belongs to Web service technology field, be specifically related to a kind of Fast W eb service QoS Forecasting Methodology based on user metadata.

Background technology

Under the promotion of Web 2.0 epoch Internet service tide, just there is huge change in the Main Morphology of traditional software systems, operational mode, the mode of production and occupation mode.The widespread deployment of Web service and application, bring power endlessly to industrial quarters and academia, and promote the sustainable development of service field subject.Under promotion in the industry cycle, QoS (Quality of Service, the service quality) attribute as the basis of service science also result in the concern of numerous service disciplinary study person.In recent years, the normal form based on QoS is also widely used in services selection, service discovery, Services Composition, the field such as service recommendation and service monitoring, greatly promotes the development that moves ahead of service subject.

At present, the development in above-mentioned field all needs to excavate and analytical work a large amount of QoS data resource.But under real conditions, this data demand ratio is more difficult to get satisfied, and reason is as follows: (1) most of Service Source is all held by commercial company and mechanism.If obtain QoS data resource by the mode calling this kind of service, researcher spends a large amount of money costs by needing.(2) contemporary internet development is maked rapid progress, and has every day a large amount of services to emerge in large numbers.If obtain QoS data resource by the mode calling the service of these magnanimity, researcher will require a great deal of time cost.Therefore, under real application scenarios, a large amount of sufficient QoS resource is difficult to be obtained by the mode of the service of calling.Meanwhile, the disappearance of QoS information is also hindering the development of service science.

In recent years, the problem solving QoS scarcity obtains the great attention of researcher.How researcher's thinking is by using the QoS data resource of machine learning method to disappearance to predict.In work on hand, the QoS of the unknown is predicted that main Based PC C (the Pearson Correlation Coefficient) method that uses is to calculate the similarity between client user or between Web service.But there are following 2 deficiencies in this computational methods: (1) PCC method needs to take statistics study to the QoS in historical record, depends critically upon accuracy and the completeness of data.But due to the complexity of contemporary internet environment, QoS record might not be all accurately, PCC method Similarity measures accuracy rate under service compute scene is caused to decline.(2) traditional PCC method is widely used in commending system field.But the application scenarios of commending system and service compute also exists the difference of essence.In QoS historical record, each is all determined by the actual use network environment of user, and the feature of this data objectivity directly reduces the precision of PCC Similarity Measure.

Meanwhile, also there is the problem of system solution overlong time in the QoS prediction work of current main flow.In fact, long running time can bring problem to each side: for researcher, and being obtained QoS data resource by tediously long learning time is not desired by them.And for system designer, they wish the inherent law being carried out learning data resource by less time cost.But current accurate QoS Forecasting Methodology needs plenty of time learning data attribute, also lack the effect analysis to core algorithm complexity simultaneously.

Summary of the invention

For the above-mentioned technical problem existing for prior art, the invention provides a kind of Fast W eb service QoS Forecasting Methodology based on user metadata, effectively can improve the accuracy of prediction, greatly promote solution efficiency, thus the personalized QoS inquiry request of real-time response multi-user.

Based on a Fast W eb service QoS Forecasting Methodology for user metadata, comprise the steps:

(1) collect the metadata that all users provide, described metadata comprises the QoS data of user to all Web services that it called and the IP address information of user; And then set up the QoS variable matrix R between user and Web service according to described QoS data;

(2) Euclidean distance about geographical position and the similarity about QoS data between user is calculated according to described metadata;

(3) according to described Euclidean distance and similarity, the neighborhood user set of each user is determined;

(4) gather according to the neighborhood user of described QoS variable matrix R and each user, by the following target function Q of SVD (singular value decomposition) method establishment, and this target function Q is minimized solve, in the hope of the implicit features matrix U about user and the implicit features matrix S about Web service; And then according to V=U ^ts rebuilds the QoS prediction matrix V between user and Web service;

$Q=\frac{1}{2}\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{I}_{ij}{\[R_{ij}-{(\mathrm{\α}\·U_i(1-\mathrm{\α})\·\frac{1}{K}\underset{g\∈N_2\left(i\right)}{\Σ}{U}_g)}^T{S}_j\]}^2+\frac{\mathrm{\λ}}{2}\left|\right|U|{|}_F^2+\frac{\mathrm{\λ}}{2}|\left|S\right|{|}_F^2$

Wherein: U _ifor the i-th column vector in implicit features matrix U, S _jfor the jth column vector in implicit features matrix S, N ₂i () is the neighborhood user set of i-th user, K is that default neighborhood user number and neighborhood user gather N ₂total number of users in (i), U _gfor neighborhood, user gathers N ₂i vector that g user in () is corresponding in implicit features matrix U, m is the sum of user, and n is the sum of Web service; R _ijfor the i-th row jth column element value in QoS variable matrix R, I _ijfor R _ijdesignator, if R _ijfor null then I _ij=0, otherwise I _ij=1; || || _ffor F-norm, ^trepresent transposition, α and λ is given control coefrficient;

(5) accept the QoS inquiry request of user about Web service, then provide the QoS data of the Web service of its requesting query to predict the outcome according to QoS prediction matrix V to this user.

Set up the QoS variable matrix R between user and Web service according to QoS data in described step (1), be specially: the dimension of described QoS variable matrix R is the i-th row jth column element value R in m × n, QoS variable matrix R _ijadopt i-th user for the QoS data of a jth Web service, if a jth Web service is crossed in never call before i-th user, then element value R _ijfor null.

By the Euclidean distance about geographical position between following formulae discovery user in described step (2):

$dist(i,p)=\sqrt{{(x_i-x_p)}^2+{(y_i-y_p)}^2}\×c$

Wherein: dist (i, p) is the Euclidean distance about geographical position between i-th user and p user, x _iand y _ibe respectively longitude and the latitude of i-th user, x _pand y _pbe respectively longitude and the latitude of p user, the latitude and longitude information of user is transformed by its IP address information and obtains; C is the constant that longitude and latitude unit conversion becomes unit rice, i ≠ p.

By the similarity about QoS data between following formulae discovery user in described step (2):

$sim(i,p)=\frac{\underset{s\∈S}{\Σ}(R_{is}-\stackrel{\‾}{R_i})(R_{ps}-\stackrel{\‾}{R_p})}{\sqrt{\underset{s\∈S}{\Σ}{(R_{is}-\stackrel{\‾}{R_i})}^2}\sqrt{\underset{s\∈S}{\Σ}{(R_{ps}-\stackrel{\‾}{R_p})}^2}}$

Wherein: sim (i, p) is the similarity about QoS data between i-th user and p user, be the QoS data mean value of i-th user to all Web services that it called, be the QoS data mean value of p user to all Web services that it called; S is the Web service set that i-th user and p user called jointly, R _isbe i-th user to the QoS data of s Web service in Web service S set, R _psbe p user to the QoS data of s Web service in Web service S set, i ≠ p.

Determine the neighborhood user set of each user in described step (3), concrete grammar is as follows:

First, for i-th user, extract from other all users and form user with it about the user that geographical position Euclidean distance is less than θ and gather N ₁(i), θ is default distance threshold;

Then, N is gathered from user ₁extract in (i) maximum about QoS data similarity with i-th user before K user neighborhood user of forming i-th user gather N ₂(i).

Preferably, adopt in described step (4) Gradient Descent iterative algorithm to minimize target function Q and solve, and take turns in iterative process at each and construct intermediate variable record sheet; The problem of duplicate keys can be solved, thus accelerating gradient decline solution procedure.

In described step (5), the last QoS data of the Web service of user institute requesting query being predicted the outcome is packaged into html page formatting and is presented to user.

The field of the present invention by using the metadata of user (geographical location information and history QoS record) accurately to select to ask user, the matrix decomposition algorithm simultaneously expanding inherent attribute by merging user improves the accuracy of prediction effectively.The present invention improves solution efficiency greatly by matrix decomposition speed technology, thus the personalized QoS inquiry request of real-time response multi-user.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the Fast W eb service QoS Forecasting Methodology that the present invention is based on user metadata.

Fig. 2 is the internal process schematic diagram of quick QoS prediction algorithm engine.

Embodiment

In order to more specifically describe the present invention, below in conjunction with the drawings and the specific embodiments, technical scheme of the present invention is described in detail.

As shown in Figure 1, the present invention is based on the Fast W eb service QoS Forecasting Methodology of user metadata, comprise with lower part:

Step 1: gather user metadata.Metadata includes two parts: the IP information of (1) user; (2) the history QoS data of user.Suppose total m user and a n service, so:

Historical data is used to produce QoS matrix R: wherein each R of the user-service of a m × n _ijthe QoS service condition of user i to service j.The IP convert information of user is the geometry geographical position coordinates (x of user by operation technique ₁, y ₁).Wherein x ₁represent the longitude station of user, y ₁represent the Position Latitude of user.

Step 2: metadata preliminary treatment.For the two parts gathered, the present invention takes following processing mode respectively:

Calculate the relative distance between user according to the geographical tuple of geometry, range formula is as follows:

$dist(i,j)=\sqrt{{(x_i-x_j)}^2+{(y_i-y_j)}^2}\×c---\left(1\right)$

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

Relative distance is calculated to all users, produces user's relative distance information matrix D of a m × m: wherein each dist (i, j) represents the relative distance information of user i and user j.

The similarity between user is calculated according to history QoS data.Similarity formula is as follows:

$sim(i,j)=\frac{\underset{s\∈S}{\Σ}(R_{is}-\stackrel{\‾}{R_i})(R_{js}-\stackrel{\‾}{R_j})}{\sqrt{\underset{s\∈S}{\Σ}{(R_{is}-\stackrel{\‾}{R_i})}^2}\sqrt{\underset{s\∈S}{\Σ}{(R_{js}-\stackrel{\‾}{R_j})}^2}}---\left(2\right)$

Wherein: s ∈ S is the public service that user i and user j called.The codomain of sim is in [-1,1], and it is higher to be worth similarity between larger expression user.

Relative distance is calculated to all users, produces user's relative distance information matrix D of a m × m: wherein each sim (i, j) represents the similarity of user i and user j.

Step 3: the front-end portal page accepts targeted customer QoS inquiry request, and ask user to input self-defining threshold value θ and TOP_K.

Step 4: the inquiry request that step 3 gathers is performed an analysis:

If targeted customer is former called request service, so will there is corresponding QoS record in system database, so QoS information can be sent to front end display engine, generates result back page.

If user did not call request service in the past, system was so needed to carry out the fast prediction algorithm engine of step 5.

Step 5: fast prediction algorithm engine is execution entity of the present invention.As shown in Figure 2, the sub-process of algorithm engine comprises following a few part:

5.1 calculate neighborhood;

Receive User Defined threshold value θ and TOP_K according to step 3, the invention process two-stage neighborhood calculates, for targeted customer selects high-quality similar users.

First stage: use the relative distance between user to delimit neighborhood choice strategy:

N ₁(i)＝{j|dist(i,j)≤θ,i≠j} (3)

Wherein: dist (i, j) is defined by equation (1), and the user j meeting above-mentioned relation can be defined as the first stage suitable neighbours of targeted customer i.

Second stage: based on first stage neighborhood, uses the similarity between user to delimit the neighborhood choice strategy of neighborhood two-stage:

N ₂(i)＝{j|TOP_K(sim(i,j)),j∈N ₁(i)} (4)

Wherein: sim (i, j) is defined by equation (2), and TOP_K function is the basic operation in database, and the upper bound is inputted by User Defined.The user j meeting above-mentioned relation can be defined as the suitable neighbours of second stage of targeted customer i.

Experiment shows, the value of θ and TOP_K can have an impact to final precision of prediction.In order to avoid occurring that the excessive situation causing estimated performance to decline of θ and TOP_K occurs, the present invention makes following restriction for user-defined threshold value θ and TOP_K:

If θ < is θ ^*with TOP_K < TOP_K ^*, the present invention accepts θ and TOP_K.

If θ>=θ ^*with TOP_K>=TOP_K ^*, the present invention accepts θ ^*and TOP_K ^*.

This restriction both can ensure that engine can delimit high-quality neighborhood for targeted customer, also can filter those incoherent neighbours simultaneously, thus promoted precision of prediction.

5.2 structure optimization equations;

In order to implementation algorithm can calculate to a nicety the qos value of targeted customer, first the present invention introduces classical SVD Predicting Technique as algorithm template:

$\min \frac{1}{2}\underset{i=1}{\overset{m}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}{(R_{ij}-{U_i}^T{S}_j)}^2+\frac{{\mathrm{\&lambda;}}_1}{2}\left|\right|U|{|}_F^2+\frac{{\mathrm{\&lambda;}}_2}{2}|\left|S\right|{|}_F^2---\left(5\right)$

Wherein: I _ijthe designator (I of original matrix _ij=1 works as R _ijthere is QoS record, on the contrary I _ij=0).U and S is respectively the implicit features matrix of user and the implicit features matrix of service.Latter two is regularization term, avoids U and S over-fitting in primal algorithm model.SVD Predicting Technique produces U and S satisfied condition, all QoS information of last reconstruction matrix R by minimum equation (5).

For the neighborhood that sub-process 5.1 delimited, the present invention uses for reference the commending system collaborative filtering thought of current main-stream, by fully learning the data resource of neighborhood, thus predicts the qos value of targeted customer.The data characteristic how learning neighborhood is the key component of collaborative filtering thought.In the present invention, the inherent attribute of user is by U _idescribe.In order to fully absorb the wisdom of neighborhood, the present invention does following expansion to the inherent attribute of targeted customer:

${U_i}^{*}=\mathrm{\&alpha;}\&CenterDot;U_i+(1-\mathrm{\&alpha;})\&CenterDot;\frac{1}{\left|N_2\left(i\right)\right|}\underset{g\&Element;N_2\left(i\right)}{\&Sigma;}{U}_g---\left(6\right)$

Wherein: N ₂i () is defined by formula (4), α is the positive coefficient controlling neighborhood effect.Formula (6) fully have learned the feature of neighborhood in average weighted mode, then expands the inherent attribute of targeted customer.Above-mentioned formula is the mathematical table method mode that the present invention proposes to suppose.

After obtaining above-mentioned formula, targeted customer is expanded inherent attribute in the present invention and traditional SVD Predicting Technique merges, and produces the optimization equation meeting Web service prediction scene:

$\begin{array}{c}Q=\frac{1}{2}\underset{i=1}{\overset{m}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}{(R_{ij}-{U_i}^{*T}{S}_j)}^2+\frac{{\mathrm{\&lambda;}}_1}{2}\left|\right|U|{|}_F^2+\frac{{\mathrm{\&lambda;}}_2}{2}|\left|S\right|{|}_F^2\\ =\frac{1}{2}\underset{i=1}{\overset{m}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}{\&lsqb;R_{ij}-{(\mathrm{\&alpha;}\&CenterDot;U_i+(1-\mathrm{\&alpha;})\&CenterDot;\frac{1}{\left|N_2\left(i\right)\right|}\underset{g\&Element;N_2\left(i\right)}{\&Sigma;}{U}_g)}^T{S}_j\&rsqb;}^2+\frac{{\mathrm{\&lambda;}}_1}{2}\left|\right|U|{|}_F^2+\frac{{\mathrm{\&lambda;}}_2}{2}|\left|S\right|{|}_F^2\end{array}---\left(7\right)$

5.3 accelerating gradients decline and solve;

For formula (7), conventional method uses gradient descent method to solve.But for the characteristic of formula (7), the simple gradient descent method that uses will produce a large amount of duplicate keys.These duplicate keys consume a large amount of computational resources, and do not produce effect to the final prediction effect of lifting.In order to solve duplicate keys problem, the present invention takes turns for each and calculates structure intermediate variable record sheet, directly accelerates the solution procedure of formula (7).

Take turns iteration for t, the present invention is the information of all targeted customers in new formula (6) first more, then builds the following information of intermediate variable table record of 1 m × n:

${{\hat{r}}_{ij}}^{\left(t\right)}={\&lsqb;{U_i}^{*\left(t\right)}\&rsqb;}^T{S_j}^{\left(t\right)}---\left(8\right)$

Last the present invention simplifies following local derviation by the project information making full use of formula (8) and calculates:

$\begin{array}{cc}{\frac{\&part;Q}{\&part;U_i}}^{(t+1)}=\mathrm{\&alpha;}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}({{\hat{r}}_{ij}}^{\left(t\right)}-r_{ij}){S_j}^{\left(t\right)}+\underset{t\&Element;M}{\&Sigma;}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}({{\hat{r}}_{\mathrm{tj}}}^{\left(t\right)}-r_{\mathrm{tj}})(1-\mathrm{\&alpha;}){S_j}^{\left(t\right)}+{\mathrm{\&lambda;}}_U{U_i}^{\left(t\right)}& \left(9\right)\\ {\frac{\&part;Q}{\&part;S_j}}^{(t+1)}=\mathrm{\&alpha;}\underset{i=1}{\overset{m}{\&Sigma;}}{I}_{ij}({{\hat{r}}_{ij}}^{\left(t\right)}-r_{ij}){U_i}^{*\left(t\right)}+{\mathrm{\&lambda;}}_S{S_j}^{\left(t\right)}& \left(10\right)\end{array}$

Then, gradient descent method enters the iterative process of matrix variables:

${U_i}^{(t+1)}={U_i}^{\left(t\right)}-\mathrm{\&eta;}{\frac{\&part;Q}{\&part;U_i}}^{(t+1)}---\left(11\right)$

${S_j}^{(t+1)}={S_j}^{\left(t\right)}-\mathrm{\&eta;}{\frac{\&part;Q}{\&part;S_j}}^{(t+1)}---\left(12\right)$

5.4 judge whether optimization equation meets final requirement;

For matrix variables U and the S of iteration generation each time, we need the result of more new formula (7).The end condition of iteration is:

J-J'≤ε (13)

Wherein: ε is iteration threshold, usual ε=0.001.

If loss function meets above-mentioned end condition, then iterative process stops.If do not meet, then return sub-process 5.3 and continue iteration, until satisfy condition.

For eigenmatrix U and S satisfied condition, the present invention is by all QoS information of matrix combination reconstruct original matrix R:

R≈U ^TS (14)

So, targeted customer i is R to the qos value of Web service j _ij.

On-line prediction algorithm engine is the core of response user QoS inquiry.Under real conditions, system needs the real-time query request in the face of numerous user, and this requires that algorithm must reduce complexity computing time while raising precision of prediction.Algorithms T-cbmplexity of the present invention is mainly formula (9) and formula (10).Mathematical proof, iteration time complexity can relax as O (ρ d) each time: wherein ρ is the density of original QoS matrix, and d is constant, is the dimension in implicit features space.Can find out that the time complexity of iteration and the density of original matrix are linear each time.Usually, original matrix is very sparse, and therefore the time complexity of single iteration is very low.Meanwhile, experiment proves that prediction algorithm of the present invention can meet pre-conditioned in about 10 times iteration usually.In sum, prediction algorithm of the present invention can the online QoS inquiry request of real-time response multi-user.

Claims (7)

1., based on a Fast W eb service QoS Forecasting Methodology for user metadata, comprise the steps:

(1) collect the metadata that all users provide, described metadata comprises the QoS data of user to all Web services that it called and the IP address information of user; And then set up the QoS variable matrix R between user and Web service according to described QoS data;

(2) Euclidean distance about geographical position and the similarity about QoS data between user is calculated according to described metadata;

(3) according to described Euclidean distance and similarity, the neighborhood user set of each user is determined;

(4) gather according to the neighborhood user of described QoS variable matrix R and each user, by the following target function Q of SVD method establishment, and this target function Q is minimized solve, in the hope of the implicit features matrix U about user and the implicit features matrix S about Web service; And then according to V=U ^ts rebuilds the QoS prediction matrix V between user and Web service;

$Q=\frac{1}{2}\underset{i=1}{\overset{m}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{I}_{ij}{\&lsqb;R_{ij}-{(\mathrm{\&alpha;}\&CenterDot;U_i(1-\mathrm{\&alpha;})\&CenterDot;\frac{1}{K}\underset{g\&Element;N_2\left(i\right)}{\&Sigma;}{U}_g)}^T{S}_j\&rsqb;}^2+\frac{\mathrm{\&lambda;}}{2}\left|\right|U|{|}_F^2+\frac{\mathrm{\&lambda;}}{2}|\left|S\right|{|}_F^2$

Wherein: U _ifor the i-th column vector in implicit features matrix U, S _jfor the jth column vector in implicit features matrix S, N ₂i () is the neighborhood user set of i-th user, K is that default neighborhood user number and neighborhood user gather N ₂total number of users in (i), U _gfor neighborhood, user gathers N ₂i vector that g user in () is corresponding in implicit features matrix U, m is the sum of user, and n is the sum of Web service; R _ijfor the i-th row jth column element value in QoS variable matrix R, I _ijfor R _ijdesignator, if R _ijfor null then I _ij=0, otherwise I _ij=1; || || _ffor F-norm, ^trepresent transposition, α and λ is given control coefrficient;

(5) accept the QoS inquiry request of user about Web service, then provide the QoS data of the Web service of its requesting query to predict the outcome according to QoS prediction matrix V to this user.

2. Fast W eb service QoS Forecasting Methodology according to claim 1, it is characterized in that: in described step (1), set up the QoS variable matrix R between user and Web service according to QoS data, be specially: the dimension of described QoS variable matrix R is the i-th row jth column element value R in m × n, QoS variable matrix R _ijadopt i-th user for the QoS data of a jth Web service, if a jth Web service is crossed in never call before i-th user, then element value R _ijfor null.

3. Fast W eb service QoS Forecasting Methodology according to claim 1, is characterized in that: by the Euclidean distance about geographical position between following formulae discovery user in described step (2):

$dist(i,p)=\sqrt{{(x_i-x_p)}^2+{(y_i-y_p)}^2}\&times;c$

Wherein: dist (i, p) is the Euclidean distance about geographical position between i-th user and p user, x _iand y _ibe respectively longitude and the latitude of i-th user, x _pand y _pbe respectively longitude and the latitude of p user, the latitude and longitude information of user is transformed by its IP address information and obtains; C is the constant that longitude and latitude unit conversion becomes unit rice, i ≠ p.

4. Fast W eb service QoS Forecasting Methodology according to claim 1, is characterized in that: by the similarity about QoS data between following formulae discovery user in described step (2):

$sim(i,p)=\frac{\underset{s\&Element;S}{\&Sigma;}(R_{is}-\stackrel{\&OverBar;}{R_i})(R_{ps}-\stackrel{\&OverBar;}{R_p})}{\sqrt{\underset{s\&Element;S}{\&Sigma;}{(R_{is}-\stackrel{\&OverBar;}{R_i})}^2}\sqrt{\underset{s\&Element;S}{\&Sigma;}{(R_{ps}-\stackrel{\&OverBar;}{R_p})}^2}}$

Wherein: sim (i, p) is the similarity about QoS data between i-th user and p user, be the QoS data mean value of i-th user to all Web services that it called, be the QoS data mean value of p user to all Web services that it called; S is the Web service set that i-th user and p user called jointly, R _isbe i-th user to the QoS data of s Web service in Web service S set, R _psbe p user to the QoS data of s Web service in Web service S set, i ≠ p.

5. Fast W eb service QoS Forecasting Methodology according to claim 1, is characterized in that: the neighborhood user set determining each user in described step (3), and concrete grammar is as follows:

First, for i-th user, extract from other all users and form user with it about the user that geographical position Euclidean distance is less than θ and gather N ₁(i), θ is default distance threshold;

Then, N is gathered from user ₁extract in (i) maximum about QoS data similarity with i-th user before K user neighborhood user of forming i-th user gather N ₂(i).

6. Fast W eb service QoS Forecasting Methodology according to claim 1, it is characterized in that: adopt in described step (4) Gradient Descent iterative algorithm to minimize target function Q and solve, and take turns in iterative process at each and construct intermediate variable record sheet.

7. Fast W eb service QoS Forecasting Methodology according to claim 1, is characterized in that: in described step (5), the last QoS data of the Web service of user institute requesting query being predicted the outcome is packaged into html page formatting and is presented to user.