CN102170449A - Web service QoS prediction method based on collaborative filtering - Google Patents

Abstract

The invention discloses a Web service QoS prediction method based on collaborative filtering. The result of the QoS predicted by the server terminal is more accurate through calculating the degree of similarity of the Web services by the A-Cosine method and solving the data sparseness problem by using the information of QoS of a similar Web service aiming at a similar client-side user.

Description

A kind of Web service QoS Forecasting Methodology based on collaborative filtering

Technical field

The invention belongs to the web service field, mainly realize a kind of Web service QoS Forecasting Methodology based on collaborative filtering.

Background technology

Along with the rise of networking application and " software is as service " theory, huge variation is just taking place in software system in main shape, operational mode, the mode of production and occupation mode under the internet environment.Reuse and dynamic aggregation becomes the important trend of future network software development with the distribution application system of the loose coupling that makes up on-demand by service.Yet along with the explosive increase of quantity of service, how a large amount of services that function is identical, nonfunctional characteristics is different that distributing on the network select the more excellent service of quality to become a ubiquitous problem in the suitable services set of function.Web service based on QoS is selected to a popular domain of research in recent years.

Suppose all that about the research of selecting based on the Web service of QoS all web services all are known at the QoS of target customer's end subscriber at present, then by solving this problem based on Optimum Theory or graph theory method.Yet, in true application scenarios, suppose all Web services at the QoS of target customer's end subscriber all be known be unpractiaca, the reasons are as follows: 1) because the difference of factors such as different clients user's geographical position or network environment, identical Web service may be different at different clients user's QoS; 2) client user is difficult to call all Web services.Therefore, in real scene, existing some Web service is unknown at the QoS of target customer's end subscriber, and these unknown QoS have influenced accuracy and the completeness of selecting based on the Web service of QoS.QoS at the unknown predicts the accuracy of the Web service selection that is based on QoS and the important leverage of completeness.

In the prior art, the QoS of the unknown predicted mainly contain following some deficiency:

1. existing Web service QOS Forecasting Methodology mainly adopts Pearson Correlation Coefficient (PCC) method to come between the computing client end subscriber or the similarity between the Web service.By observation to real Web service QoS data, we find the different clients user the QoS range differences apart from obviously, it is all lower to be that some client user calls the QoS of all Web services, and has some client users to call the QoS of any Web service all than higher.Because residing network gateway is more, and fail safe is stronger as, client user A, and it is all long that he calls the time that all Web services need, and client user B is owing to be in the fast network, and it is all shorter to call all Web service times.

2. existing Web service QoS Forecasting Methodology all is to use based on the client user or based on the QoS of Web service prediction, also has some researchs all will combine based on the client user with based on the Forecasting Methodology of Web service and predicts.Yet because the sparse property of QoS data, the information that obtains from similar client user and similar Web service also is not enough to provide accurate QoS prediction.

Summary of the invention

At the deficiencies in the prior art, the present invention proposes a kind of Web service QoS Forecasting Methodology based on collaborative filtering, by calculating similarity between the Web service with the A-Cosine method, utilizing similar web service that similar client user's QoS information is solved the sparse problem of data, make prediction result more accurate.

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

A kind of Web service QoS Forecasting Methodology based on collaborative filtering comprises the steps:

1) client user proposes to select demand based on the Web service of QoS to service end;

2) described service end if described web service is crossed in existing client user's never call, then is changed to sky according to described client user's demand and the QoS data generating feature vector between existing client user and the described service end web service;

3) described server end is by the similarity between the described client user of formula 1 calculating, by the similarity between the described web service of formula 2 calculating

$\mathrm{Sim}(u_1,u_2)=\frac{{\mathrm{\Σ}}_{s\∈S}(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}{\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})}^2}\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}^2}}---1$

$S=S_{u_1}\∩S_{u_2}$

$\mathrm{Sim}(s_1,s_2)=\frac{{\mathrm{\Σ}}_{u\∈U}(r_{u,s_1}-\stackrel{\‾}{r_u})(r_{u,s_2}-\stackrel{\‾}{r_u})}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_1}-\stackrel{\‾}{r_u})}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_2}-\stackrel{\‾}{r_u})}^2}}---2$

$U=U_{s_1}\∩U_{s_2};$

Wherein

Be client user u ₁With the set of all Web services of crossing,

For all called Web service s ₁Client user's set, r _{U, s}The QoS of expression client user u called Web service s,

The mean value of the QoS of the Web service of being called for client user u;

4) according to the Top-K principle, described server end is selected client user similar in the step 3) and similar web service;

5) described server end calculates the similarity of QoS between the similar web service of similar client user by formula 3;

Sim(u ₁s ₁，us)＝Sim(u ₁，u)×Sim(s ₁，s) 3

6) described server end calculates predicting the outcome based on the client user by formula 4, calculate predicting the outcome by formula 5 based on the web service, calculate based on predicting the outcome that the similar web of similar client user serves by formula 6, and calculate the confidence level that predicts the outcome separately respectively;

$\mathrm{UPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_u}+\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}S{\mathrm{im}}^{\′}(u_1,u)(r_{u_1,s}-\stackrel{\‾}{r_{u_2}})}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}{\mathrm{Sim}}^{\′}(u_1,u)}---4$

${\mathrm{con}}_u={\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\frac{\mathrm{Sim}(u_1,u)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\mathrm{Sim}(u_1,u)}\×\mathrm{Sim}(u_1,u),$

Con wherein _uBe the confidence level that predicts the outcome based on the client user,

$\mathrm{SPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_s}+\frac{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}S{\mathrm{im}}^{\′}(s_1,s)(r_{u,s_1}-\stackrel{\‾}{r_{s_1}})}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}{\mathrm{Sim}}^{\′}(s_1,s)}---5$

${\mathrm{con}}_s={\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\frac{\mathrm{Sim}(s_1,s)}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\mathrm{Sim}(s_1,s)}\×\mathrm{Sim}(s_1,s),$

Con _sBe the confidence level that predicts the outcome based on the web service

$\mathrm{USPre}\left(r_{u,s}\right)=\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})P_{u,s}(u_1,s_1)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})},---6$

${\mathrm{con}}_{\mathrm{us}}={\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\frac{\mathrm{Sim}{(u_1{s}_1,\mathrm{us})}^2}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})}$

Con wherein _UsBe the confidence level that predicts the outcome based on the similar Web service of similar client user, P _{U, s}(u ₁, s ₁) expression by

To r _{U, s}What go out predicts the outcome, and it is calculated as follows:

$P_{u,s}(u_1,s_1)=(r_{u_1,s_1}-\frac{\stackrel{\‾}{r_{u_1}}+\stackrel{\‾}{r_{s_1}}-\stackrel{\‾}{r_u}-\stackrel{\‾}{r_s}}{2});$

7) described server end calculates described three kinds of weights that predict the outcome respectively by formula 7,8,9, and obtains final predicting the outcome by formula 10

$w_u=\frac{\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},---7$

$w_s=\frac{(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},---8$

$w_{\mathrm{us}}=\frac{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}---9$

Pre(r _u，s)＝w _u×UPre(r _u，s)+w _s×SPre(r _u，s)+w _us×USPre(r _u，s) 10

Wherein, w _u, w _s, w _UsBe based on predicting the outcome of client user respectively, based on predicting the outcome and the weight that predicts the outcome of serving of web service based on the similar web of similar client user;

8) described server end is according to finally predicting the outcome, and choosing is deleted in service to candidate web;

9) the candidate web service that screening obtains according to step 8) according to described client user's QoS demand, is carried out the web services selection.

As possibility: comprise also between described step 4) and the step 5 that Qos to described similar client user and described similar web service carries out Data Smoothing by formula 11 and handles;

${r^{\′}}_{u_{\mathrm{\τ}},s}=\stackrel{\‾}{r_{u_{\mathrm{\τ}}}}+\mathrm{\Δ}{r}_{C_u}\left(s\right),---11$

Wherein

Expression cluster C _uMiddle client user is to the average QoS deviation of Web service s, and concrete computing formula is as follows:

$\mathrm{\Δ}{r}_{C_u}\left(s\right)=\frac{{\mathrm{\Σ}}_{u^{\′}\∈C_u\left(s\right)}(r_{u^{\′},s}-\stackrel{\‾}{r_{u^{\′}}})}{\left|C_u\left(s\right)\right|},$

Wherein, C _u(s) expression cluster C _uIn all called the client user's of web service s set.Beneficial effect of the present invention is: by calculate the similarity between the Web service with the A-Cosine method, use

Eliminated the influence that different clients user's different QoS scope predicts the outcome to final QoS; And use Data Smoothing method to improve the accuracy rate of prediction more; Utilize similar web service that similar client user's QoS information is solved the sparse problem of data, make prediction result more accurate.

Description of drawings

Fig. 1 selects Organization Chart based on the Web service of collaborative filtering;

Fig. 2 is Web service QoS Forecasting Methodology internal process figure.

Embodiment

The present invention will be further described below in conjunction with the drawings and specific embodiments.

As shown in Figure 1, select to comprise following components based on the Web service of collaborative filtering:

(1) Request: the client user proposes to select demand based on the Web service of QoS;

(2) QoS Prediction: server end is predicted the QoS of each web service to target customer's end subscriber according to client user's demand and the QoS data of collecting.

(3) Pruning: server end is according to predicting the outcome, and service is deleted and cut to candidate web, with some obviously by other services " domination " and service from candidate queue, remove.

(4) QoS-based Selection: the candidate web service that screening obtains according to the Pruning process, according to client user's QoS demand, carry out Web service and select.

Above-mentioned steps 2) in, it is a ring of most critical that server end QoS predicts the outcome, and mainly adopts the PCC method to come between the computing client end subscriber or the similarity between the Web service in existing Web service QoS Forecasting Methodology.But by to the observation of real Web service QoS data, find the different clients user the QoS range differences apart from obviously, in order to eliminate the gap of this QoS scope, propose in this example to calculate the similarity of web between serving with the A-Cosine method.

At first variable and the formula that needs in the Forecasting Methodology to use carried out some definition;

Be client user u ₁With the set of all Web services of crossing, For all called Web service s ₁Client user's set, r _{U, s}The QoS of expression client user u called Web service s,

The mean value of the QoS of the Web service of being called for client user u.

Client user's calculating formula of similarity:

$\mathrm{Sim}(u_1,u_2)=\frac{{\mathrm{\Σ}}_{s\∈S}(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}{\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})}^2}\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}^2}}---\left(2.1\right)$

Wherein $S=S_{u_1}\∩S_{u_2}.$

Web service calculating formula of similarity based on A-Cosine:

$\mathrm{Sim}(s_1,s_2)=\frac{{\mathrm{\Σ}}_{u\∈U}(r_{u,s_1}-\stackrel{\‾}{r_u})(r_{u,s_2}-\stackrel{\‾}{r_u})}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_1}-\stackrel{\‾}{r_u})}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_2}-\stackrel{\‾}{r_u})}^2}}---\left(2.2\right)$

Wherein $U=U_{s_1}\∩U_{s_2}.$

In formula 2.2, we have introduced the A-Cosine method, by using

Eliminated the influence that different clients user's different QoS scope predicts the outcome to final QoS.

In the prior art about the Forecasting Methodology of the QoS of Web service in final prediction, often the QoS with the Web service of client user's never call is made as Null, though method is simple for this, final predicted value is reduced, influenced the precision of prediction of QoS.The method of a DataSmoothing has been proposed in the present invention, QoS to the Web service of client user's never call carries out an interpolation processing, by the client user is carried out cluster according to similarity, from the residing cluster of this client user, get average qos value to this Web service as the qos value of this client user, thereby improve the accuracy rate of prediction this Web service.The client user gathers U={u ₁, u ₂..., u _n, the client user's similarity that obtains according to formula 1 by the K-Means clustering method obtains K cluster, wherein client user u with its cluster _tBelong to cluster C _u

Data?Smoothing：

${r^{\′}}_{u_{\mathrm{\τ}},s}=\stackrel{\‾}{r_{u_{\mathrm{\τ}}}}+\mathrm{\Δ}{r}_{C_u}\left(s\right),---\left(2.3\right)$

Wherein

Expression cluster C _uMiddle client user is to the average QoS deviation of Web service s, and concrete computing formula is as follows:

$\mathrm{\Δ}{r}_{C_u}\left(s\right)=\frac{{\mathrm{\Σ}}_{u^{\′}\∈C_u\left(s\right)}(r_{u^{\′},s}-\stackrel{\‾}{r_{u^{\′}}})}{\left|C_u\left(s\right)\right|},$

Wherein, C _u(s) expression cluster C _uIn all called the client user's of Web service s set.

The advantage of above-mentioned Data Smoothing is to make the accuracy improved prediction more, but at some accuracy to be required be not under the very harsh situation, uses prior art also can.

QoS prediction based on the client user

$\mathrm{UPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_u}+\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}S{\mathrm{im}}^{\′}(u_1,u)(r_{u_1,s}-\stackrel{\‾}{r_{u_2}})}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}{\mathrm{Sim}}^{\′}(u_1,u)}---(2.4)$

QoS prediction based on Web service

$\mathrm{SPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_s}+\frac{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}S{\mathrm{im}}^{\′}(s_1,s)(r_{u,s_1}-\stackrel{\‾}{r_{s_1}})}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}{\mathrm{Sim}}^{\′}(s_1,s)}---\left(2.5\right)$

Existing Web service QoS Forecasting Methodology all be to use in formula 2.4 and the formula 2.5 based on the client user or based on the QoS of Web service prediction, also there are some researchs all will combine and predict " Zibin Zheng, ICWS ' 09 " based on the client user with based on the Forecasting Methodology of Web service.Yet because the sparse property of QoS data, the information that obtains from similar client user and similar Web service also is not enough to provide accurate QoS prediction.Therefore, can utilize similar Web service that similar client user's QoS information is solved the sparse problem of data (Data Sparsity Problem).Given u ₁Be the similar client user of u, s ₁Be the similar Web service of Web service s, then and r _{U, s}Between similarity calculate by following formula 2.6.

Sim(u ₁s ₁，us)＝Sim(u ₁，u)×Sim(s ₁，s) (2.6)

Based on formula 2.7, this example proposes a kind of QoS Forecasting Methodology based on the similar Web service of similar client user.

QoS prediction based on the similar Web service of similar client user

$\mathrm{USPre}\left(r_{u,s}\right)=\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})P_{u,s}(u_1,s_1)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})},---\left(2.7\right)$

Wherein

Expression is by right

That makes predicts the outcome, and it is calculated as follows:

$P_{u,s}(u_1,s_1)=(r_{u_1,s_1}-\frac{\stackrel{\‾}{r_{u_1}}+\stackrel{\‾}{r_{s_1}}-\stackrel{\‾}{r_u}-\stackrel{\‾}{r_s}}{2})$

By with formula 2.4,2.5, three in 2.7 predict the outcome and carry out combination, obtain final predicting the outcome.

Because three predict the outcome and drawn by different data, so its credibility also differs, and can calculate its result's credibility by its similarity.Concrete formula is as follows:

The reliability forecasting formula:

${\mathrm{con}}_u={\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\frac{\mathrm{Sim}(u_1,u)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\mathrm{Sim}(u_1,u)}\×\mathrm{Sim}(u_1,u),$

${\mathrm{con}}_s={\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\frac{\mathrm{Sim}(s_1,s)}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\mathrm{Sim}(s_1,s)}\×\mathrm{Sim}(s_1,s),---\left(2.8\right)$

${\mathrm{con}}_{\mathrm{us}}={\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\frac{\mathrm{Sim}{(u_1{s}_1,\mathrm{us})}^2}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})}$

Wherein, con _uBe the confidence level that predicts the outcome based on the client user, con _sBe the confidence level that predicts the outcome based on Web service, con _UsBe the confidence level that predicts the outcome based on the similar Web service of similar client user.Simultaneously, in the present invention, designed two parameter δ and θ and adjusted three kinds of weights that predict the outcome in final result.

Weight predicts the outcome:

$w_u=\frac{\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},$

$w_s=\frac{(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},---\left(2.9\right)$

$w_{\mathrm{us}}=\frac{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}$

Wherein, w _u, w _s, w _UsBe based on predicting the outcome of client user respectively, based on predicting the outcome and the weight that predicts the outcome of Web service based on the similar Web service of similar client user.The final formula that predicts the outcome is as follows:

Finally predict the outcome:

Pre(r _u，s)＝w _u×UPre(r _u，s)+w _s×SPre(r _u，s)+w _us×USPre(r _u，s)(2.10)

Web service QoS predicts internal process as shown in Figure 2:

(1) according to the QoS data between client user who has had and the web service, the generating feature vector if this web service is crossed in client user's never call, then is changed to null;

(2) according between formula 2.1, the 2.2 computing client end subscribers and the similarity between the Web service;

(3) according to Top-K principle (K Web service before selecting), select similar client user and similar Web service;

(4) QoS to similar client user and similar Web service uses formula 2.3 to carry out Data Smoothing processing;

(5) pass through the similarity that formula 2.6 calculates QoS between the similar Web service of similar client users;

(6) calculate based on the predicting the outcome of client user, based on predicting the outcome of Web service and predicting the outcome by formula 2.4,2.5,2.7 based on the similar Web service of similar client user;

(7) calculate three kinds of confidence levels that predict the outcome and weight by formula 2.8,2.9,2.10, and obtain final predicting the outcome.

Test result:

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

$\mathrm{MAE}=\frac{{\mathrm{\Σ}}_{u,s}|r_{u,s}-{\hat{r}}_{u,s}|}{N}$

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 represents the disappearance QoS sum predicted, MAE be all predict the outcome with separately between the actual value the total institute of the mean value of error known, different Web services qos value size separately is can difference very big, uses the problem that MAE can not very objective reflection accuracy therefore merely.For the influence of the gap between the QoS span of eliminating different Web services to the quantitative evaluation objectivity, we have adopted the NMAE after the normalization:

$\mathrm{NMAE}=\frac{\mathrm{MAE}}{{\mathrm{\Σ}}_{i,j}\frac{r_{i,j}}{N}}---\left(2.11\right)$

From formula 2.11 we as can be seen, QoS predicts the outcome more accurately, the NMAE of generation calculates in institute will be more little.

Data centralization that experiment is used has comprised the detailed recalls information of 100 client users to 150 Web services, so we use the client user-Web service matrix of a 150*100 size to store.In experiment, this matrix has been divided into two parts, wherein, the capable training matrix that is used as of p wherein, remaining (150-p) row is then as test matrix.Simultaneously, in order to simulate actual environment for use truly as far as possible, we suppose that each client user in the training set only called 5% in 100 Web services separately, and therefore, we are sparse in 5% with the density of training matrix randomly.In addition, we have also carried out to a certain degree sparse to test matrix, for each client user only keeps Given qos value, promptly suppose the Web service that each client user to be predicted only called Given quantity, the value of Given is respectively 5,10,20.Simultaneously, we are provided with θ=0.1, δ=0.05, top-K=10.

With present method UPCC, IPCC compares with WSRec, and the NMAE value of the method ADF in this patent 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, without departing from the inventive concept of the premise; can also make some improvements and modifications, these improvements and modifications also should be considered within the scope of protection of the present invention.

Claims (2)

1. the Web service QoS Forecasting Methodology based on collaborative filtering is characterized in that: comprise the steps:

1) client user proposes to select demand based on the Web service of QoS to service end;

2) described service end if described web service is crossed in existing client user's never call, then is changed to sky according to described client user's demand and the QoS data generating feature vector between existing client user and the described service end web service;

3) described server end is by the similarity between the described client user of formula 1 calculating, by the similarity between the described web service of formula 2 calculating

$\mathrm{Sim}(u_1,u_2)=\frac{{\mathrm{\Σ}}_{s\∈S}(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}{\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_1,s}-\stackrel{\‾}{r_{u_1}})}^2}\sqrt{{\mathrm{\Σ}}_{s\∈S}{(r_{u_2,s}-\stackrel{\‾}{r_{u_2}})}^2}}---1$

$S=S_{u_1}\∩S_{u_2}$

$\mathrm{Sim}(s_1,s_2)=\frac{{\mathrm{\Σ}}_{u\∈U}(r_{u,s_1}-\stackrel{\‾}{r_u})(r_{u,s_2}-\stackrel{\‾}{r_u})}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_1}-\stackrel{\‾}{r_u})}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(r_{u,s_2}-\stackrel{\‾}{r_u})}^2}}---2$

$U=U_{s_1}\∩U_{s_2}$

Wherein Be client user u ₁With the set of all Web services of crossing,

For all called Web service s ₁Client user's set, r _{U, s}The QoS of expression client user u called Web service s,

The mean value of the QoS of the Web service of being called for client user u;

4) according to the Top-K principle, described server end is selected client user similar in the step 3) and similar web service;

5) described server end calculates the similarity of QoS between the similar web service of similar client user by formula 3;

Sim(u ₁s ₁，us)＝Sim(u ₁，u)×Sim(s ₁，s) 3

6) described server end calculates predicting the outcome based on the client user by formula 4, calculate predicting the outcome by formula 5 based on the web service, calculate based on predicting the outcome that the similar web of similar client user serves by formula 6, and calculate the confidence level that predicts the outcome separately respectively;

$\mathrm{UPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_u}+\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}S{\mathrm{im}}^{\′}(u_1,u)(r_{u_1,s}-\stackrel{\‾}{r_{u_2}})}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}{\mathrm{Sim}}^{\′}(u_1,u)}---4$

${\mathrm{con}}_u={\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\frac{\mathrm{Sim}(u_1,u)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right)}\mathrm{Sim}(u_1,u)}\×\mathrm{Sim}(u_1,u),$

Con wherein _uBe the confidence level that predicts the outcome based on the client user,

$\mathrm{SPre}\left(r_{u,s}\right)=\stackrel{\‾}{r_s}+\frac{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}S{\mathrm{im}}^{\′}(s_1,s)(r_{u,s_1}-\stackrel{\‾}{r_{s_1}})}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}{\mathrm{Sim}}^{\′}(s_1,s)}---5$

${\mathrm{con}}_s={\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\frac{\mathrm{Sim}(s_1,s)}{{\mathrm{\Σ}}_{s_1\∈S\left(s\right)}\mathrm{Sim}(s_1,s)}\×\mathrm{Sim}(s_1,s),$

Con _sBe the confidence level that predicts the outcome based on the web service

$\mathrm{USPre}\left(r_{u,s}\right)=\frac{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})P_{u,s}(u_1,s_1)}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})},---6$

${\mathrm{con}}_{\mathrm{us}}={\mathrm{\Σ}}_{u_1\∈S\left(u\right),s_1\∈S\left(s\right)}\frac{\mathrm{Sim}{(u_1{s}_1,\mathrm{us})}^2}{{\mathrm{\Σ}}_{u_1\∈S\left(u\right),s\∈S\left(s\right)}\mathrm{Sim}(u_1{s}_1,\mathrm{us})}$

Con wherein _UsBe the confidence level that predicts the outcome based on the similar service of similar client user, P _{U, s}(u ₁, s ₁) expression by

To r _{U, s}That makes predicts the outcome, and it is calculated as follows:

$P_{u,s}(u_1,s_1)=(r_{u_1,s_1}-\frac{\stackrel{\‾}{r_{u_1}}+\stackrel{\‾}{r_{s_1}}-\stackrel{\‾}{r_u}-\stackrel{\‾}{r_s}}{2});$

7) described server end calculates described three kinds of weights that predict the outcome respectively by formula 7,8,9, and obtains final predicting the outcome by formula 10

$w_u=\frac{\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},---7$

$w_s=\frac{(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s},---8$

$w_{\mathrm{us}}=\frac{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}}{\mathrm{\δ}{\mathrm{con}}_{\mathrm{us}}+\mathrm{\θ}(1-\mathrm{\δ}){\mathrm{con}}_u+(1-\mathrm{\θ})(1-\mathrm{\δ}){\mathrm{con}}_s}---9$

Pre(r _u，s)＝w _u×UPre(r _u，s)+w _s×SPre(r _u，s)+w _us×USPre(r _u，s)?10

Wherein, w _u, w _s, w _UsBe based on predicting the outcome of client user respectively, based on predicting the outcome and the weight that predicts the outcome of serving of web service based on the similar web of similar client user;

8) described server end is according to finally predicting the outcome, and choosing is deleted in service to candidate web;

9) the candidate web service that screening obtains according to step 8) according to described client user's QoS demand, is carried out the web services selection.

2. a kind of Web service QoS Forecasting Methodology based on collaborative filtering according to claim 1 is characterized in that: comprise also between described step 4) and the step 5 that the Qos to described similar client user and described similar web service carries out Data Smoothing processing by formula 11;

${r^{\′}}_{u_{\mathrm{\τ}},s}=\stackrel{\‾}{r_{u_{\mathrm{\τ}}}}+\mathrm{\Δ}{r}_{C_u}\left(s\right),---11$

Wherein

(s) expression cluster C _uMiddle client user is to the average QoS deviation of Web service s, and concrete computing formula is as follows:

$\mathrm{\Δ}{r}_{C_u}\left(s\right)=\frac{{\mathrm{\Σ}}_{u^{\′}\∈C_u\left(s\right)}(r_{u^{\′},s}-\stackrel{\‾}{r_{u^{\′}}})}{\left|C_u\left(s\right)\right|},$

Wherein, C _u(s) expression cluster C _uIn all called the client user's of web service s set.

Images