CN103793505B - Network service collaborative filtering method based on user-service characteristics - Google Patents

Abstract

Disclosed is a network service collaborative filtering method based on user-service characteristics. The method comprises the steps that firstly, a user is selected randomly to serve as an active user and a user-service distance matrix is worked out; then, k users which are most similar to the active user are found out and k services which are most similar to a service are found out; finally, a final QoS recommended value is worked out according to the k most similar users and the k most similar services. According to the method, the inter-user similarity, the user-service distance similarity and the user-service responding time are calculated on three dimensions respectively, and therefore the defects that the concept is ambiguous and the calculation precision is low when the inter-user similarity, the user-service distance similarity and the user-service responding time are calculated together in a linear regression mode are overcome. Compared with a UMEAN algorithm and an IMEAN algorithm, the network service collaborative filtering method is obvious superior in classification precision, and the average prediction precision is improved by about twenty percent.

Description

Network service collaborative filtering method based on user-service features

Technical field

The invention belongs to data digging method technical field, it is related to a kind of association of the network service based on user-service features Same filter method.

Background technology

Collaborative filtering is the major technique of information filtering, is also the most successful technology in commending system, grinds for it Study carefully and start from the nineties in 20th century, promote the prosperity of whole commending system research.Collaborative filtering is widely used in digitized map In the individuation service system in the fields such as book shop, ecommerce.

The basic thought of collaborative filtering is：Obtain and active user c_curSimilar（Such as interest is similar with hobby）Other User c_j, calculate the value of utility u (c for user for the object s_j, s), using value of utility, all s are ranked up or weight etc. with behaviour Make, find be best suitable for cur its basic thought of object s*. be highly susceptible to understand, in daily life, we often make good use of The recommendation of friend come to carry out some selection.Collaborative filtering exactly applies to commending system this thought, that is, be based on other User recommends to targeted customer to the evaluation of a certain content.Be can be described as from user's based on the commending system of collaborative filtering Angle is recommended, and is automatic that is to say, that the recommendation that user is obtained is that system can browse from user's purchase In behavior implicit expression obtain it is not necessary to user's active go search be suitable for oneself interest recommendation information, such as fill in some application forms Lattice etc..Its another one advantage is that do not have special requirement to recommended（And content-based recommendation needs to recommended Carry out feature analysiss）, non-structured complex object can be processed, such as music, film etc..Meanwhile, the pass between research user System need substantial amounts of user access activity historical data, with community network research have cross point, have abundant Research foundation and Wide prospect.The research earliest to collaborative filtering has Grundy system, and achievement in research later includes Tapestry System, GroupLens, Ringo, PHOAKS system, Jester system etc..The main performance of QoS in network service Parameter includes availability, response time, reliability, throughput etc..These qos values all with the address of user-service both sides and Apart from closely related, and these all do not have comprehensively and reasonably consider in existing Collaborative Filtering Recommendation Algorithm.Study first That carry out QoS prediction based on collaborative filtering is Shao.They propose a kind of prediction of the collaborative filtering method based on user Qos value.Zheng et al. proposes WSRec algorithm, and this is a kind of collaborative filtering method having merged based on user with based on service Carry out recommendation network service, and large-scale experiment has been carried out based on live network service data collection.In their algorithm also Propose a kind of Pearson correlation coefficients of strengthening, it solves PCC and often over-evaluates use because calling few same services The problem of the similarity of the service that family is called.Above-mentioned both of which be only according to user call similarity between service Lai Recommended, without considering the various parameters problem in qos value, therefore, its precision is relatively low.

Content of the invention

It is an object of the invention to provide a kind of network service collaborative filtering method-USCFA based on user-service features (An user-service based web service collaborative filtering algorithm), solves existing With the presence of the relatively low problem of the qos value precision of prediction of the large scale network service of technology.

The technical scheme is that, based on the network service collaborative filtering method of user-service features, select at random first Take a user as active user, calculate user-served distance matrix；Then, find out the k the most similar to this active user Individual user, finds out the k service the most similar to this service；Finally according to this k most like user and k most like service meter Calculate final QoS recommendation.

The feature of the present invention also resides in：

The method finding out the k user the most similar to this active user is as follows：

Step one：For specified active user u, calculate Sim_L(u, u '), and be designated asCodomain is [0,1]；

Step 2：For specified active user u, for destination service, calculate Sim_D(u, u '), and be designated asCodomain It is [0,1]；

Step 3：For specified active user u, for destination service, calculate Sim_RT(u, u '), and be designated asCodomain It is [- 1,1]；

Step 4：Calculate final similarity S based on user_u, its computing formula is as follows：

Here, α₁、α₂、α₃For the combination parameter of three components,It is the similarity component based on address of theenduser,It is base Call the similarity component of service in user,It is the similarity component of the response time calling service based on user, codomain is [-1,1]；

Step 5：According to the above-mentioned Similarity value based on user, find out the user of k maximum, use as with positive Family u user u' the most similar, then the QoS predictive value of prediction user is calculated by the qos value of these users, it is designated as r_u(u, s)；

The method finding out the k service the most similar to this service is as follows：

Step one：For specified destination service s, calculate Sim_L(s, s '), and it is designated as S_i1, codomain is [0,1]；

Step 2：For specified destination service s, for calling its active user u, calculate Sim_D(s, s '), and It is designated as S_i2, codomain is [0,1]；

Step 3：For specified destination service s, for calling its active user u, calculate Sim_RT(s, s '), and It is designated as S_i3, codomain is [- 1,1]；

Step 4：Calculate final similarity S based on service_s, its computing formula is as follows：

S_s=β₁·S_i1+β₂·S_i2+β₃·S_i3(2)

Here, β₁、β₂、β₃For the combination parameter of three components, S_i1It is the similarity component based on service, S_i2It is based on mesh The similarity component of the distance to test user for the mark service, S_i3It is the response time being produced to test user based on destination service Similarity component, codomain is [- 1,1]；

Step 5：According to the above-mentioned Similarity value based on service, find out the service of k maximum, take as with target Business s service s' the most similar, then the QoS predictive value of destination service is calculated by the qos value of these services, it is designated as r_s(u, s).

Integrate the prediction of the qos value based on user and final QoS predictive value is predicted as based on the qos value of service, and to survey Family on probation recommendation network service；To r_u(u, s) and r_s(u, s) linear regression calculate, obtain final qos value predict the outcome r (u, s).

Similarity Measure based on user

I. active user u and the similarity based on address of theenduser for other users u'：

Here Dis (u, u ') is the distance to other users u' for active user u, and R is the radius of terrestrial equator circle, is abbreviated as

II. the similarity of the distance that active user u and other users u' are produced based on the same services called：

Calculate the distance of service s the normalized that user u calls to it first, obtain Dis (u, s), then calculate use The distance of service s that family u' calls to it normalized, (u ' s), then calculates similarity with following formula to obtain Dis

Here from (distance), SD refers to the set of the service that user u and user u' calls simultaneously to D span, because here It is the similarity of distance, so set identification is SD,Refer to user u to the average distance value of service s, Refer to user u' to the average distance value of service s, be abbreviated as

III. the response time similarity that active user u and other users u' are produced based on the same services called

Here RT refers to response time (response time), and SRT refers to the collection of the service that user u and user u' calls simultaneously Close, because being in response to the similarity of time here, set identification is SRT,Refer to the average qos value of user u, Refer to the average qos value of user u'.It is abbreviated as

The computational methods of the Similarity Measure based on service are as follows：

I. destination service s services the similarity based on address of service for the s' with other

Here Dis (s, s ') is the distance to other services s' for destination service s, and R is the radius of terrestrial equator circle.It is abbreviated as S_i1.

II. the similarity of the distance that destination service s is called based on same subscriber and produced with other services s'

Calculate service s first to the distance calling its user u and normalized, obtain Dis (s, u), then calculate clothes S' is to the distance calling its user u and normalized for business, obtain Dis (s ', u).Then calculate similarity with following formula

Here from (distance), UD refers to call the set of the user u of service s and service s', because here D span simultaneously It is the similarity of distance, so set identification is UD,Refer to service the average distance value of s to user u,Refer to The average distance value of service s' to user u.It is abbreviated as S_i2.III. destination service s is based on other services s' and is adjusted by same subscriber With and produce response time similarity

Here RT refers to response time (response time), and URT refers to call the collection of the user of service s and service s' simultaneously Close, because being in response to the similarity of time here, set identification is URT,Refer to service the average qos value of s,Refer to the average qos value of service s'.It is abbreviated as S_i3.

The computational methods of last QoS predictive value are as follows：

（1）Qos value prediction based on user

Here N (u) refers to the nearest k user set of user u,Refer to the average qos value of test user,Refer to it The average qos value of his user.

（2）Qos value prediction based on service

Here N (s) refers to service the nearest k set of service of s,Refer to the average qos value of service s,Refer to other clothes The average qos value of business.

Final qos value predictor formula is as follows

R (u, s)=λ r_u(u,s)+(1-λ)·r_s(u,s) (11)

Wherein the scope of λ is 0.1 to 0.5, and step value is 0.1.

The present invention has the advantages that：

Present invention employs the similarity of three types, i.e. the geographical position similarity of user and user, service and service Geographical similarity, user and service similarity geographically；The present invention is by the similarity between user, user-served distance Similarity, user-service response time are calculated respectively in three dimensions, this avoid their linear regressions together Calculate concept ambiguous, the low defect of computational accuracy.In contrast to UMEAN, IMEAN algorithm, the present invention is bright in nicety of grading Show better than other 2 contrast algorithms, consensus forecast precision improves about 20%.

Brief description

The network service collaborative filtering method flow chart based on user-service features for Fig. 1 present invention.

The network service collaborative filtering method based on user-service features for Fig. 2 present invention and other two kinds of arithmetic accuracy pair Than.

The network service collaborative filtering method based on user-service features for Fig. 3 present invention is in different combination parameter S₁、S₂、 S₃Under Comparative result.α here₁=β₁, α₂=β₂, α₃=β₃, use S₁Unified representation α₁And β₁, use S₂Unified representation α₂And β₂, use S₃ Unified representation α₃And β₃.Find that result and the selection of combination parameter affect less.

Result pair under different parameters λ for the network service collaborative filtering method based on user-service features for Fig. 4 present invention Than.

Specific embodiment

The present invention is described in detail with reference to the accompanying drawings and detailed description.

Identifier declaration in the present invention：

U={ u₁,u₂,...,u_mBe user set, m is the sum of user.

S={ s₁,s₂,...,s_nIt is the set servicing, n is the sum of service.

M={ r { u_i,s_j| 1≤i≤m, 1≤j≤n } it is user-service matrix.r{u_i,s_jIt is user u_iCall service s_j Parameter vector.

R (u) is the mean parameter that user u calls service.

R (s) is the mean parameter of service s.

R (u ') is the mean parameter that user u ' calls service.

R (s ') is the mean parameter of service s '.

U is the test user specifying.

U ' is other user.

S is the destination service specified.

S ' is other service.

K is nearest-neighbors number parameter.

Sim_L(u, u ') is active user u and the similarity based on address of theenduser for the other users u '.

Sim_D(u, u ') is similar to the distance that other users u ' is produced based on the same services called for active user u Degree.

Sim_RTThe response time that (u, u ') is produced based on the same services called with other users u ' for active user u Similarity.

r_u(u, s) is the recommendation results based on user.

Sim_L(s, s ') is that destination service s services the similarity based on address of service for the s ' with other.

Sim_D(s, s ') is similar to the distance that other services s ' is called and produced based on tested user for destination service s Degree.

Sim_RTThe response time that (s, s ') is called and produced based on tested user with other services s ' for destination service s Similarity.

r_s(u, s) is the recommendation results based on service.

R (u, s) is final recommendation results.

Define 1：Similarity between user, refers to the geographical similarity between user, computing formula such as formula（3）.

Define 2：Similarity between service, the geographical similarity between referring to service, computing formula such as formula（6）.

Define 3：User-served distance matrix.Referring to element is corresponding row（User）To respective column（Service）The square of distance value Battle array.It is expressed as M '={ d (u_i,s_j) | 1≤i≤m, 1≤j≤n }, wherein d (u_i,s_j) it is user u_i, to service s_jNormalization away from From value, for actual range divided by terrestrial equator girth half, the result after normalization is in the range of [0,1] for method for normalizing.

Referring to Fig. 1, specify active user u and destination service s first, be the recommendation network service of active user u.

（One）Similarity Measure based on user and qos value predicted portions

Step one：For specified active user u, calculate Sim_L(u, u '), and be designated asCodomain is [0,1].

Step 2：For specified active user u, for destination service, calculate Sim_D(u, u '), and be designated asCodomain It is [0,1].

Step 3：For specified active user u, for destination service, calculate Sim_RT(u, u '), and be designated asCodomain It is [- 1,1].

Step 4：Calculate final similarity S based on user_u, its computing formula is as follows：

Here, α₁、α₂、α₃For the combination parameter of three components,It is the similarity component based on address of theenduser,It is Call the similarity component of service based on user,It is the similarity component of the response time calling service based on user.Codomain For [- 1,1].

Step 5：According to the above-mentioned Similarity value based on user, find out the user of k maximum, use as with positive Family u user u' the most similar, then the QoS predictive value of prediction user is calculated by the qos value of these users, it is designated as r_u(u, s).

（Two）Similarity Measure based on service and qos value predicted portions

Step one：For specified destination service s, calculate Sim_L(s, s '), and it is designated as S_i1, codomain is [0,1].

Step 2：For specified destination service s, for calling its active user u, calculate Sim_D(s, s '), and It is designated as S_i2, codomain is [0,1].

Step 3：For specified destination service s, for calling its active user u, calculate Sim_RT(s, s '), and It is designated as S_i3, codomain is [- 1,1].

Step 4：Calculate final similarity S based on service_s, its computing formula is as follows：

S_s=β₁·S_i1+β₂·S_i2+β₃·S_i3(2)

Here, β₁、β₂、β₃For the combination parameter of three components, α in the present invention₁=β₁, α₂=β₂, α₃=β₃, use below S₁Unified representation α₁And β₁, use S₂Unified representation α₂And β₂, use S₃Unified representation α₃And β₃.S_i1It is the similarity component based on service, S_i2It is the similarity component of the distance to test user based on destination service, S_i3It is to be produced to test user based on destination service Response time similarity component.Codomain is [- 1,1].The experimental result difference of different parameter combinations is less.As Fig. 3 institute Show, different combination parameter S₁、S₂、S₃Under Comparative result, selected by discovery, combination parameter is little for the impact of result.

Step 5：According to the above-mentioned Similarity value based on service, find out the service of k maximum, take as with target Business s service s' the most similar, then the QoS predictive value of destination service is calculated by the qos value of these services, it is designated as r_s(u, s).

（Three）Integrate based on user qos value prediction and based on service qos value be predicted as final QoS predictive value and to Test user's recommendation network service

To r_u(u, s) and r_s(u, s) linear regression calculates, and obtains final recommendation results r (u, s).

Study now following problem：（1）How specifically to carry out the meter based on user's similarity with based on service similarity Calculate；（2）How specifically to carry out based on user with based on service qos value prediction；（3）How to determine various combination parameters.

（Four）Similarity Measure based on user

I. active user u and the similarity based on address of theenduser for other users u'

Here Dis (u, u ') is the distance to other users u' for active user u, and R is the radius of terrestrial equator circle.

II. the similarity of the distance that active user u and other users u' are produced based on the same services called

Calculate the distance of service s the normalized that user u calls to it first, obtain Dis (u, s), then calculate use The distance of service s that family u' calls to it normalized, obtain Dis (u ', s).Then calculate similarity with following formula

Here from (distance), SD refers to the set of the service that user u and user u' calls simultaneously to D span, because here It is the similarity of distance, so set identification is SD,Refer to user u to the average distance value of service s,Refer to User u' is to the average distance value of service s.

III. the response time similarity that active user u and other users u' are produced based on the same services called

Here RT refers to response time (response time), and SRT refers to the collection of the service that user u and user u' calls simultaneously Close, because being in response to the similarity of time here, set identification is SRT,Refer to the average qos value of user u,Refer to the average qos value of user u'.

（Five）Similarity Measure based on service

I. destination service s services the similarity based on address of service for the s' with other

Here Dis (s, s ') is the distance to other services s' for destination service s, and R is the radius of terrestrial equator circle.

II. the similarity of the distance that destination service s is called based on same subscriber and produced with other services s'

Calculate service s first to the distance calling its user u and normalized, obtain Dis (s, u), then calculate clothes S' is to the distance calling its user u and normalized for business, obtain Dis (s ', u).Then calculate similarity with following formula

Here from (distance), UD refers to call the set of the user u of service s and service s', because here D span simultaneously It is the similarity of distance, so set identification is UD,Refer to service the average distance value of s to user u,Refer to The average distance value of service s' to user u.

III. the response time similarity that destination service s is produced based on being called by same subscriber with other services s'

Here RT refers to response time (response time), and URT refers to call the collection of the user of service s and service s' simultaneously Close, because being in response to the similarity of time here, set identification is URT,Refer to service the average qos value of s,Refer to the average qos value of service s'.

Similarity Measure completes, and task below is to calculate last QoS predictive value.

（1）Qos value prediction based on user

Here N (u) refers to the nearest k user set of user u,Refer to the average qos value of test user,Refer to it The average qos value of his user.

（2）Qos value prediction based on service

Here N (s) refers to service the nearest k set of service of s,Refer to the average qos value of service s,Refer to other clothes The average qos value of business.

Final qos value predict the outcome for

R (u, s)=λ r_u(u,s)+(1-λ)·r_s(u,s) (11)

Wherein the scope of λ is 0.1 to 0.5, and step value is 0.1.When the number of users for prediction is less, the impact of λ Larger.

Fig. 2 is the network service collaborative filtering method based on user-service features for the present invention and other two kinds of arithmetic accuracy Contrast, given parameters are：S₁=0.5, S₂=0.25, S₃=0.25, λ=0.4.Because the present invention carries out similarity meter on three dimensionality Calculate, it is to avoid only with user's meansigma methodss and service mean value calculation precision low defect.So contrast UMEAN, IMEAN calculate Method, average error value of the present invention is relatively low, and predictablity rate improves.

Fig. 4 is result under different parameters λ for the network service collaborative filtering method based on user-service features for the present invention Contrast, given parameters are：S₁=0.5, S₂=0.25, S₃=0.25.λ value gets over trend 0.5, and average error value is less.

Claims (5)

1. the network service collaborative filtering method based on user-service features is made it is characterised in that randomly selecting a user first For active user, calculate user-served distance matrix；Then, find out the k user the most similar to this active user, look for Go out the k service the most similar to destination service；Calculate finally according to this k most like user and k most like service Whole QoS recommendation；

The method finding out the k user the most similar to this active user is as follows：

Step one：For specified active user u, calculate Sim_L(u, u '), and be designated asCodomain is [0,1], and wherein u ' is it He is user；

Step 2：For specified active user u, for destination service, calculate Sim_D(u, u '), and be designated asCodomain be [0, 1]；

Step 3：For specified active user u, for destination service, calculate Sim_RT(u, u '), and be designated asCodomain be [- 1,1]；

Step 4：(u' u), and is designated as S to final similarity Sim based on user for the calculating_u, its computing formula is as follows：

$S_u={\mathrm{\α}}_1\·S_{u_1}+{\mathrm{\α}}_2\·S_{u_2}+{\mathrm{\α}}_3\·s_{u_3}---\left(1\right)$

Here, α₁、α₂、α₃For the combination parameter of three components,It is the similarity component based on address of theenduser,It is based on use The similarity component of service is called at family,The similarity component of the response time calling service based on user, codomain be [- 1, 1]；

Step 5：According to the above-mentioned Similarity value based on user, find out the user of k maximum, as with active user u For similar user u_s', then the QoS predictive value predicting user is calculated by the qos value of these users, it is designated as r_u(u,s)；

The method finding out the k service the most similar to destination service is as follows：

Step one：For specified destination service s, calculate Sim_L(s, s '), and it is designated as S_i1, codomain is [0,1], and wherein s ' is it He services；

Step 2：For specified destination service s, for calling its active user u, calculate Sim_D(s, s '), and be designated as S_i2, codomain is [0,1]；

Step 3：For specified destination service s, for calling its active user u, calculate Sim_RT(s, s '), and be designated as S_i3, codomain is [- 1,1]；

Step 4：(s' s), and is designated as S to final similarity Sim based on service for the calculating_s, its computing formula is as follows：

S_s=β₁·S_i1+β₂·S_i2+β₃·S_i3(2)

Here, β₁、β₂、β₃For the combination parameter of three components, α in the present invention₁=β₁, α₂=β₂, α₃=β₃, use S below₁System One expression α₁And β₁, use S₂Unified representation α₂And β₂, use S₃Unified representation α₃And β₃；S_i1It is the similarity component based on service, S_i2 It is the similarity component of the distance to test user based on destination service, S_i3It is the sound being produced to test user based on destination service Similarity component between seasonable, codomain is [- 1,1]；

Step 5：According to the above-mentioned Similarity value based on service, find out the service of k maximum, as with destination service s For similar service s_s', then the QoS predictive value of destination service is calculated by the qos value of these services, it is designated as r_s(u,s).

2. the network service collaborative filtering method based on user-service features as claimed in claim 1 is it is characterised in that base Similarity Measure in user

I. active user u and the similarity based on address of theenduser for other users u'：

${\mathrm{Sim}}_L(u,u^{\′})=1-\frac{Dis(u,u^{\′})}{\mathrm{\π}\·R}---\left(3\right)$

Here Dis (u, u ') is the distance to other users u' for active user u, and R is the radius of terrestrial equator circle, is abbreviated as

II. the similarity of the distance that active user u and other users u' are produced based on the same services called：

Calculate the distance of service s the normalized that user u calls to it first, obtain Dis (u, s), then calculate user u' The distance of service s called to it normalized, (u ' s), then calculates similarity with following formula to obtain Dis

${\mathrm{Sim}}_D(u,u^{\′})=\frac{\underset{s\∈SD}{\Σ}\left(Dis\right(u,s)-\stackrel{\‾}{Dis}(u\left)\right)\·\left(Dis\right(u^{\′},s)-\stackrel{\‾}{Dis}(u^{\′}\left)\right)}{\sqrt{\underset{s\∈SD}{\Σ}{\left(Dis\right(u,s)-\stackrel{\‾}{Dis}(u\left)\right)}^2}\·\sqrt{\underset{s\∈SD}{\Σ}{\left(Dis\right(u^{\′},s)-\stackrel{\‾}{Dis}(u^{\′}\left)\right)}^2}}---\left(4\right)$

Here from (distance), SD refers to the set of the service that user u and user u' calls simultaneously to D span because be here away from From similarity, so set identification be SD,Refer to user u to the average distance value of service s,Refer to user U', to the average distance value of service s, is abbreviated as

III. the response time similarity that active user u and other users u' are produced based on the same services called

${\mathrm{Sim}}_{RT}(u,u^{\′})=\frac{\underset{s\∈SRT}{\Σ}\left(r\right(u,s)-\stackrel{\‾}{r}(u\left)\right)\·\left(r\right(u^{\′},s)-\stackrel{\‾}{r}(u^{\′}\left)\right)}{\sqrt{\underset{s\∈SRT}{\Σ}{\left(r\right(u,s)-\stackrel{\‾}{r}(u\left)\right)}^2}\·\sqrt{\underset{s\∈SRT}{\Σ}{\left(r\right(u^{\′},s)-\stackrel{\‾}{r}(u^{\′}\left)\right)}^2}}---\left(5\right)$

Here RT refers to response time (response time), and SRT refers to the set of the service that user u and user u' calls simultaneously, because For being in response to the similarity of time here, so set identification is SRT,Refer to the average qos value of user u,Refer to use The average qos value of family u', r (u, s) refers to the qos value for service s for the user u, and (u' s) refers to other users u' for service to r The qos value of s, is abbreviated as

3. the network service collaborative filtering method based on user-service features as claimed in claim 1 is it is characterised in that base As follows in the computational methods of the Similarity Measure of service：

I. destination service s services the similarity based on address of service for the s' with other

${\mathrm{Sim}}_L(s,s^{\′})=1-\frac{Dis(s,s^{\′})}{\mathrm{\π}\·R}---\left(6\right)$

Here Dis (s, s ') is the distance to other services s' for destination service s, and R is the radius of terrestrial equator circle, is abbreviated as S_i1；

II. the similarity of the distance that destination service s is called based on same subscriber and produced with other services s'

Calculate service s first to the distance calling its user u and normalized, obtain Dis (s, u), then calculate and service s' To the distance calling its user u and normalized, obtain Dis (s ', u)；Then calculate similarity with following formula

${\mathrm{Sim}}_D(s,s^{\′})=\frac{\underset{u\∈UD}{\Σ}\left(Dis\right(s,u)-\stackrel{\‾}{Dis}(s\left)\right)\·\left(Dis\right(s^{\′},u)-\stackrel{\‾}{Dis}(s^{\′}\left)\right)}{\sqrt{\underset{u\∈UD}{\Σ}{\left(Dis\right(s,u)-\stackrel{\‾}{Dis}(s\left)\right)}^2}\·\sqrt{\underset{u\∈UD}{\Σ}{\left(Dis\right(s^{\′},u)-\stackrel{\‾}{Dis}(s^{\′}\left)\right)}^2}}---\left(7\right)$

Here D span is from (distance), and UD refers to call the set of the user u of service s and service s' simultaneously because be here away from From similarity, so set identification be UD,Refer to service the average distance value of s to user u,Refer to service The average distance value of s' to user u, is abbreviated as S_i2；

III. the response time similarity that destination service s is produced based on being called by same subscriber with other services s'

${\mathrm{Sim}}_{RT}(s,s^{\′})=\frac{\underset{u\∈URT}{\Σ}\left(r\right(s,u)-\stackrel{\‾}{r}(s\left)\right)\·\left(r\right(s^{\′},u)-\stackrel{\‾}{r}(s^{\′}\left)\right)}{\sqrt{\underset{u\∈URT}{\Σ}{\left(r\right(s,u)-\stackrel{\‾}{r}(s\left)\right)}^2}\·\sqrt{\underset{u\∈URT}{\Σ}{\left(r\right(s^{\′},u)-\stackrel{\‾}{r}(s^{\′}\left)\right)}^2}}---\left(6\right)$

Here RT refers to response time (response time), and URT refers to call the set of the user of service s and service s' simultaneously, because For being in response to the similarity of time here, so set identification is URT,Refer to service the average qos value of s,Refer to clothes The average qos value of business s', r (s, u) refers to service the qos value for user u for the s, and (s' u) refers to other services s' for user to r The qos value of u, is abbreviated as S_i3.

4. the network service collaborative filtering method based on user-service features as claimed in claim 1 is it is characterised in that The computational methods of QoS predictive value afterwards are as follows：

(1) qos value based on user is predicted

$r_u(u,s)=\stackrel{\‾}{r}\left(u\right)+\frac{\underset{u^{\′}\∈N\left(u\right)}{\Σ}Sim(u^{\′},u)\·\left(r\right(u^{\′},u)-\stackrel{\‾}{r}(u^{\′}\left)\right)}{\underset{u^{\′}\∈N\left(u\right)}{\Σ}Sim(u^{\′},u)}---\left(9\right)$

Here N (u) refers to the nearest k user set of user u,Refer to the average qos value of test user,Refer to other users Average qos value, r (u', s) refer to other users u' for service s qos value；

(2) the qos value prediction based on service

$r_s(u,s)=\stackrel{\‾}{r}\left(s\right)+\frac{\underset{s^{\′}\∈N\left(s\right)}{\Σ}Sim(s^{\′},s)\·\left(r\right(s^{\′},s)-\stackrel{\‾}{r}(s^{\′}\left)\right)}{\underset{s^{\′}\∈N\left(s\right)}{\Σ}Sim(s^{\′},s)}---\left(10\right)$

Here N (s) refers to service the nearest k set of service of s,Refer to the average qos value of service s,Refer to other services Average qos value, (s' s) refers to the qos value for destination service for other services to r.

5. the network service collaborative filtering method based on user-service features as claimed in claim 1 is it is characterised in that to r_u (u, s) and r_s(u, s) linear regression calculate, obtain the i.e. final qos value of final recommendation results r (u, s) predict the outcome for

R (u, s)=λ r_u(u,s)+(1-λ)·r_s(u,s) (11)

Wherein the scope of λ is 0.1 to 0.5, and step value is 0.1, r_u(u, s) is the QoS predictive value based on user, r_s(u, s) is base QoS predictive value in service.