CN107464032A - A kind of online service measures of reputation method based on KendallTau distances - Google Patents

A kind of online service measures of reputation method based on KendallTau distances Download PDF

Info

Publication number
CN107464032A
CN107464032A CN201710484114.1A CN201710484114A CN107464032A CN 107464032 A CN107464032 A CN 107464032A CN 201710484114 A CN201710484114 A CN 201710484114A CN 107464032 A CN107464032 A CN 107464032A
Authority
CN
China
Prior art keywords
msub
service
mrow
reputation
user
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201710484114.1A
Other languages
Chinese (zh)
Inventor
付晓东
郑苏苏
刘骊
冯勇
刘利军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN201710484114.1A priority Critical patent/CN107464032A/en
Publication of CN107464032A publication Critical patent/CN107464032A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/01—Customer relationship services

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • Theoretical Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The present invention relates to a kind of online service measures of reputation method based on KendallTau distances, belong to online measures of reputation and group decision-making field.The present invention weighs the uniformity between two service scoring vectors by range index first；Then measures of reputation is modeled as an optimization problem, this optimization problem is exactly to find a reputation vectors minimum with user service rating matrix distance；Finally solves the optimization problem using simulated annealing, using obtained reputation vectors as services reputation.The present invention considers the inconsistency of user's interpretational criteria, efficiently avoid online service prestige and does not possess the problem of comparability；In addition, finding optimal reputation vectors as services reputation using simulated annealing, the efficiency of measures of reputation can effectively ensure that；Meanwhile by considering relation of the user to different service scorings, improve the anti-maneuverability of measures of reputation method.

Description

A kind of online service measures of reputation method based on KendallTau distances
Technical field
The present invention relates to a kind of online service measures of reputation method based on KendallTau distances, belong to online credit worthiness Amount and group decision-making technical field.
Background technology
With the development and popularization of internet, traditional business environment is just being converted into open, shared, polynary towards online The new environment of service.More and more important effect is played in online service in fields such as ecommerce, enterprise operation, management, interconnection The online online service for being available for user to select is also more and more, meanwhile, user is also required to take more time with energy to seek Look for oneself desired service.First, huge online service quantity causes user can not possibly produce with each online service to interact, User can not possibly obtain the complete information of each online service.Secondly, under network environment, due to allowing anonymous interaction and each other it Between again be not directly contacted with, certain user or online service supplier may provide false evaluation information.Therefore, user needs More preferable service recommendation is provided the user by the prestige of the online service formed based on third party's viewpoint.Prestige is service The result of some behavior of credit accumulations, plays an important role for online service selection.One objective online service prestige Measure can effectively aid in the good and bad trade-off decision of user's progress online service.
Current online reputation model is roughly divided into following a few classes：Method of average model, summation model, based on Bayesian network The reputation model of network, discrete reputation model, the reputation model based on fuzzy logic, the reputation model based on chain and based on card Reputation model according to theory etc..In recent years, research of the domestic and foreign scholars to online service prestige obtains positive progress, Haiteng(<Scientific World Journal>, 2014,2014 (1)：145-156) propose that one kind is based on service quality The Web service measures of reputation model of similarity, by comparing the similitude of service quality bulletin value and actual value, to carry out Web Services reputation is assessed.Zhang Xin Zhou(<IEEE International Conference on Services Computing>, 2016,211-218) and a kind of KMRW model based on consumer's recessiveness behavior is proposed to overcome grading is rare to ask Topic, and solution method is proposed to cold start-up problem.Tingwei Chen(<International Journal of Intelligent Information Systems>, 2015,4 (1)：A kind of evaluation method based on cloud 8-15) is proposed, is led to The historical behavior of evaluation services consumer is crossed, and considers scoring similarity to generate grading quality cloud, and then utilizes cloud model Parameter carries out services reputation assessment.Hossein Shad Manaman(<Computers in Human Behavior 54>, 2016,94-100) a kind of Reputation Management System is proposed, social media data are analyzed based on N metagrammars learning method, To measure the reputation of given company.However, Audun(<International Conference on Modeling Decisions for Artificial Intelligence>, 2013,8234：126-138) point out these known methods all That the one-dimensional evaluating data submitted according to user calculates the credit worthiness of service, and assume all users according to identical standard come Evaluation service, and do not consider user because interpretational criteria is inconsistent caused by subjective preferences the problem of.Due to being accustomed to by itself, The influence of the experience of life and consumption history, the standard that user is evaluated online service can not possibly be identical, some users tendency In to higher evaluation is serviced, some users tend to relatively low evaluation is serviced, even if causing essence performance identical service Also different evaluations can be obtained.Therefore, the obtained prestige result calculated according to known method does not possess just comparability. For example the method for average is worth to its services reputation by calculating all users to being averaged for service scoring, and due to interpretational criteria not Score and not may compare with the service caused between different user, the prestige result being calculated according to the method for average does not possess comparable yet One sided result can be produced compared with property, thus when carrying out services selection using this prestige, so as to being misled to user.This Outside, the anti-maneuvering capability of known method is weaker, for example summation obtains its service by cumulative scoring of all users to service Prestige, therefore only need repeatedly to award high marks to same service (or low point), it is possible to reach the purpose for manipulating services reputation.
The present invention considers the inconsistency of user's interpretational criteria, based on scoring of the user to service, proposes to be based on The computational methods of uniformity between Kendalltau range indexs measurement service scoring vector, will be with user-service rating matrix The minimum reputation vectors of distance efficiently avoid the incomparable problem of online service prestige as final services reputation. Meanwhile optimal reputation vectors are found using simulated annealing, it ensure that the efficiency of measures of reputation；In addition, by considering user The anti-maneuvering capability of prestige is improved to the relation of different service scorings, provided for the measures of reputation and its related application of online service A kind of new technical method.
The content of the invention
The invention provides a kind of online service measures of reputation method based on KendallTau distances, for solving user Services reputation caused by inconsistent to service evaluation criterion does not possess the problem of comparability, and overcomes known method measurement Efficiency is low and the problem of anti-maneuvering capability is weak.
The technical scheme is that：A kind of online service measures of reputation method based on KendallTau distances, first Uniformity between two service scoring vectors is weighed by range index；Then measures of reputation is modeled as an optimization Problem, this optimization problem are exactly to find one and user-minimum reputation vectors of service rating matrix distance；Finally use Simulated annealing solves the optimization problem, using obtained reputation vectors as services reputation.
Methods described comprises the following steps that：
Step 1, by range index come weigh two service scoring vector between uniformity；
1.1 set U={ u1,u2,...,uxIt is to have the set of x user, S={ s1,s2,...,syIt is to have y service Gather, user is represented with x × y matrix R the score information of service, the element r in RusRepresent user u to servicing s Scoring；
1.2 pass through range index K (ri,rj) two vectorial r of service scoring of measurementi=(ri1,...,rip,...,riy) and rj =(rj1,...,rjq,...,rjy) between uniformity：
For two different service sp,sq∈ S (p ≠ q), calculate two vectorial r of service scoringi=(ri1,..., rip,...,riy) and rj=(rj1,...,rjq,...,rjy) between Kendall tau distances：
Wherein, rip(i=1,2 ..., x, p=1,2 ..., y) it is i-th of user uiTo p-th of service spScoring, table Show user uiTo servicing spSatisfaction, ripIt is more big, represent uiTo spIt is more satisfied；rip＞ riqRepresent user uiThink to service sp Better than sq, rip＜ riqRepresent user uiThink to service sqBetter than sp, rip=riqRepresent user uiThink to service spAnd sqIndifference, rip=0 represents user uiIt is not given to service spScoring or user did not produce with the service and interacted, at this point for service sp,sq∈ S (p ≠ q), do not calculate the vectorial r of service scoringiAnd rjThe distance between；
For set of service S={ s1,s2,...,sy, calculate two vectorial r of service scoringi=(ri1,...,rip,..., riy) and rj=(rj1,...,rjq,...,rjyThe distance between)：
Wherein K (ri,rj) represent riAnd rjBetween preference uniformity, K (ri,rj) smaller, represent riAnd rjBetween preference Uniformity is bigger；K(ri,rj) bigger, represent riAnd rjBetween preference uniformity it is smaller；K(ri,rj)=0 represents riAnd rjBetween Preference it is completely the same；
Step 2 and then measures of reputation is modeled as an optimization problem, this optimization problem be exactly find one with User-minimum the reputation vectors of service rating matrix distance；
R in 2.1 counting users-service rating matrix Rip＞ riq, rip=riq, rip＜ riqService to quantity, and respectively With N (rip＞ riq), N (rip=riq), N (rip<riq) represent；
2.2 use a reputation vectors rz=(rz1,rz2...,rzy) represent set of service S={ s1,s2,...,sy Prestige, based on K (r in formula (1) and formula (2)i,rj) definition, with reference to the r counted in step 2.1ip＞ riq, rip=riq, rip＜ riqService to quantity, calculate reputation vectors rzThe distance between user-service rating matrix R；
For two different service sp,sq∈ S (p ≠ q), calculate reputation vectors rzWith user-service rating matrix R it Between distance：
For set of service S={ s1,s2,...,sy, rzThe distance between user-service rating matrix R is：
Wherein K (rz, R) and represent reputation vectors rzPreference uniformity between user-service rating matrix R, K (rz,R) It is smaller, represent rzPreference uniformity between R is bigger；K(rz, R) and it is bigger, represent rzPreference uniformity between R is smaller；K (rz, R)=0 represent rzPreference between R is completely the same.
2.3 determine that measures of reputation optimization objective function is as follows：
In formula (5)：rbFor object function f (rb) one may solution, rb=(rb1,rb2...,rby), wherein rb1, rb2...,rbyRepresent set of service S={ s1,s2,...,syPrestige；Reputation vectors r in step 2.2zIt is object function f (rb) one may solution, all possible solution vector constitutes the solution space M of optimization objective function；
Step 3, using the optimization problem in simulated annealing settlement steps to deal 2, using obtained reputation vectors as clothes Business prestige；
The parameter of 3.1 setting simulated annealings, including initial temperature t0, final temperature te, move back warm factor alpha, markov Chain length L, initial parameter, which is chosen, uses following empirical values：t0=100, te=1, α=0.9, L=100, in calculating process, order terminates Condition is：New explanation under temperature t in continuous iterative process several times is not all received or temperature is reduced to final temperature, meets it Then algorithm terminates middle either condition；
3.2 setting initial solution r0, a possible solution is randomly selected from the solution space M of step 2.3 and is used as initial solution r0, r0 =(r01,r02...,r0y), calculate r0Target function value K (r0,R)；
3.3 random change initial solution r0In Partial Elements value (as change r02And r04Value), produce one it is empty positioned at solution Between M new explanation rn, rn=(rn1,rn2...,rny), calculate rnTarget function value K (rn,R)；
3.4 calculate rnObject function K (rn, R) and r0Object function K (r0, R) difference DELTA k, Δ k=K (rn,R)-K (r0,R)；
The acceptance probability of 3.5 new explanations is：
In formula (6)：T is Current Temperatures, the temperature T=α t after cooling
The receiving of new explanation follows Metropolis criterions：As Δ k ＜ 0, receive rnAs current optimal solution, when Δ k >=0 When, the random number β on (0, a 1) section is provided, in the acceptance probability P ＞ β of new explanation, receives rnAs current optimal solution, Otherwise r is not receivedn, now initial solution r0Realize an iteration；L iteration is carried out altogether under Current Temperatures, if in iterative process New solution is received, then is unsatisfactory for end condition, temperature is reduced according to T=α t；
3.6 make Current Temperatures t=T, repeat step 3.3 to 3.5, until meeting end condition, now stop calculating, try to achieve Current solutionFor optimal solution；
3.7 optimal solutions that will be obtainedServices reputation as user-service rating matrix R.
Validation verification is carried out to measures of reputation result to assess using following aspects：
Comparability：The present invention is found with all user preferences with maximum consistent based on Kendall tau range indexs The reputation vectors of property meet the preference of majority as services reputation.Because interpretational criteria is inconsistent, different user is commented Divide and do not possess comparability, so as to which the prestige result obtained according to the score calculation of different user does not also possess comparability.And The scoring of same user is comparable, such as user upTo servicing siScoring ripMore than to servicing sjScoring rjp, can be with Infer user upThink to service siBetter than sj.The present invention consider user to it is different service scoring relations, so as to get online service Prestige result possesses comparability.
High efficiency：The present invention finds optimal reputation vectors using simulated annealing, moves back initial temperature t during temperature0, eventually Only temperature teAnd warm formula T=α t setting is moved back so that algorithm terminates in finite time, and by right in searching process Current solution carries out simple transformation and produces new explanation, reduces algorithm and takes；Meanwhile simulated annealing has generally in searching process Rate kick, in addition to it can receive optimization solution, also connect in a limited degree with a random acceptance criterion (Metropolis criterions) By poor solution, and the probability for receiving poor solution is slowly intended to zero so that avoids falling into local optimum during algorithm performs Trap in.Therefore the optimization time can be shortened while low optimization accuracy is not reduced using simulated annealing, it is ensured that this hair The high efficiency of bright measures of reputation method.
Anti- maneuverability：Give user upThe service si not scored increases scoring (or lower assessment point) r one highip, then according to this hair Method in bright, according to formula (4), cause upDistance between si and other services is changed, upIt is inclined between other services Good uniformity also changes therewith.Therefore only by servicing siHigh scoring (or lower assessment point) is given, more than improve (or drop It is low) service siPrestige, the prestige of (or reduce) other online services can be also improved, so as to improve the difficulty of prestige manipulation Degree.
The beneficial effects of the invention are as follows：
1. due to being influenceed by itself custom, the experience of life and consumption history, the standard that user is evaluated online service Can not possibly be identical.In the case of user's interpretational criteria is inconsistent, scoring of the different user to same service has different Implication, so as to the services reputation for obtaining different user scoring using known methods such as cumulative or averaging methods do not possess it is comparable Property.And the present invention finds the reputation vectors for having most homogeneous with all user preferences based on Kendall tau range indexs As services reputation, the inconsistent influence of different user interpretational criteria have ignored so that the services reputation that the present invention obtains possesses Comparability.
2. it is a np hard problem to find optimal reputation vectors, solution space scale | M | with online service quantity y increase into Index increase, optimal solution is such as found by traveling through all possibility solutions using exhaustive optimization method, amount of calculation is huge, and efficiency is too It is low.The present invention finds optimal reputation vectors using simulated annealing, can significantly not reduced while the optimization time is shortened Low optimization accuracy, obtains approximate optimal solution in polynomial time, objectively ensure that the efficient of measures of reputation method of the invention Property.
3. known method when calculating the credit worthiness of a certain service, only considers scoring of the user to the service.It is such as flat Equal method is worth to the services reputation by calculating all users to being averaged for certain service scoring, thus need to only improve special services Scoring can improves the prestige of the service, so as to realize that prestige manipulates.The present invention considers scoring of the user to different services Between relation, the change of single service scoring can not only influence the prestige of the service, can also be to the prestige generation of other services Influence.User only repeatedly gives high scoring to same service, while other services are given with low scoring, can just make manipulation The prestige of service is improved.Therefore, method known to the services reputation measurement method ratio in the present invention more is difficult to manipulate.
In a word, the online service measures of reputation method based on KendallTau distances is led for user's interpretational criteria is inconsistent The incomparable problem of user service prestige of cause, optimal reputation vectors are found as services reputation using simulated annealing, The efficiency of online service measures of reputation is ensure that while the services reputation most consistent with user's scoring is obtained.Meanwhile consider User provides one kind to relation between the scoring of different services to improve the anti-maneuverability of the measures of reputation of online service New thinking.
Brief description of the drawings
Fig. 1 is method flow diagram in the present invention；
Fig. 2 is simulated annealing flow chart in step 3.
Embodiment
Embodiment 1：As shown in Figure 1-2, a kind of online service measures of reputation method based on KendallTau distances, first Uniformity between two service scoring vectors is weighed by range index；Then measures of reputation is modeled as an optimization Problem, this optimization problem are exactly to find one and user-minimum reputation vectors of service rating matrix distance；Finally use Simulated annealing solves the optimization problem, using obtained reputation vectors as services reputation.
1st, the uniformity between two service scoring vectors is weighed by range index；
1.1 give scoring of 5 users to 5 online services, and user's collection is combined into U={ u1,u2,u3,u4,u5, services set It is combined into S={ s1,s2,s3,s4,s5, user-service rating matrix R is as shown in table 1, and wherein user-service scoring represents user couple Service performance satisfaction, using in ecommerce evaluation mechanism commonly use 5 grades, 1-5 levels represent respectively it is very dissatisfied, Dissatisfied, general, satisfied and feel quite pleased, numeral 0 represents that user is not given to the service evaluation or user and the service and do not had Produced interactive.
Table 1
rip s1 s2 s3 s4 s5
u1 4 0 2 0 1
u2 2 1 2 3 3
u3 5 1 5 4 0
u4 0 2 3 0 5
u5 4 5 0 0 2
1.2 pass through range index K (ri,rj) two vectorial r of service scoring of measurementi=(ri1,...,rip,...,ri5) and rj =(rj1,...,rjq,...,rj5) between uniformity：For example calculate two vectorial r of service scoring1,r2Between distance K (r1, r2)：
For two different service sp,sq∈ S (p ≠ q), calculate r1,r2Two service the distance between scoring vectorAs shown in table 2
Table 2
For set of service S={ s1,s2,...,s5, calculate r1,r2Two service the distance between scoring vector：
Therefore, r1,r2It is 3 that two, which service the distance between scoring vector,.
2 and then measures of reputation is modeled as an optimization problem, this optimization problem is exactly to find one with using Family-minimum the reputation vectors of service rating matrix distance；
R in 2.1 counting users-service rating matrix Rip＞ riq, rip=riq, rip＜ riqService to quantity, and respectively With N (rip＞ riq), N (rip=riq), N (rip<riq) represent；
N(ri1＞ ri2)=2, N (ri1=ri2)=0, N (ri1<ri2)=1；
N(ri1＞ ri3)=1, N (ri1=ri3)=2, N (ri1＜ ri3)=0；
N(ri1＞ ri4)=1, N (ri1=ri4)=0, N (ri1＜ ri4)=1；
N(ri1＞ ri5)=2, N (ri1=ri5)=0, N (ri1＜ ri5)=1；
N(ri2＞ ri3)=0, N (ri2=ri3)=0, N (ri2＜ ri3)=3；
N(ri2＞ ri4)=0, N (ri2=ri4)=0, N (ri2＜ ri4)=2；
N(ri2＞ ri5)=1, N (ri2=ri5)=0, N (ri2＜ ri5)=2；
N(ri3＞ ri4)=1, N (ri3=ri4)=0, N (ri3＜ ri4)=1；
N(ri3＞ ri5)=1, N (ri3=ri5)=0, N (ri3＜ ri5)=2；
N(ri4＞ ri5)=0, N (ri4=ri5)=1, N (ri4＜ ri5)=0.
2.2 use a reputation vectors rz=(rz1,rz2...,rz5) represent set of service S={ s1,s2,...,s5 Prestige, based on K (r in formula (1) and formula (2)i,rj) definition, with reference to the r counted in step 2.1ip＞ riq, rip=riq, rip＜ riqService to quantity, calculate reputation vectors rzThe distance between user-service rating matrix R.Such as rz=(2,1, 4,1,5) K (r, are calculatedz,R)
rz=(2, Isosorbide-5-Nitrae, 1,5) in,
rz1＞ rz2, therefore
rz1＜ rz3, therefore
rz1＞ rz4, therefore
rz1＜ rz5, therefore
rz2＜ rz3, therefore
rz2=rz4, therefore
rz2＜ rz5, therefore
rz3＞ rz4, therefore
rz3＜ rz5, therefore
rz4＜ rz5, therefore
Therefore, reputation vectors rzThe distance between user-service rating matrix R is 13.
2.3 determine that measures of reputation optimization objective function is as follows：
In formula (5)：rbFor object function f (rb) one may solution, rb=(rb1,rb2...,rb5), wherein rb1, rb2...,rbyRepresent set of service S={ s1,s2,...,s5Prestige.Reputation vectors r in step 2.2zIt is object function f (rb) one may solution, all possible solution vector constitutes the solution space M of optimization objective function.
3rd, using the optimization problem in simulated annealing settlement steps to deal 2, believe obtained reputation vectors as service Reputation
The initial parameter of 3.1 setting simulated annealings, including initial temperature t0=100, final temperature te=1, move back warm system Number α=0.9, markov chain length L=100, in calculating process, the end condition is made to be：Under temperature t in continuous 40 iterative process New explanation is not all received or temperature is reduced to final temperature, meets that then algorithm terminates any of which condition；
3.2 setting initial solution r0, randomly select in the solution space M set from step 2.3 one may solution as initial Solve r0, r0=(3,2,2,4,5), calculate r0Target function value K (r0,R)
r0In=(3,2,2,4,5),
r01＞ r02, therefore
r01＞ r03, therefore
r01＜ r04, therefore
r01＜ r05, therefore
r02=r03, therefore
r02＜ r04, therefore
r02＜ r05, therefore
r03＜ r04, therefore
r03＜ r05, therefore
r04＜ r05, therefore
3.3 random change initial solution r0In Partial Elements value (as change r02And r04Value), produce one it is empty positioned at solution Between M new explanation rn, rn=(3,1,2,3,5), calculate rnTarget function value K (rn,R)
rnIn=(3,1,2,3,5),
rn1＞ rn2, therefore
rn1＞ rn3, therefore
rn1=rn4, therefore
rn1＜ rn5, therefore
rn2＜ rn3, therefore
rn2＜ rn4, therefore
rn2＜ rn5, therefore
rn3＜ rn4, therefore
rn3＜ rn5, therefore
rn4＜ rn5, therefore
3.4 calculate rnObject function K (rn, R) and r0Object function K (r0, R) difference DELTA k,；
K(r0, R) and=13, K (rn, R)=11, Δ k=K (rn,R)-K(r0, R)=- 2
The acceptance probability of 3.5 new explanations is：
In formula (6)：T is Current Temperatures, the temperature T=0.9t after cooling
According to Metropolis criterions, now Δ k ＜ 0, receive rnAs current optimal solution, if Δ k >=0, one is provided Random number β on (0,1) section, in the acceptance probability P ＞ β of new explanation, receive rnAs current optimal solution, otherwise do not receive rn, Now initial solution r0Realize an iteration.100 iteration are carried out altogether under Current Temperatures, if being received in iterative process new Solution, is unsatisfactory for end condition, then reduces temperature according to T=0.9t；
3.6 make Current Temperatures t=T, repeat step 3.3 to 3.5, until meeting end condition, now stop calculating, try to achieve Current solutionFor optimal solution.
3.7 users-service rating matrix R services reputation is
Efficiency assessment is carried out to measures of reputation result：
Comparability：5 are shared in solution space M5Kind may solve, whereinIt is minimum with user-service rating matrix R distance, There is most homogeneous with the preference of all users, therefore willMeet the preference of majority as services reputation.In addition, according to R in table 111＞ r13＞ r15It necessarily may infer that user u1Think to service s1Better than s3, s1Better than s5, s3Better than s5, for user u1, its scoring must be comparable, and similarly, scoring of the other users to different services is also comparable, thus is obtained Services reputationPossesses comparability.In contrast, in the case of user's interpretational criteria is inconsistent, different user is to same The scoring of service has different implications, so as to the clothes for obtaining different user scoring using known methods such as cumulative or averaging methods Business prestige does not have comparability clearly.
High efficiency：5 are shared in solution space M5Kind may solve, and exhaustive optimization method will calculate 55It is secondary just to obtain most Excellent solution, and because initial temperature and final temperature are certain in the present invention, according to warm formula T=0.9t is moved back, at most carry out 43 drops Temperature, and be cooled to every time carry out 39 iteration, and the generation of new explanation has randomness, thus amount of calculation be necessarily less than 43 × 39 times, the far smaller than amount of calculation of exhaustive optimization method for solving；Meanwhile produced by changing the Partial Elements value in current solution at random Raw new explanation, reduces algorithm and takes；In addition, the probabilistic jumping property of simulated annealing makes searching process not to be absorbed in local optimum Trap, shorten optimize the time while, do not reduce low optimization accuracy, improve the efficiency of measures of reputation.
Anti- maneuverability：It is according to the obtained user-service rating matrix R services reputation that the known method of average calculates (3.75,2.25,3,3.5,2.75), if one user u of increase1To servicing s2The evaluation of 5 points of high score, then by known flat The obtained services reputation that equal method calculates is (3.75,2.8,3,3.5,2.75) so that the originally minimum service s of prestige2It is better than s3, so as to reach the purpose of prestige manipulation.And if one user u of increase1To servicing s25 points of high score evaluation, use this The services reputation that inventive method obtains is that (3,1,3,4,4) are compared with the services reputation (3,2,3,5,5) obtained in step 3, is not had It is improved manipulation service s2Ranking with reach prestige manipulation purpose.Therefore known in the services reputation measurement method ratio of the present invention The anti-maneuverability of method is higher.
Above in conjunction with accompanying drawing to the present invention embodiment be explained in detail, but the present invention be not limited to it is above-mentioned Embodiment, can also be before present inventive concept not be departed from those of ordinary skill in the art's possessed knowledge Put that various changes can be made.

Claims (2)

  1. A kind of 1. online service measures of reputation method based on KendallTau distances, it is characterised in that：Referred to first by distance Mark to weigh the uniformity between two service scoring vectors；Then measures of reputation is modeled as an optimization problem, this Optimization problem is exactly to find one and user-minimum reputation vectors of service rating matrix distance；Finally use simulated annealing Algorithm solves the optimization problem, using obtained reputation vectors as services reputation.
  2. 2. the online service measures of reputation method according to claim 1 based on KendallTau distances, it is characterised in that： Methods described comprises the following steps that：
    Step 1, by range index come weigh two service scoring vector between uniformity；
    1.1 set U={ u1,u2,...,uxIt is to have the set of x user, S={ s1,s2,...,syTo there is the set of y service, User represents with x × y matrix R the score information of service, the element r in RusRepresent that user u is commented service s Point；
    1.2 pass through range index K (ri,rj) two vectorial r of service scoring of measurementi=(ri1,...,rip,...,riy) and rj= (rj1,...,rjq,...,rjy) between uniformity：
    For two different service sp,sq∈ S (p ≠ q), calculate two vectorial r of service scoringi=(ri1,...,rip,..., riy) and rj=(rj1,...,rjq,...,rjy) between Kendall tau distances：
    Wherein, rip(i=1,2 ..., x, p=1,2 ..., y) it is i-th of user uiTo p-th of service spScoring, represent use Family uiTo servicing spSatisfaction, ripIt is more big, represent uiTo spIt is more satisfied；rip＞ riqRepresent user uiThink to service spIt is better than sq, rip＜ riqRepresent user uiThink to service sqBetter than sp, rip=riqRepresent user uiThink to service spAnd sqIndifference, rip=0 Represent user uiIt is not given to service spScoring or user did not produce with the service and interacted, at this point for service sp,sq∈ S (p ≠ q), do not calculate the vectorial r of service scoringiAnd rjThe distance between；
    For set of service S={ s1,s2,...,sy, calculate two vectorial r of service scoringi=(ri1,...,rip,...,riy) and rj=(rj1,...,rjq,...,rjyThe distance between)：
    <mrow> <mi>K</mi> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>r</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>s</mi> <mi>p</mi> </msub> <mo>,</mo> <msub> <mi>s</mi> <mi>q</mi> </msub> <mo>&amp;Element;</mo> <mi>S</mi> </mrow> </munder> <msub> <mi>K</mi> <mrow> <msub> <mi>s</mi> <mi>p</mi> </msub> <mo>,</mo> <msub> <mi>s</mi> <mi>q</mi> </msub> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>r</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>
    Wherein K (ri,rj) represent riAnd rjBetween preference uniformity, K (ri,rj) smaller, represent riAnd rjBetween preference it is consistent Property is bigger；K(ri,rj) bigger, represent riAnd rjBetween preference uniformity it is smaller；K(ri,rj)=0 represents riAnd rjBetween it is inclined It is good completely the same；
    Step 2 and then measures of reputation is modeled as an optimization problem, this optimization problem is exactly to find one with using Family-minimum the reputation vectors of service rating matrix distance；
    R in 2.1 counting users-service rating matrix Rip＞ riq, rip=riq, rip＜ riqService to quantity, and use N respectively (rip＞ riq), N (rip=riq), N (rip<riq) represent；
    2.2 use a reputation vectors rz=(rz1,rz2...,rzy) represent set of service S={ s1,s2,...,syPrestige, Based on K (r in formula (1) and formula (2)i,rj) definition, with reference to the r counted in step 2.1ip＞ riq, rip=riq, rip＜ riq Service to quantity, calculate reputation vectors rzThe distance between user-service rating matrix R；
    For two different service sp,sq∈ S (p ≠ q), calculate reputation vectors rzBetween user-service rating matrix R Distance：
    For set of service S={ s1,s2,...,sy, rzThe distance between user-service rating matrix R is：
    <mrow> <mi>K</mi> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>z</mi> </msub> <mo>,</mo> <mi>R</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>s</mi> <mi>p</mi> </msub> <mo>,</mo> <msub> <mi>s</mi> <mi>q</mi> </msub> <mo>&amp;Element;</mo> <mi>S</mi> </mrow> </munder> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>R</mi> </mrow> </munder> <msub> <mi>K</mi> <mrow> <msub> <mi>s</mi> <mrow> <mi>p</mi> <mo>,</mo> </mrow> </msub> <msub> <mi>s</mi> <mi>q</mi> </msub> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>z</mi> </msub> <mo>,</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>
    Wherein K (rz, R) and represent reputation vectors rzPreference uniformity between user-service rating matrix R, K (rz, R) and it is smaller, Represent rzPreference uniformity between R is bigger；K(rz, R) and it is bigger, represent rzPreference uniformity between R is smaller；K(rz, R r)=0 is representedzPreference between R is completely the same.
    2.3 determine that measures of reputation optimization objective function is as follows：
    <mrow> <mi>f</mi> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>b</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mi>min</mi> <mi> </mi> <mi>K</mi> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>b</mi> </msub> <mo>,</mo> <mi>R</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>s</mi> <mi>p</mi> </msub> <mo>,</mo> <msub> <mi>s</mi> <mi>q</mi> </msub> <mo>&amp;Element;</mo> <mi>S</mi> </mrow> </munder> <munder> <mo>&amp;Sigma;</mo> <mrow> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>&amp;Element;</mo> <mi>R</mi> </mrow> </munder> <msub> <mi>K</mi> <mrow> <msub> <mi>s</mi> <mrow> <mi>p</mi> <mo>,</mo> </mrow> </msub> <msub> <mi>s</mi> <mi>q</mi> </msub> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>b</mi> </msub> <mo>,</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>
    In formula (5)：rbFor object function f (rb) one may solution, rb=(rb1,rb2...,rby), wherein rb1,rb2..., rbyRepresent set of service S={ s1,s2,...,syPrestige；Reputation vectors r in step 2.2zIt is object function f (rb) one Individual to solve, all possible solution vector constitutes the solution space M of optimization objective function；
    Step 3, using the optimization problem in simulated annealing settlement steps to deal 2, believe obtained reputation vectors as service Reputation；
    The parameter of 3.1 setting simulated annealings, including initial temperature t0, final temperature te, move back warm factor alpha, markov chain length L, initial parameter, which is chosen, uses following empirical values：t0=100, te=1, α=0.9, L=100, in calculating process, make end condition For：New explanation under temperature t in continuous iterative process several times is not all received or temperature is reduced to final temperature, and satisfaction is wherein appointed Then algorithm terminates one condition；
    3.2 setting initial solution r0, a possible solution is randomly selected from the solution space M of step 2.3 and is used as initial solution r0, r0= (r01,r02...,r0y), calculate r0Target function value K (r0,R)；
    3.3 random change initial solution r0In element value, produce one be located at solution space M new explanation rn, rn=(rn1,rn2..., rny), calculate rnTarget function value K (rn,R)；
    3.4 calculate rnObject function K (rn, R) and r0Object function K (r0, R) difference DELTA k, Δ k=K (rn,R)-K(r0, R)；
    The acceptance probability of 3.5 new explanations is：
    In formula (6)：T is Current Temperatures, the temperature T=α t after cooling
    The receiving of new explanation follows Metropolis criterions：As Δ k ＜ 0, receive rnAs current optimal solution, as Δ k >=0, give The random number β gone out on (0, a 1) section, in the acceptance probability P ＞ β of new explanation, receive rnAs current optimal solution, otherwise not Receive rn, now initial solution r0Realize an iteration；L iteration is carried out altogether under Current Temperatures, if being received in iterative process New solution, then be unsatisfactory for end condition, and temperature is reduced according to T=α t；
    3.6 make Current Temperatures t=T, repeat step 3.3 to 3.5, until meeting end condition, now stop calculating, that tries to achieve works as Preceding solutionFor optimal solution；
    3.7 optimal solutions that will be obtainedServices reputation as user-service rating matrix R.
CN201710484114.1A 2017-06-23 2017-06-23 A kind of online service measures of reputation method based on KendallTau distances Pending CN107464032A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710484114.1A CN107464032A (en) 2017-06-23 2017-06-23 A kind of online service measures of reputation method based on KendallTau distances

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710484114.1A CN107464032A (en) 2017-06-23 2017-06-23 A kind of online service measures of reputation method based on KendallTau distances

Publications (1)

Publication Number Publication Date
CN107464032A true CN107464032A (en) 2017-12-12

Family

ID=60546066

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710484114.1A Pending CN107464032A (en) 2017-06-23 2017-06-23 A kind of online service measures of reputation method based on KendallTau distances

Country Status (1)

Country Link
CN (1) CN107464032A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108763206A (en) * 2018-05-22 2018-11-06 南京邮电大学 A method of quicksort is carried out to single text keyword

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108763206A (en) * 2018-05-22 2018-11-06 南京邮电大学 A method of quicksort is carried out to single text keyword
CN108763206B (en) * 2018-05-22 2022-04-05 南京邮电大学 Method for quickly sequencing keywords of single text

Similar Documents

Publication Publication Date Title
Wang et al. MeLoDy: A long-term dynamic quality-aware incentive mechanism for crowdsourcing
US10832269B2 (en) API pricing based on relative value of API for its consumers
Opricovic et al. Compromise solution by MCDM methods: A comparative analysis of VIKOR and TOPSIS
CN103744917B (en) Method and system are recommended in mixing
Zhou et al. Particle swarm optimization for trust relationship based social network group decision making under a probabilistic linguistic environment
Lu et al. HBGG: A hierarchical Bayesian geographical model for group recommendation
Lambert et al. Eliciting truthful answers to multiple-choice questions
CN107256241B (en) Movie recommendation method for improving multi-target genetic algorithm based on grid and difference replacement
Zhang et al. Bayesian hybrid collaborative filtering-based residential electricity plan recommender system
CN105354260B (en) The Mobile solution of a kind of mosaic society&#39;s network and item characteristic recommends method
CN107016569A (en) The targeted customer&#39;s account acquisition methods and device of a kind of networking products
CN104462383A (en) Movie recommendation method based on feedback of users&#39; various behaviors
Sha et al. A Network‐Based Approach to Modeling and Predicting Product Coconsideration Relations
CN109902823A (en) A kind of model training method and equipment based on generation confrontation network
CN116244513B (en) Random group POI recommendation method, system, equipment and storage medium
CN109840702A (en) A kind of new projects&#39; collaborative recommendation method based on multi-core integration
Guo et al. Social trust aware item recommendation for implicit feedback
Pliszczuk et al. Forecasting sales in the supply chain based on the LSTM network: the case of furniture industry
Rostamzadeh et al. An application of a hybrid MCDM method for the evaluation of entrepreneurial intensity among the SMEs: a case study
CN107909498B (en) Recommendation method based on area below maximized receiver operation characteristic curve
CN117216382A (en) Interactive processing method, model training method and related device
CN109857928A (en) User preference prediction technique based on polynary credit evaluation
CN107464032A (en) A kind of online service measures of reputation method based on KendallTau distances
Hurley et al. Attacking recommender systems: A cost-benefit analysis
Faltings et al. Peer truth serum: incentives for crowdsourcing measurements and opinions

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20171212