CN106095761A - A kind of multiple criteria decision make method and device - Google Patents

Abstract

The invention discloses a kind of multiple criteria decision make method, comprising: S1. user inputs decision object (target)；S2. determining set of criteria, this set must be complete and nonredundant, defines semantic variant, inputs user preference information, other parameter informations that input system is used；S3. foundation criterion is dissimilar, calculates the weight of each criterion；S4. according to user's cuit, the criterion of definition, parameter information, the original rating matrix of multiple criteria is generated；Then successively original rating matrix Triangular Fuzzy Number and weighting are processed and obtain weighting user-project rating matrix；S5. use modified fuzzy TOPSIS method that the score value under different criterions is gathered into the overall approach degree of target.The approach degree that S5 is obtained by the linear combination S6. calculating with content-based recommendation and collaborative filtering recommending respectively enters the process of progressive step, and carries out linear combination to two kinds of arithmetic result；S7. for certain specific user by result from optimum to the most bad sequence, it is recommended that give user.The invention also discloses a kind of multiple criteria decision make device.Invention can significantly provide accuracy and the user satisfaction of recommendation under multiple criteria environment.

Description

A kind of multiple criteria decision make method and device

Technical field

The present invention relates to field of computer technology, particularly relate to a kind of multiple criteria decision make method and device.

Background technology

With developing rapidly of information technology, on the one hand, people enjoy the facility that information sharing brings to the full；The opposing party Face, people are also more and more universal recognizes the adverse consequences that information overload brings.The people thirsting for obtaining useful information expect Can there is a kind of efficient way helping them to be better understood by self-demand, and more effectively utilize information resources.

Commending system, is a kind of special Information Filtering Technology, and this technology is closely related with many subjects, for example: letter Breath system, machine learning, data mining etc.；By providing a user with the recommendation service of personalization, it is recommended that system is also a kind of The method of important solution problem of information overload.As a rule, it is recommended that system refers to some specific technology and software, they will " project " recommends user；Here, " project " is a kind of general name, refers to commending system and recommends certain individually defined thing of user Product.

In general, it is recommended that first system is assembled item information and user preference information；That is: commending system needs to use certain The mode of kind expresses the feature of user/project, and these features will be utilized for user preference modeling.Secondly the recommendation of commending system It is that one " project " is estimated one " scoring " that problem will be simplified as in the ordinary course of things according to user preferences modeling, and before This project being estimated scoring was not also contacted by certain user.

Although recommended technology quickly grows, reality also has the application of a large amount of commending system, but present most recommendation The many attributive character carrying in technology still neglected items feature.Most of commending systems still use single evaluation standard one The two-dimensional space of individual user × project is expressed the value of utility for certain user for certain project.

Many attributive character of project can be naturally converted into Multiple-criteria Decision Problems, i.e. MCDM problem.MCDM problem Multi criteriaproblem when being to specialize in decision-making, is generally used to find the optimal solution in multiple feasible solution.At present, even if Use multiple criteria points-scoring system, it often uses single algorithm, it was predicted that accuracy is not high, it is recommended that result often undesirable.

Content of the invention:

The present invention is to solve the defect that prior art exists, a kind of multiple criteria decision make method and device is provided, should Method uses a kind of new mixing proposed algorithm, by be combineding with each other of proposed algorithm, fuzzy logic and multiple MCDM method, Commending system can be improved greatly and recommend the forecasting accuracy under environment in multiple criteria.

The invention discloses a kind of multiple criteria decision make method, comprising:

S1. user inputs decision object (target)；

S2. the decision object for input, determines one group of set of criteria, and this set must be complete and nonredundant, and Definition semantic variant, inputs user preference information；Other parameter informations of input system；

S3. foundation criterion is dissimilar, calculates the weight of each criterion；

S4. according to S2, multiple criteria scoring is generated $r_{u,i}({\left(r_i,s_i,t_i\right)}_{c_1},{\left(r_i,s_i,t_i\right)}_{c_2},...{\left(r_i,s_i,t_i\right)}_{c_k});$ U is to represent to use Family, i represents project；Multiple decision object of input, constitute user-project rating matrix R, enter matrix R by Triangular Fuzzy Number Row is processed, and obtains fuzzy user-project rating matrixUtilize the weight that S3 obtains, to fuzzy user-project rating matrix It is weighted processing, obtain weighting user-project rating matrix

S5. use the fuzzy TOPSIS method changed that the score value under different for project criterions is gathered into destination object Approach degree.Specific practice is with the fuzzy positive ideal solution setting $A^{*}=({\tilde{v}}_1^{*},{\tilde{v}}_2^{*},...{\tilde{v}}_k^{*})$ With fuzzy minus ideal result $A^{-}=({\tilde{v}}_1^{-},{\tilde{v}}_2^{-},...{\tilde{v}}_k^{-})$ Processing the matrix of S4, processing mode isWithThen target is calculated The approach degree of object, computational methods are

S6. the approach degree obtaining S5 by content-based recommendation technology and collaborative filtering recommending technology respectively is counted Calculate, and linear combination is carried out to two kinds of arithmetic result, more met user psychology demand, more accurately recommendation results.

S7. according to the approach degree result doping, for certain specific user by result from optimum to the most bad sequence, it is recommended that give User.

In a kind of multiple criteria decision make method of the present invention, following steps can be resolved in step S3:

S31. the type of judgment criterion, if the type of criterion is priority and importance, then enters step S33；If criterion Type be to judge to influence each other relation, then enter step S32.

S32. AHP method is used to carry out calculation criterion weight.AHP (Analytic Hierarchy Process) i.e. level divides Analysis method, is that the U.S. scholar T.L.Saaty that plans strategies for teaches the multi-scheme of a kind of practicality proposing the seventies in last century or multiple target Decision-making technique, be the method for decision analysis of a kind of combination of qualitative and quantitative analysis.

S33. DEMATEL method is used to carry out calculation criterion weight.DEMATEL(Decision Making Trial and Evaluation Laboratory) i.e. decision-making test and evaluation experimental method are a kind of for screening mainly wanting of complication system Element, the process of relieving system structural analysis and the methodology that proposes.

In a kind of multiple criteria decision make method of the present invention, following steps can be resolved in step S6:

S61. the approach degree collaborative filtering recommending technology obtaining S5 is processed, the formula of collaborative filtering: ${\mathrm{CC}}_{u,i}\left(CF\right)=\frac{1}{{\mathrm{\Σ}}_{u^{\′}\∈U}\left|sim(u,u^{\′})\right|}{\mathrm{\Σ}}_{u^{\′}\∈U}\left(sim\right(u,u^{\′})\×C_{u,i});$

S62. the approach degree obtaining S5 is processed by based on commending contents technology, based on the formula of commending contents: ${\mathrm{CC}}_{u,i}\left(CB\right)=\frac{\underset{all-similar-items}{\Σ}\left(sim\right(i,i^{\′})\×C_{u,i})}{\underset{all-similar-items}{\Σ}\left|sim(i,i^{\′})\right|}.$

S63. the result that S61 and S62 obtains is carried out linear combination, user-defined linear combination weight,It is used for The importance of measure algorithm, when multiple projects are similar, then content-based recommendation technology can be more accurate, then now proportion Arrange larger；If project is mostly the scoring of friendly neighbour's project, then collaborative filtering recommending technology can be more accurate, then now than WeightArrange smaller, the formula of linear combination: ${\mathrm{CC}}_{u,i}=\widehat{\mathrm{\ω}}*{\mathrm{CC}}_{u,i}\left(CB\right)+(1-\widehat{\mathrm{\ω}})*\left(CF\right).$

The invention also discloses a kind of multiple criteria decision make device, be used for realizing said method, comprising:

Input and define decision object or object module: input decision object, and be defined as bullets or project Collection, it is recommended that these projects of system default are the projects of user's more preference, system decision-making object definition is " sequence ", described " row Sequence " definition is to be ranked up project to the most bad from optimum；

Definition criterion and parameter module: in order to project is analyzed, needs to define a plurality of criterion, and determine that criterion is weighed Weight, two kinds of methods of this module DEMATEL and AHP determine criterion weight；This module other also requires that user inputs other systems The parameter information used；

Realize user's multiple criteria score data module: this module is according to user's cuit, the criterion of definition, parameter letter Breath, can generate the original rating matrix of multiple criteria, it may be assumed that user-project rating matrix R；Then successively original rating matrix is used Triangular Fuzzy Number processes and obtains fuzzy user-project rating matrix, and is weighted processing, and obtains weighting user-project scoring Matrix；

Generate the overall approach degree module of target: this module uses modified fuzzy TOPSIS method by under difference criterion Score value be gathered into the overall approach degree of target.

Generate recommendation results module: this module passes through the linear combination of content-based recommendation and collaborative filtering recommending technology Obtain final recommendation results, and by recommendation results from excellent to bad, sort recommendations.

Brief description

Fig. 1 is the method flow diagram that the embodiment of the present invention provides；

Fig. 2 is the structure drawing of device that the embodiment of the present invention provides

Detailed description of the invention

In order to make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the accompanying drawings with specific embodiment pair The present invention describes in detail.

Embodiment: a kind of multiple criteria decision make method implementation method of the present embodiment, as it is shown in figure 1, by fuzzy number and Semantic variant describes user preference information.By DEMATEL and the AHP algorithm in MCDM algorithm come between weighing criteria mutual Relation and significance level.Obtain in conjunction with the collaborative filtering in TOPSIS algorithm and conventional recommendation technology and based on the algorithm of content More preferable recommendation results.Its step is decomposed into:

S1. user inputs decision object (target)；

S2. the decision object for input, determines one group of set of criteria, and this set must be complete and nonredundant, and Definition semantic variant, inputs user preference information；Other parameter informations of input system；

S3. foundation criterion is dissimilar, calculates the weight of each criterion；S3 is decomposed into following steps:

S31. the type of judgment criterion, if the type of criterion is priority and importance, then enters step S33；If criterion Type be to judge to influence each other relation, then enter step S32.

S32. AHP method is used to carry out calculation criterion weight.AHP (Analytic Hierarchy Process) i.e. level divides Analysis method, is that the U.S. scholar T.L.Saaty that plans strategies for teaches the multi-scheme of a kind of practicality proposing the seventies in last century or multiple target Decision-making technique, be the method for decision analysis of a kind of combination of qualitative and quantitative analysis.

S33. DEMATEL method is used to carry out calculation criterion weight.DEMATEL(Decision Making Trial and Evaluation Laboratory) i.e. decision-making test and evaluation experimental method are a kind of for screening mainly wanting of complication system Element, the process of relieving system structural analysis and the methodology that proposes.

S4. according to S2, multiple criteria scoring is generated $r_{u,i}({\left(r_i,s_i,t_i\right)}_{c_1},{\left(r_i,s_i,t_i\right)}_{c_2},...{\left(r_i,s_i,t_i\right)}_{c_k});$ U is to represent to use Family, i represents project；Multiple decision object of input, constitute user-project rating matrix R, enter matrix R by Triangular Fuzzy Number Row is processed, and obtains fuzzy user-project rating matrixUtilize the weight that S3 obtains, to fuzzy user-project rating matrix It is weighted processing, obtain weighting user-project rating matrix

S5. use the fuzzy TOPSIS method changed that the score value under different for project criterions is gathered into destination object Approach degree.Specific practice is with the fuzzy positive ideal solution setting $A^{*}=({\tilde{v}}_1^{*},{\tilde{v}}_2^{*},...{\tilde{v}}_k^{*})$ With fuzzy minus ideal result $A^{-}=({\tilde{v}}_1^{-},{\tilde{v}}_2^{-},...{\tilde{v}}_k^{-})$ Processing the matrix of S4, processing mode isWithThen target is calculated The approach degree of object, computational methods are

S6. the approach degree obtaining S5 by content-based recommendation technology and collaborative filtering recommending technology respectively is counted Calculate, and linear combination is carried out to two kinds of arithmetic result, more met user psychology demand, more accurately recommendation results.S6 can To resolve into following steps:

S61. the approach degree collaborative filtering recommending technology obtaining S5 is processed, the formula of collaborative filtering: ${\mathrm{CC}}_{u,i}\left(CF\right)=\frac{1}{{\mathrm{\Σ}}_{u^{\′}\∈U}\left|sim(u,u^{\′})\right|}{\mathrm{\Σ}}_{u^{\′}\∈U}\left(sim\right(u,u^{\′})\×C_{u,i});$

S62. the approach degree obtaining S5 is processed by based on commending contents technology, based on the formula of commending contents: ${\mathrm{CC}}_{u,i}\left(CB\right)=\frac{\underset{all-similar-items}{\Σ}\left(sim\right(i,i^{\′})\×C_{u,i})}{\underset{all-similar-items}{\Σ}\left|sim(i,i^{\′})\right|}.$

S63. the result that S61 and S62 obtains is carried out linear combination, user-defined linear combination weight,It is used for The importance of measure algorithm, when multiple projects are similar, then content-based recommendation technology can be more accurate, then now proportion Arrange larger；If project is mostly the scoring of friendly neighbour's project, then collaborative filtering recommending technology can be more accurate, then now than WeightArrange smaller, the formula of linear combination: ${\mathrm{CC}}_{u,i}=\widehat{\mathrm{\ω}}*{\mathrm{CC}}_{u,i}\left(CB\right)+(1-\widehat{\mathrm{\ω}})*\left(CF\right).$

S7. according to the approach degree result doping, for certain specific user by result from optimum to the most bad sequence, it is recommended that give User.

Correspondingly, the present invention provides a kind of multiple criteria decision make device, as shown in Figure 2, comprising:

Input and define decision object or object module 10: input decision object, and be defined as bullets or item Mesh collection, it is recommended that these projects of system default are the projects of user's more preference, system decision-making object definition is " sequence ", described Definition of " sorting " is to be ranked up project to the most bad from optimum；

Definition criterion and parameter module 20: in order to project is analyzed, needs to define a plurality of criterion, and determine that criterion is weighed Weight, two kinds of methods of this module DEMATEL and AHP determine criterion weight；This module other also requires that user inputs other systems The parameter information used；

Realize user's multiple criteria score data module 30: this module is according to user's cuit, the criterion of definition, parameter letter Breath, can generate the original rating matrix of multiple criteria, it may be assumed that user-project rating matrix R；Then successively original rating matrix is used Triangular Fuzzy Number processes and obtains fuzzy user-project rating matrix, and is weighted processing, and obtains weighting user-project scoring Matrix；

Generate the overall approach degree module 40 of target: this module uses modified fuzzy TOPSIS method by difference criterion Under score value be gathered into the overall approach degree of target.

Generate recommendation results module 50: this module passes through linear group of content-based recommendation and collaborative filtering recommending technology Incompatible obtain final recommendation results, and by recommendation results from excellent to bad, sort recommendations.

To sum up, technical staff is without departing from the present invention, can carry out suitable adjustment to disclosed device, by This, as described above being for illustration only and the purpose that not limits, technical staff should be distinctly understood that inconspicuous change above-mentioned Operational circumstances under the slightly modified purpose reaching same effect can be carried out to disclosed device or technique, the present invention is by weighing Profit claim makes restriction.

Claims (4)

1. a multiple criteria decision make method, it is characterised in that include:

S1. user inputs decision object (target)；

S2. the decision object for input, determines one group of set of criteria, and this set must be complete and nonredundant, and defines Semantic variant, inputs user preference information；Other parameter informations of input system；

S3. foundation criterion is dissimilar, calculates the weight of each criterion；

S4. according to S2, multiple criteria scoring is generatedU is to represent user, i generation Table entry；Multiple decision object of input, constitute user-project rating matrix R, by Triangular Fuzzy Number to matrix R process, Obtain fuzzy user-project rating matrixUtilize the weight that S3 obtains, to fuzzy user-project rating matrixIt is weighted Process, obtain weighting user-project rating matrix

S5. use the fuzzy TOPSIS method changed that the score value under different for project criterions is gathered into the patch of destination object Recency.Specific practice is with the fuzzy positive ideal solution settingWith fuzzy minus ideal result Processing the matrix of S4, processing mode isWithThen target is calculated The approach degree of object, computational methods are

S6. the approach degree obtaining S5 by content-based recommendation technology and collaborative filtering recommending technology respectively calculates, and Linear combination is carried out to two kinds of arithmetic result, is more met user psychology demand, more accurately recommendation results.

S7. according to the approach degree result doping, for certain specific user by result from optimum to the most bad sequence, it is recommended that give and use Family.

2. a kind of multiple criteria decision make method as claimed in claim 1, it is characterised in that be decomposed into following in step S3 Step:

S31. the type of judgment criterion, if the type of criterion is priority and importance, then enters step S33；If the class of criterion Type is to judge to influence each other relation, then enter step S32.

S32. AHP method is used to carry out calculation criterion weight.AHP (Analytic Hierarchy Process) i.e. step analysis Method, is that the U.S. scholar T.L.Saaty that plans strategies for teaches the multi-scheme or multiobject of a kind of practicality proposing the seventies in last century Decision-making technique, is the method for decision analysis of a kind of combination of qualitative and quantitative analysis.

S33. DEMATEL method is used to carry out calculation criterion weight.DEMATEL(Decision Making Trial and Evaluation Laboratory) i.e. decision-making test and evaluation experimental method are a kind of for screening mainly wanting of complication system Element, the process of relieving system structural analysis and the methodology that proposes.

3. a kind of multiple criteria decision make method as claimed in claim 1, it is characterised in that walk below decomposable asymmetric choice net in step S6 Rapid:

S61. the approach degree collaborative filtering recommending technology obtaining S5 is processed, the formula of collaborative filtering: ${\mathrm{CC}}_{u,i}\left(CF\right)=\frac{1}{{\mathrm{\Σ}}_{u^{\′}\∈U}\left|sim(u,u^{\′})\right|}{\mathrm{\Σ}}_{u^{\′}\∈U}\left(sim\right(u,u^{\′})\×C_{u,i});$

S62. the approach degree obtaining S5 is processed by based on commending contents technology, based on the formula of commending contents: ${\mathrm{CC}}_{u,i}\left(CB\right)=\frac{\underset{all-similar-items}{\Σ}\left(sim\right(i,i^{\′})\×C_{u,i})}{\underset{all-similar-items}{\Σ}\left|sim(i,i^{\′})\right|}.$

S63. the result that S61 and S62 obtains is carried out linear combination, user-defined linear combination weightIt is used for weighing The importance of quantity algorithm, when multiple projects are similar, then content-based recommendation technology can be more accurate, then now proportionIf Put larger；If project is mostly the scoring of friendly neighbour's project, then collaborative filtering recommending technology can be more accurate, then now proportionArrange smaller, the formula of linear combination:

4. a multiple criteria decision make device, for realizing the method described in claim 1, it is characterised in that include:

Input and define decision object or object module: input decision object, and be defined as bullets or Item Sets, push away Recommending the project that these projects of system default are user's more preferences, system decision-making object definition is " sequence ", and described " sequence " is fixed Justice is to be ranked up project to the most bad from optimum；

Definition criterion and parameter module: in order to project is analyzed, needs to define a plurality of criterion, and determine criterion weight, this Module DEMATEL and two kinds of methods of AHP determine criterion weight；This module other also requires that user inputs what other systems were used Parameter information；

Realize user's multiple criteria score data module: this module is according to user's cuit, the criterion of definition, parameter information, energy Enough generate the original rating matrix of multiple criteria, it may be assumed that user-project rating matrix R；Then successively to original rating matrix Triangle Module Stick with paste number process and obtain fuzzy user-project rating matrix, and be weighted processing, obtain weighting user-project rating matrix；

Generate the overall approach degree module of target: this module uses modified fuzzy TOPSIS method by commenting under different criterions Score value is gathered into the overall approach degree of target.

Generate recommendation results module: this module is come by the linear combination of content-based recommendation and collaborative filtering recommending technology To final recommendation results, and by recommendation results from excellent to bad, sort recommendations.