CN104820719A - Web service creditworthiness measuring method based on context data of user - Google Patents

Abstract

A Web service creditworthiness measuring method based on context data of a user comprises the following steps: acquiring the context data of the user when the user accesses Web services, and pre-processing the context data, represented by vectors, of the user by using a normalization manner; clustering the pre-processed context data of the user by using a k-means method so as to conveniently represent the context data of the different users by using different types respectively; then, calculating the influences on feedback grades by using the user context data with the different types according to the user feedback grades and the clustered user context data; converting the feedback grades of the user context data with the different types into a unified reference grade so as to weaken the influences on the feedback grades by the user context data with the different types; and finally, calculating the user similarity according to the converted feedback grades to obtain service creditworthiness corresponding to the context data type of the reference user, and converting the service creditworthiness into the service creditworthiness corresponding to the context data types of the other users.

Description

Based on the Web service credit worthiness measure of user context data

Technical field

The present invention relates to a kind of Web service credit worthiness measure based on user's context, belong to the technical field of computer utility.

Background technology

Services Oriented Achitecture SOA (Service-Oriented Architecture) is a component model, and it defines good interface between being served by these by the different function units (being also called service) of application program and contract connects.The existence of SOA technology more than 20 years, but, be not used widely always.Along with the appearance of Web service, received gradually by people, SOA has welcome oneself " spring " finally.

In open SOA environment, due to the impact by several factors, user can not Deterministic service supplier issues simultaneously service QoS information be objectively, reliably and real.Therefore, effectively weighing authenticity, the credibility of the QoS of ISP's issuing service, is very important for services selection.Credit worthiness is just to the key factor that QoS authenticity, credibility are weighed.Credit worthiness is generally the feedback levels given after using a certain service according to a large number of users, and calculating evaluates a value relevant to services reputation.It represent true, the credibility of a Web service.Therefore, when ISP provides certain function services, not only to meet traditional Q oS requirement, also will meet the demand of consumer for credit worthiness simultaneously.Web service credit worthiness is for important in inhibitings such as services selection.Credit degree of service measure accurately, being conducive to service requester (service consumer/user) can select ISP with a high reputation on the one hand, thus obtains safe and reliable service; On the other hand, be also conducive to ISP promotes self service by measuring the credit worthiness obtained, thus attract more service requester, form the situation of Web service health and stable development and service.Therefore, Web service credit worthiness tolerance becomes study hotspot.

External prior art situation is: document " A trust management framework forservice-oriented environments " (publishing in In:Proc.of the 18th Int ' l Conf.on World WideWeb " WWW 2009 " 2009.289-302. [doi:10.1145/1526709.1526829]) proposes a credible Governance framework of the service based on credit worthiness towards Open Distributed service environment, its core is trusted service management, not only support the trusting relationship of multiple different entities, also support that each entity adopts different credit worthiness score functions to assess same feedback data simultaneously.A remarkable advantage of this framework supports multiple credit assessment, has higher practical value.

Document " A class of hierarchical fuzzy systems with constraints on the fuzzyrules [J]. " (publishes in " Fuzzy Systems; IEEE Transactions " on, 2005,13 (2), 194-203) consider service QoS and user's similarity two kinds of angles, Web service credit worthiness has been measured.This article is expanded on the basis of Traditional Web services model, adds service detection center.Propose the angle combined based on user's Objective and subjective evaluations.The value of the QoS data obtained from service detection center and issue is carried out contrast to upgrade QoS.Then by the similarity of user, based on similar users to service recommendation etc.

Domestic prior art situation is: the people such as the Mei Hong of Peking University propose a kind of authenticity of the user feedback of consideration effectively grade and the credit worthiness measure of accuracy.The method, by the method for statistics, by the enough user feedback data of statistic as standard value, revises other feedback data according to standard value, effectively to revise malice feedback, thus obtains more believable credit worthiness Evaluation Environment.

The people such as the Xu Lanfang of the Central China University of Science and Technology for exist in traditional analysis Malicious clients falseness recommend, propose based on gray system theory, with gray clustering evaluation arithmetic be main contents prestige reporting mechanism scheme.The grading power that the program overcomes each user in traditional analysis is considered as equal way, data can be made more objective, have and assess the advantages such as reliable, workable.

The people such as the Wang Shangguang of Beijing University of Post & Telecommunication propose a kind of appraisal procedure for credit worthiness in QoS perception Web service selection.The main thought of the method is, is verified, corrects and detect this three credit worthiness evaluation modules reply from the impact on credit worthiness assessment objectivity and accuracy of the confusion of user, preference and malice three kinds feedback by feedback.The method proposed effectively improves the objectivity of credit worthiness assessment, also reduced significantly the irrelevance of services selection.

In sum, although existing Web service credit worthiness measure can improve the accuracy of credit degree of service tolerance effectively, but mostly there is following shortcoming in it: all users are considered as same weight by (1) said method, namely think that all effective feedback user factors of influence are identical, and this point is inaccurate obviously.(2) said method have ignored residing environment when user uses service, and the impact that device therefor causes, and namely have ignored this influence factor of user's context, distinguishes the feedback data under varying environment and treats.These problems all result in method existence inaccuracy to a certain extent.

Therefore, how effectively consider the impact that user's context causes user feedback grade and impact is weakened, with measurement service credit worthiness more accurately, becoming the new problem that scientific and technical personnel in the industry pay close attention to, and a large amount of Exploration & stu dy has been carried out to it.

Summary of the invention

In view of this, the object of this invention is to provide a kind of Web service credit worthiness measure based on user context data, the inventive method can in mobile environment, the accurate tolerance of Web service credit worthiness is carried out when user context data is larger on service QoS impact, the method is that when first using service to user, the user context data information at place carries out pre-service, to classify to user context data, user context data according to a large amount of user feedback level data and correspondence is classified, infer the impact that different user context data classification causes feedback levels, and then the impact that reduction user context data causes feedback levels, according to the feedback levels weakened after user's context impact, accurate tolerance credit degree of service.

In order to reach foregoing invention object, the invention provides a kind of Web service credit worthiness measure based on user context data, it is characterized in that: the context data first gathering this user when user accesses Web service, and utilize normalization mode to carry out pre-service to this user with the context data of vector representation, recycling k-means method these pretreated context datas to this user carry out cluster, so that with the different classes of context data representing different user respectively; Then, the impact that different classes of user context data causes feedback levels is calculated according to the user context data after user feedback grade and cluster, the feedback levels of different classes of user context data is converted to unified reference rank, to weaken the impact of different classes of user context data on feedback levels; Finally, calculate user's similarity according to the feedback levels after conversion, and obtain with reference to credit degree of service corresponding to the context data classification of user with this, be finally converted into the credit degree of service that other user context data classifications are corresponding; Described method comprises following operative step:

Step 1, context data when accessing Web service to the user collected performs pre-service: because to user context data have an impact multiple because have, at least comprise CPU, internal memory, screen size and the network bandwidth four factors, and the physics meaning of Different factor, span and measurement unit are different again, therefore first choose and wherein affect large factor, and each factor large on this impact is normalized, vector representation is adopted, as the user context data after normalization again by unified for each factor after normalization; Then by k-means method, cluster is carried out to the user context data after normalization, so as can according to user context data user feedback grade returned be divided into different classes of;

Step 2, calculate different classes of user context data to the impact of feedback levels: according to the user context data after cluster and corresponding feedback levels thereof, for each user calculates the impact that different classes of user context data causes its feedback levels respectively, namely solve each user when accessing same service, under different classes of user context data environment, different classes of user context data feedback levels is converted to unified reference rank, and the difference between the feedback levels that this user is submitted to realizes Weakening treatment;

Step 3, according to the unified user context data after conversion with reference to level calculation user's similarity and feedback similarity thereof: according to the feedback levels of the reference user context data classification that step 2 obtains, the collaborative filtering method based on user is adopted to calculate the Pearson correlation coefficient PCC of two users in its Web service of accessing (Pearson Correlation Coefficient), as the user's similarity between these two users; Then the user choosing similarity higher forms set, then solves the feedback levels similarity between each user;

Step 4, the feedback levels of reference classification user context data obtained respectively according to step 2 and 3 and the feedback levels similarity of user, for the credit degree of service that user metric reference classification user context data is corresponding, finally according to different classes of user context data and with reference to the feedback levels difference between classification user context data, credit worthiness is transformed into sight corresponding to different classes of user context data.

The advantage of the inventive method is: carry out cluster by simple clustering method to user context data, use the multiattribute user context data of categorized representation again, thus solve the statement problem of the user context data that attribute is many, scope meaning is different preferably.Then by solving the impact that different classes of user context data causes feedback levels, to realize the conversion of feedback levels between different user context data classification, the data user rate of feedback levels is added.Be that different user distributes different weight according to user's similarity again, effectively improve the accuracy of credit worthiness tolerance, introduce user's hobby each other and scoring similarity better.Therefore, the advantage of the inventive method is: applicability is strong, can be used in the tolerance of Web service credit worthiness under the complicated network environment such as all kinds mobile network, UNE.The most important thing is, the processing mode of the inventive method is simple, easily realizes, and computation complexity is low.Therefore, the present invention has higher calculating accuracy and stronger practicality, has good application value.

Accompanying drawing explanation

Fig. 1 is the operation steps process flow diagram of the Web service credit worthiness measure that the present invention is based on user context data.

Embodiment

For making the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, the present invention is described in further detail.

The Web service credit worthiness measure that the present invention is based on user context data is the context data of this user first gathered when user accesses Web service, and utilize normalization mode to carry out pre-service to this user with the context data of vector representation, recycling k-means method these pretreated context datas to this user carry out cluster, so that with the different classes of context data representing different user respectively; Then, the impact that different classes of user context data causes feedback levels is calculated according to the user context data after user feedback grade and cluster, the feedback levels of different classes of user context data is converted to unified reference rank, to weaken the impact of different classes of user context data on feedback levels; Finally, calculate user's similarity according to the feedback levels after conversion, and obtain with reference to credit degree of service corresponding to the context data classification of user with this, be finally converted into the credit degree of service that other user context data classifications are corresponding.

See Fig. 1, introduce concrete operation step of the present invention:

Step 1, context data when accessing Web service to the user collected performs pre-service: because to user context data have an impact multiple because have, at least comprise CPU, internal memory, screen size and the network bandwidth four factors, and the physics meaning of Different factor, span and measurement unit are different again, therefore first choose and wherein affect large factor, and each factor large on this impact is normalized, vector representation is adopted, as the user context data after normalization again by unified for each factor after normalization; Then by k-means method, cluster is carried out to the user context data after normalization, so as can according to user context data user feedback grade returned be divided into different classes of.This step 1 comprises following content of operation:

(11) the set WS={ws of all Web services composition is set ₁, ws ₂..., ws _j..., ws _n, in formula, ws is the Web service during this WS gathers, and natural number subscript j is the sequence number of Web service, and its maximal value n is the Web service sum during this WS gathers;

(12) the set U={u of whole user composition is set ₁, u ₂..., u _i..., u _m, in formula, u is the user during this U gathers, and natural number subscript i is the sequence number of user, and its maximal value m is the total number of users during this U gathers;

(13) described user context data refers to the user's context of narrow sense, i.e. user terminal context; Large factor is affected on it and comprises CPU, internal memory, screen size and network bandwidth performance, by each user u _iaccess each Web service ws _jtime corresponding user context data vector representation be in formula, natural number subscript w is the sequence number of user context data attribute, and its maximal value r is the sum of user context data attribute, chooses four attribute: CPU, internal memory, screen size and network bandwidth performance here altogether, therefore r is 4, so just user context data is expressed as $C_{i,j}=(c_{i,j}^1,c_{i,j}^2,c_{i,j}^3,c_{i,j}^4);$

(14) because of the span of each attribute of user context data and measurement unit different, directly use C _i,jrepresent that user context data mistake will occur, therefore will be normalized respectively each attribute of user context data, wherein each factor will be converted into the numerical value between setting district on [0,1]: in formula, for the primitive attribute of user context data, for after normalized numerical value, with the minimum value in user context data w the community set in whole user and maximal value respectively; User context data after such normalization is just expressed as

(15) with k-means clustering method the user context data after normalization returned and be divided into basic, normal, high three classifications, distinguish respective user context data by bad to three good appropriate levels, namely the attribute performance of user context data is better, then its generic is higher, and vice versa: each user u _ieach Web service ws is accessed when use three kind user context data _jafter, the feedback levels obtained is not r respectively _{i, j, L}, r _{i, j, M}, r _{i, j, H}, wherein, subscript L, M, H represent the user context data of Three Estate respectively; And this Three Estate is divided into two parts composition: Internal feedback grade rf _i,jwith the difference rc being subject to user context data impact _i,j; Wherein, Internal feedback grade depends on the feedback of user self to Web service, has nothing to do with user context data, and therefore the difference rc being subject to user context data impact is all depended in the difference of Three Estate _i,j; Respectively by middle rank and the user context data of rudimentary two kinds to user u _ithe difference definition that affects of feedback levels is Δ _{i, 1}, the user context data of senior and intermediate two kinds is to user u _ithe difference definition that affects of feedback levels is Δ _{i, 2}, so just by each user u _ithe feedback levels impact vector representation that is subject to different classes of user context data be (Δ _{i, 1}, Δ _{i, 2}).

Step 2, calculate different classes of user context data to the impact of feedback levels: according to the user context data after cluster and corresponding feedback levels thereof, for each user calculates the impact that different classes of user context data causes its feedback levels respectively, namely solve each user when accessing same service, under different classes of user context data environment, different classes of user context data feedback levels is converted to unified reference rank, and the difference between the feedback levels that this user is submitted to realizes Weakening treatment.This step 2 comprises following content of operation:

(21) impact of different classes of user context data on user feedback grade is calculated: be respectively each user u _icalculate (Δ _{i, 1}, Δ _{i, 2}): ${\mathrm{\Δ}}_{i,1}=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈{\mathrm{WS}}_{i,1}}(r_{i,j,M}-r_{i,j,L})}{n_{i,1}},{\mathrm{\Δ}}_{i,2}=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈{\mathrm{WS}}_{i,2}}(r_{i,j,H}-r_{i,j,M})}{n_{i,2}};$ In formula, WS _{i, 1}represent user u _iused the Web service set of middle rank and rudimentary two kind user context data access, WS simultaneously _{i, 2}represent user u _iused the Web service set of middle rank and senior two kind user context data access, ws simultaneously _jset WS _{i, 1}or WS _{i, 2}in Web service, r _{i, j, L}, r _{i, j, M}, r _{i, j, H}each user u respectively _ito Web service ws when using basic, normal, high level three class users context datas _jfeedback levels, n _{i, 1}, n _{i, 2}set WS respectively _{i, 1}and WS _{i, 2}in the sum of Web service;

(22) Weakening treatment is carried out on the difference that affects of different classes of user context data: (the Δ obtained according to step (21) _{i, 1}, Δ _{i, 2}), by each user u _ibe transformed in unified reference classification user context data in the feedback levels of different classes of user context data; If when being middle rank selected with reference to user context data, then less advanced users context data feedback levels is expressed as r' via after conversion _{i, j, L}, and r' _{i, j, L}=r _{i, j, L}+ Δ _{i, 1}; Advanced level user's contextual feedback grade is expressed as r' via after conversion _{i, j, H}, and r' _{i, j, H}=r _{i, j, H}-Δ _{i, 2}; Thus by feedback levels corresponding for all different classes of user context data after Weakening treatment, be all transformed into unified reference classification, feedback levels when all corresponding to the user context data of intermediate classification: r' _{i, j, L}, r _{i, j, M}, r' _{i, j, H}.

Step 3, according to the unified user context data after conversion with reference to level calculation user's similarity and feedback similarity thereof: according to the feedback levels of the reference user context data classification that step 2 obtains, the collaborative filtering method based on user is adopted to calculate the Pearson correlation coefficient PCC of two users in its Web service of accessing (Pearson Correlation Coefficient), as the user's similarity between these two users; Then the user choosing similarity higher forms set, then solves the feedback levels similarity between each user.

Here it should be noted that: user's similarity is the similarity degree of different user between operation behavior.And the similarity degree numerical value in the feedback levels that to be different user provide identical Web service of the feedback similarity between each user, and the similarity degree numerical value of feedback levels is larger, both are more similar, otherwise then both are more dissimilar.This step 3 comprises following content of operation:

(31) two different user u are calculated _aand u _bsimilarity time, first find the Web service set P=P that these two users accessed jointly _a∩ P _b, in formula, natural number a and b are the sequence numbers of two different users, P _a, P _bbe expressed as user u _aand u _bduring the Web service set of respective access, what obtain that these two users accessed jointly at it is total to | the Pearson correlation coefficient in P| Web service: $\mathrm{Sim}(a,b)=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}(r_{a,j}-\stackrel{\‾}{r_a})(r_{b,j}-\stackrel{\‾}{r_b})}{\sqrt{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}{(r_{a,j}-\stackrel{\‾}{r_a})}^2}\sqrt{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}{(r_{b,j}-\stackrel{\‾}{r_b})}^2}},$ As the similarity between these two different users; In formula, r _a,jand r _b,jtwo user u respectively _aand u _bto Web service ws _jrespective scoring, with two user u respectively _aand u _bto each self feed back expectation value of serving in Web service set P; In formula, r _a,jby user u according to step (22) _abe transformed into reference to rank (c in the feedback levels of three kinds of different classes of user context data (L, M, H) _r) after user context data to Web service ws _jfeedback levels referred to as r _a,j.

(32) return execution step (31), solve respectively and obtain each user u _aand the user's similarity between every other user;

(33) from each user u _aand in user's similarity between other users, choose Top-k similar users, namely select user u _alarger front k similar users composition similar users S set (a) of similarity numerical value: $S\left(a\right)=\left\{b\right|\mathrm{Sim}(a,b)\≥{\mathrm{Sim}}_a^k,\mathrm{Sim}(a,b)>0,a\≠b\},$ Wherein, k is natural number, for user u _auser's similarity that the similar users that a kth numerical value is larger is corresponding;

(34) user u is obtained _asimilar users set after, according to the common used Web service collection of this k user in S set (a) calculate each user u _awith other each user u _bbetween feedback similarity $\mathrm{FSim}(a,b)=\left\{\begin{array}{c}1-\sqrt{\frac{\underset{s_j\∈S^k}{\mathrm{\Σ}}{(r_{a,j}-r_{b,j})}^2}{{10}^2\·l}},l\≠0\\ 0,l=0\end{array}\right.,$ And FSim (a, b) ∈ [0,1]; In formula, l is Web service S set ^kin service sum, r _a,jand r _b,juser u respectively _awith another user u _bto service s _jfeedback levels.

Step 4, the feedback levels of reference classification user context data obtained respectively according to step 2 and 3 and the feedback levels similarity of user, for the credit degree of service that user metric reference classification user context data is corresponding, finally according to different classes of user context data and with reference to the feedback levels difference between classification user context data, credit worthiness is transformed into sight corresponding to different classes of user context data.This step 4 comprises following content of operation:

(41) be intermediate, i.e. c according to reference user context data selected in step (22) _rduring for M, the feedback levels of the intermediate class users context data obtained i.e. r _{a, j, M}basis on, according to formula calculate this user u _awhen with reference to classification user context data, the Web service ws obtained _jcredit worthiness result in formula, c _rfor the selected a certain classification with reference to user context data, be chosen to be middle rank here; S (a) is the user u calculated in step (33) _asimilar users set, FSim (a, b) is two different user u _aand u _bfeedback similarity;

(42) according to step (41) user u _athe Web service ws obtained when being middle rank with reference to classification user context data _jcredit worthiness with this user u that step (21) obtains _a(Δ _{i, 1}, Δ _{i, 2}), calculate this user respectively at other classifications, Web service credit worthiness r namely corresponding to rudimentary and senior two class users context datas _{a, j, L}=r _{a, j, M}-Δ _{a, 1}, r _{a, j, H}=r _{a, j, M}+ Δ _{a, 2}, to obtain under different classes of user context data environment, user u _aaccess Web service ws _jthe different classification measurement results of credit worthiness.

The present invention is under Web service environment, and the feedback levels data that multiple terminals service access collects and user context data have carried out repeatedly Case Experiments On A, and the result of test is successful, achieves goal of the invention.

Claims (6)

1. the Web service credit worthiness measure based on user context data, it is characterized in that: the context data first gathering this user when user accesses Web service, and utilize normalization mode to carry out pre-service to this user with the context data of vector representation, recycling k-means method these pretreated context datas to this user carry out cluster, so that with the different classes of context data representing different user respectively; Then, the impact that different classes of user context data causes feedback levels is calculated according to the user context data after user feedback grade and cluster, the feedback levels of different classes of user context data is converted to unified reference rank, to weaken the impact of different classes of user context data on feedback levels; Finally, calculate user's similarity according to the feedback levels after conversion, and obtain with reference to credit degree of service corresponding to the context data classification of user with this, be finally converted into the credit degree of service that other user context data classifications are corresponding; Described method comprises following operative step:

Step 1, context data when accessing Web service to the user collected performs pre-service: because to user context data have an impact multiple because have, at least comprise CPU, internal memory, screen size and the network bandwidth four factors, and the physics meaning of Different factor, span and measurement unit are different again, therefore first choose and wherein affect large factor, and each factor large on this impact is normalized, vector representation is adopted, as the user context data after normalization again by unified for each factor after normalization; Then by k-means method, cluster is carried out to the user context data after normalization, so as can according to user context data user feedback grade returned be divided into different classes of;

Step 2, calculate different classes of user context data to the impact of feedback levels: according to the user context data after cluster and corresponding feedback levels thereof, for each user calculates the impact that different classes of user context data causes its feedback levels respectively, namely solve each user when accessing same service, under different classes of user context data environment, different classes of user context data feedback levels is converted to unified reference rank, and the difference between the feedback levels that this user is submitted to realizes Weakening treatment;

Step 3, according to the unified user context data after conversion with reference to level calculation user's similarity and feedback similarity thereof: according to the feedback levels of the reference user context data classification that step 2 obtains, the collaborative filtering method based on user is adopted to calculate the Pearson correlation coefficient PCC of two users in its Web service of accessing (Pearson Correlation Coefficient), as the user's similarity between these two users; Then the user choosing similarity higher forms set, then solves the feedback levels similarity between each user;

Step 4, the feedback levels of reference classification user context data obtained respectively according to step 2 and 3 and the feedback levels similarity of user, for the credit degree of service that user metric reference classification user context data is corresponding, finally according to different classes of user context data and with reference to the feedback levels difference between classification user context data, credit worthiness is transformed into sight corresponding to different classes of user context data.

2. method according to claim 1, is characterized in that: described step 1 comprises following content of operation:

(11) the set WS={ws of all Web services composition is set ₁, ws ₂..., ws _j..., ws _n, in formula, ws is the Web service during this WS gathers, and natural number subscript j is the sequence number of Web service, and its maximal value n is the Web service sum during this WS gathers;

(12) the set U={u of whole user composition is set ₁, u ₂..., u _i..., u _m, in formula, u is the user during this U gathers, and natural number subscript i is the sequence number of user, and its maximal value m is the total number of users during this U gathers;

(13) described user context data refers to the user's context of narrow sense, i.e. user terminal context; Large factor is affected on it and comprises CPU, internal memory, screen size and network bandwidth performance, by each user u _iaccess each Web service ws _jtime corresponding user context data vector representation be in formula, natural number subscript w is the sequence number of user context data attribute, and its maximal value r is the sum of user context data attribute, chooses four attribute: CPU, internal memory, screen size and network bandwidth performance here altogether, therefore r is 4, so just user context data is expressed as $C_{i,j}=(c_{i,j}^1,c_{i,j}^2,c_{i,j}^3,c_{i,j}^4);$

(14) because of the span of each attribute of user context data and measurement unit different, directly use C _i,jrepresent that user context data mistake will occur, therefore will be normalized respectively each attribute of user context data, wherein each factor will be converted into the numerical value between setting district on [0,1]: in formula, for the primitive attribute of user context data, for after normalized numerical value, with the minimum value in user context data w the community set in whole user and maximal value respectively; User context data after such normalization is just expressed as

(15) with k-means clustering method the user context data after normalization returned and be divided into basic, normal, high three classifications, distinguish respective user context data by bad to three good appropriate levels, namely the attribute performance of user context data is better, then its generic is higher, and vice versa: each user u _ieach Web service ws is accessed when use three kind user context data _jafter, the feedback levels obtained is not r respectively _{i, j, L}, r _{i, j, M}, r _{i, j, H}, wherein, subscript L, M, H represent the user context data of Three Estate respectively; And this Three Estate is divided into two parts composition: Internal feedback grade rf _i,jwith the difference rc being subject to user context data impact _i,j; Wherein, Internal feedback grade depends on the feedback of user self to Web service, has nothing to do with user context data, and therefore the difference rc being subject to user context data impact is all depended in the difference of Three Estate _i,j; Respectively by middle rank and the user context data of rudimentary two kinds to user u _ithe difference definition that affects of feedback levels is Δ _{i, 1}, the user context data of senior and intermediate two kinds is to user u _ithe difference definition that affects of feedback levels is Δ _{i, 2}, so just by each user u _ithe feedback levels impact vector representation that is subject to different classes of user context data be (Δ _{i, 1}, Δ _{i, 2}).

3. method according to claim 1, is characterized in that: described step 2 comprises following content of operation:

(21) impact of different classes of user context data on user feedback grade is calculated: be respectively each user u _icalculate (Δ _{i, 1}, Δ _{i, 2}): ${\mathrm{\Δ}}_{i,1}=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈{\mathrm{WS}}_{i,1}}(r_{i,j,M}-r_{i,j,L})}{n_{i,1}},{\mathrm{\Δ}}_{i,2}=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈{\mathrm{WS}}_{i,2}}(r_{i,j,H}-r_{i,j,M})}{n_{i,2}};$ In formula, WS _{i, 1}represent user u _iused the Web service set of middle rank and rudimentary two kind user context data access, WS simultaneously _{i, 2}represent user u _iused the Web service set of middle rank and senior two kind user context data access, ws simultaneously _jset WS _{i, 1}or WS _{i, 2}in Web service, r _{i, j, L}, r _{i, j, M}, r _{i, j, H}each user u respectively _ito Web service ws when using basic, normal, high level three class users context datas _jfeedback levels, n _{i, 1}, n _{i, 2}set WS respectively _{i, 1}and WS _{i, 2}in the sum of Web service;

(22) Weakening treatment is carried out on the difference that affects of different classes of user context data: (the Δ obtained according to step (21) _{i, 1}, Δ _{i, 2}), by each user u _ibe transformed in unified reference classification user context data in the feedback levels of different classes of user context data; If when being middle rank selected with reference to user context data, then less advanced users context data feedback levels is expressed as r' via after conversion _{i, j, L}, and r' _{i, j, L}=r _{i, j, L}+ Δ _{i, 1}; Advanced level user's contextual feedback grade is expressed as r' via after conversion _{i, j, H}, and r' _{i, j, H}=r _{i, j, H}-Δ _{i, 2}; Thus by feedback levels corresponding for all different classes of user context data after Weakening treatment, be all transformed into unified reference classification, feedback levels when all corresponding to the user context data of intermediate classification: r' _{i, j, L}, r _{i, j, M}, r' _{i, j, H}.

4. method according to claim 1, is characterized in that: described step 3 comprises following content of operation:

(31) two different user u are calculated _aand u _bsimilarity time, first find the Web service set P=P that these two users accessed jointly _a∩ P _b, in formula, natural number a and b are the sequence numbers of two different users, P _a, P _bbe expressed as user u _aand u _bduring the Web service set of respective access, what obtain that these two users accessed jointly at it is total to | the Pearson correlation coefficient in P| Web service: $\mathrm{Sim}(a,b)=\frac{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}(r_{a,j}-{\stackrel{\‾}{r}}_a)(r_{b,j}-{\stackrel{\‾}{r}}_b)}{\sqrt{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}{(r_{a,j}-{\stackrel{\‾}{r}}_a)}^2}\sqrt{{\mathrm{\Σ}}_{{\mathrm{ws}}_j\∈P}{(r_{b,j}-{\stackrel{\‾}{r}}_b)}^2}},$ As the similarity between these two different users; In formula, r _a,jand r _b,jtwo user u respectively _aand u _bto Web service ws _jrespective scoring, with two user u respectively _aand u _bto each self feed back expectation value of serving in Web service set P; In formula, r _a,jby user u according to step (22) _abe transformed into reference to rank (c in the feedback levels of three kinds of different classes of user context data (L, M, H) _r) after user context data to Web service ws _jfeedback levels referred to as r _a,j.

(32) return execution step (31), solve respectively and obtain each user u _aand the user's similarity between every other user;

(33) from each user u _aand in user's similarity between other users, choose Top-k similar users, namely select user u _alarger front k similar users composition similar users S set (a) of similarity numerical value: $S\left(a\right)=\left\{b\right|\mathrm{Sim}(a,b)\≥{\mathrm{Sim}}_a^k,\mathrm{Sim}(a,b)>0,a\≠b\},$ Wherein, k is natural number, for user u _auser's similarity that the similar users that a kth numerical value is larger is corresponding;

(34) user u is obtained _asimilar users set after, according to the common used Web service collection of this k user in S set (a) calculate each user u _awith other each user u _bbetween feedback similarity $\mathrm{FSim}(a,b)=\left\{\begin{array}{c}1-\sqrt{\frac{\underset{s_j\∈S^k}{\mathrm{\Σ}}{(r_{a,j}-r_{b,j})}^2}{{10}^2\·l}},l\≠0,\\ 0,l=0\end{array}\right.$ And FSim (a, b) ∈ [0,1]; In formula, l is Web service S set ^kin service sum, r _a,jand r _b,juser u respectively _awith another user u _bto service s _jfeedback levels.

5. method according to claim 4, it is characterized in that: the similarity degree numerical value in the described feedback similarity feedback levels that to be two different users provide identical Web service, and the similarity degree numerical value of feedback levels is larger, both are more similar, otherwise then both are more dissimilar; And user's similarity is the similarity degree of different user between operation behavior.

6. method according to claim 1, is characterized in that: described step 4 comprises following content of operation:

(41) be intermediate, i.e. c according to reference user context data selected in step (22) _rduring for M, the feedback levels of the intermediate class users context data obtained i.e. r _{a, j, M}basis on, according to formula calculate this user u _awhen with reference to classification user context data, the Web service ws obtained _jcredit worthiness result in formula, c _rfor the selected a certain classification with reference to user context data, be chosen to be middle rank here; S (a) is the user u calculated in step (33) _asimilar users set, FSim (a, b) is two different user u _aand u _bfeedback similarity;

(42) according to step (41) user u _athe Web service ws obtained when being middle rank with reference to classification user context data _jcredit worthiness with this user u that step (21) obtains _a(Δ _{i, 1}, Δ _{i, 2}), calculate this user respectively at other classifications, Web service credit worthiness r namely corresponding to rudimentary and senior two class users context datas _{a, j, L}=r _{a, j, M}-Δ _{a, 1}, r _{a, j, H}=r _{a, j, M}+ Δ _{a, 2}, to obtain under different classes of user context data environment, user u _aaccess Web service ws _jthe different classification measurement results of credit worthiness.