CN102508907A - Dynamic recommendation method based on training set optimization for recommendation system - Google Patents

Abstract

The invention discloses a dynamic recommendation method based on training set optimization for a recommendation system, which specifically includes: (1) establishing a preliminary recommendation portion: generating an original recommendation model according to original user grading data; (2) performing AdaBoost trainings: utilizing the original recommendation model as a classifying and judging basis to classify the data and adjust learning times of samples by means of multiple iterative learning training data; (3) screening incorrect samples: data of selected difficult samples are removed as the incorrect samples after multiple AdaBoost trainings so as to construct a new training data collection; (4) reconstructing a recommendation model: combining training results to regenerate the recommendation model based on the new training data; and (5) generating recommendation results: utilizing the new recommendation model to generate the recommendation results. The method is capable of removing the data without referential meaning in recommended service by the aid of great relevance of original training set data in content, so that validity of the training data and precision of the final recommendation model are improved.

Description

A kind of dynamic recommend method of the commending system of optimizing based on training set

Technical field

The present invention relates to the technical field of user's commending system, particularly a kind of dynamic recommend method of the commending system of optimizing based on training set.

Background technology

The personalized recommendation service is customer-centric, is the basis to understand user preferences, is the method for service that the user provides customized customized information to appear, and also is to solve a kind of effective way of from the magnanimity Internet resources, extracting user's information needed.Compare the personalized recommendation service with the generic services pattern following characteristics are arranged: at first; The personalized recommendation service can save the user to come out from the predicament of information overload; Make the user can have an opportunity to enjoy the hommization network information service really rich and varied, convenient appropriateness, greatly promote user experience and satisfaction; Secondly, the personalized recommendation service can fully improve the service quality and the access efficiency of Web website, can also find the point of interest that the user is potential simultaneously, thereby excavate potential commercial value, for Internet service provider provides considerable economy return.

Since the commending system based on collaborative filtering technology is born, the proposition of the latent vectorial recommended models of especially decomposing based on normalized matrix, the personalized recommendation technology has had quite high lifting in the recommendation precision of theoretical aspect.Original score data as recommending important evidence has decisive influence to final recommendation results, and obvious one group of data with high accuracy can obtain good recommendation effect in final recommendation.

The data set that general personalized recommendation service to the user all is based on existing historical accumulation carries out, and the data volume of this data set is very huge.Data set scale huge is difficult to avoid in the collection of data, exist irrational data, replaces phenomenons such as scoring such as user's mistake scoring or non-user.These data itself do not have with reference to property, and these data should not adopted in to user's recommendation service.Therefore, processing and the selection to original score data can help to improve the recommendation precision to a great extent.Adopting the training set of determination methods screening comparatively accurately, and carrying out the foundation of recommended models on this basis, the recommended models that is obtained so can recommend on the precision more remarkable must the lifting arranged.

Summary of the invention

The technical matters that the present invention will solve is: the deficiency that overcomes prior art; A kind of dynamic recommend method of the commending system of optimizing based on training set is provided; This method can be through screening the original training data as the personalized recommendation foundation; And serve as according to obtaining to have the more recommended models of high accurancy and precision, the accuracy that has improved personalized recommendation with the new training set of removing error sample.

The technical scheme that the present invention solves the problems of the technologies described above is: a kind of dynamic recommend method of the commending system of optimizing based on training set, and these method concrete steps are following:

Step (1) is set up preliminary recommended models: according to original user's score data, utilize based on the modeling method in the normalized matrix factorization recommended models and generate initial recommended models;

Step (2) AdaBoost training: the foundation of utilizing the middle recommended models that generates of step (1) to judge as primary classification makes up grader; The recommendation that calculates based on recommended models and the classification of the difference condition decision data between the raw value; Utilize AdaBoost algorithm study original training sample, and take turns the new grader of end back generation at each;

Step (3) screening error sample: each is taken turns training and all need filter out difficult sample in the training process that utilizes the AdaBoost algorithm; The division of difficult sample can adopt the otherness between predicted value and the actual value to judge in the method, promptly judges awkward sample during greater than a certain threshold value when this species diversity.After the AdaBoost training that warp is too much taken turns, the data that repeatedly are chosen awkward sample can be used as the error sample removal, are used for the required training data set of next iteration thereby construct;

Step (4) reconstruct recommended models: the training data to obtain in the step (3) is the basis, in conjunction with the AdaBoost training data, regenerates recommended models.

Step (5) produces recommendation results: as input, the recommended models of utilizing step (4) to obtain calculates recommendation results and returns to the user with the user characteristics vector.

, specific as follows in the said step (2) to the AdaBoost cluster training of raw data set:

1. step revises normalized matrix factorization recommended models, no longer original score data set T is divided into two sub-set T ₁, T ₂, wherein, data set T ₁Be used to learn data set T ₂Be used for judging that study stops wherein, but all data among the data set T all learnt, set the AdaBoost training iteration wheel number I, whenever take turns the number of times R of study, the error range errPermission of permission, and initialization feature vector set;

2. step utilizes normalized matrix factorization recommended models learning training data R time in the iteration of the first round, on the set of eigenvectors that training obtains in the calculation training data respective user to the estimated value of the scoring of project And obtain itself and actual value r _{U, i}Absolute error, i.e. absolute error

The AbsE value that 3. step calculates in 2. when step is judged the awkward sample of secondary data during greater than errPermission, travels through the total errCount that whole training datas obtains difficult sample ⁿAnd go out the error rate ε of sample by computes ⁿ, wherein | T| representes the number of samples in the training set;

${\mathrm{\ϵ}}^n=\frac{{\mathrm{ErrCount}}^n}{\left|T\right|}$ (1) formula;

In the formula: ε ⁿExpression sample error rate, | T| representes the number of samples in the training set, errCount ⁿThe whole training data of expression traversal obtains the sum of difficult sample;

The error rate ε that 4. step is calculated in 3. according to step ⁿThe adjustment study number of times of training sample in the next round iteration is specially when the AbsE of training sample data value during less than errPermission, and the study number of times of this sample in the next round iteration is trainTime ^N+1=trainTime ⁿ* ε ⁿ(if tramTime ^N+1Get 1 for＜1), when the AbsE of training sample data value determined that it is difficult sample in (2) during greater than errPermission, the study number of times of this sample in the next round iteration did ${\mathrm{TrainTime}}^{n+1}=\frac{{\mathrm{TrainTime}}^n}{{\mathrm{\ϵ}}^n},$ Available as shown in the formula expression:

${\mathrm{TrainTime}}^{n+1}=\left\{\begin{array}{cc}\frac{{\mathrm{TrainTime}}^n}{{\mathrm{\&epsiv;}}^n},& (\mathrm{AbsE}\&GreaterEqual;\mathrm{ErrPermission})\\ \mathrm{Max}({\mathrm{TrainTime}}^n*{\mathrm{\&epsiv;}}^n,1),& (\mathrm{AbsE}<\mathrm{ErrPermission})\end{array}\right.$ (2) formula;

In the formula: trainTime ⁿThe study number of times of sample in n next round iteration, trainTime ^N+1The study number of times of sample in n+1 next round iteration, ε ⁿThe error rate of calculating in expression step (3), the absolute error that the substep of AbsE value representation step (2) 2. calculates, errPermission representes the error range that allows;

Step 5. with errPermission with after the fixed proportion declineRate reduction, the iteration of a beginning new round, and the study number of times of each sample calculates in 4. according to step and carries out in the iteration that this is taken turns.

The method of the screening error sample in the said step (3), specific as follows:

Each is taken turns training and all need filter out difficult sample go forward side by side row labels and statistics in the training process that utilizes the AdaBoost algorithm in steps A, the step (2);

After the AdaBoost iteration training that step B, warp are too much taken turns, training data is traveled through the number of times that each sample of statistics is judged into difficult sample;

Step C, according to clearance delRate, remove from training set judging into the higher sample of difficult sample number of times, thereby obtain new training set.

The present invention's advantage compared with prior art is:

The present invention can be that automatic cluster is carried out on the basis with the available data; Iterative learning through repeatedly can filter out the misdata that is unfavorable for improving the recommended models precision; Thereby remove misdata and obtain high-quality training data, finally obtain higher recommendation precision.Only do not handle original training data is filtered to recommended models than classic method, this method can more effectively improve the recommendation precision.

Description of drawings

Fig. 1 is a summary workflow diagram of the present invention;

Fig. 2 is a detailed operation process flow diagram of the present invention.

Embodiment

Existing accompanying drawings embodiments of the invention.

As shown in Figure 2, the present invention includes four key steps: set up recommended models, AdaBoost training, screening error sample and reconstruct recommended models.

Step (1) is set up recommended models: read original user's score data and test data; And based on the Customs Assigned Number UserID of maximum in two data and the dimension that bullets ItemID confirms user characteristics vector sum item feature vector; Based on the modeling method in the normalized matrix factorization recommended models, newly-built and random initializtion user characteristics vector sum item feature vector;

Step (2) the AdaBoost training stage: the foundation structure sorter that utilizes the middle recommended models that generates of step (1) to judge as classification; The recommendation that calculates according to recommended models and the classification of the difference condition decision data between the raw value, concrete steps are:

1. step revises normalized matrix factorization recommended models, no longer original score data set T is divided into two sub-set T ₁, T ₂(utilize data set T ₁Learn, utilize T ₂Judge that study stops), but all data among the data set T are all learnt, set the AdaBoost training iteration wheel number I, whenever take turns the number of times R of study, the error range errPermission of permission, and initialization feature vector set;

Step 2. AdaBoost every take turns utilize in the iteration above-mentioned amended based on the modeling in the normalized matrix factorization recommended models and training method learning training data R time.After each study, calculating the RMSE of resulting user characteristics vector sum commodity proper vector model on training data, the RMSE value that reaches this wheel hour promptly when RMSE value can finish this in advance when big and takes turns iteration than learning value afterwards last time.This is taken turns and obtains that new recommended models---this takes turns final user characteristics vector sum commodity proper vector after the iteration.Utilize in the proper vector calculation training data respective user to the estimated value of the scoring of project

And obtain itself and actual value r _{U, i}Absolute error, i.e. absolute error

The AbsE value that 3. step calculates in 2. when step is judged the awkward sample of secondary data during greater than errPermission, and the mistake that these data are corresponding judges that number of times increases by 1, travels through the total errCount that whole training datas obtains difficult sample ⁿAnd calculate the error rate ε of sample by (1) formula ⁿ

${\mathrm{\&epsiv;}}^n=\frac{{\mathrm{ErrCount}}^n}{\left|T\right|}$ (1) formula;

In the formula: ε ⁿExpression sample error rate, | T| representes the number of samples in the training set, errCount ⁿThe whole training data of expression traversal obtains the sum of difficult sample;

The error rate ε that 4. step is calculated in 3. according to step ⁿ, utilize the study number of times of (2) formula adjustment training sample in the next round iteration;

${\mathrm{TrainTime}}^{n+1}=\left\{\begin{array}{cc}\frac{{\mathrm{TrainTime}}^n}{{\mathrm{\&epsiv;}}^n},& (\mathrm{AbsE}\&GreaterEqual;\mathrm{ErrPermission})\\ \mathrm{Max}({\mathrm{TrainTime}}^n*{\mathrm{\&epsiv;}}^n,1),& (\mathrm{AbsE}<\mathrm{ErrPermission})\end{array}\right.$ (2) formula;

In the formula: trainTime ⁿThe study number of times of sample in n next round iteration, trainTime ^N+1The study number of times of sample in n+1 next round iteration, ε ⁿThe error rate of calculating in expression step (3), the absolute error that the substep of AbsE value representation step (2) 2. calculates, errPermission representes the error range that allows;

5. step reduces errPermission with fixed proportion declineRate after; The iteration of a beginning new round; And the study number of times of each sample calculates in 4. according to step and carries out in the iteration that this is taken turns, should the stage if accomplished the iteration of pre-determined number then finished;

Step (3) screening error sample: each is taken turns training and all need filter out difficult sample in the training process that utilizes the AdaBoost algorithm; The division of difficult sample can adopt the otherness between predicted value and the actual value to judge in the method, promptly judges awkward sample during greater than a certain threshold value when this species diversity.After the AdaBoost training that warp is too much taken turns, the data that repeatedly are chosen awkward sample can be used as the error sample removal, are used for the required training data set of next iteration thereby construct;

Each is taken turns training and all need filter out difficult sample go forward side by side row labels and statistics in the training process that utilizes the AdaBoost algorithm in steps A, the step (2);

After the AdaBoost iteration training that step B, warp are too much taken turns, training data is traveled through the number of times that each sample of statistics is judged into difficult sample;

Step C, statistics calculates the maximum errors number that allows according to clearance del_rate, total sample number and 2., and frequency of training is set in training set is 0 with judging into sample that difficult sample number of times is higher than this value, thereby obtains new training set;

Step (4) reconstruct recommended models: the training data to obtain in the step (3) is the basis, in conjunction with the AdaBoost training data, regenerates recommended models based on the normalized matrix factorization method;

Step (5) produces recommendation results: with the product of user characteristics vector sum item feature vector as the estimation score value of specific user, with estimating that the higher project recommendation of score value gives this user to specific project.

The present invention can be that automatic cluster is carried out on the basis with the available data; Iterative learning through repeatedly can filter out the misdata that is unfavorable for improving the recommended models precision; Thereby remove misdata and obtain high-quality training data, finally obtain higher recommendation precision.Only do not handle original training data is filtered than classic method to recommended models; This method can more effectively improve commending system and recommend precision; As shown in table 1 below; When not adopting screening technique of the present invention, the RMSE of RMF method on test set is 0.792933, and several groups of data in the table 1 have all obtained less RMSE value.

The RMSE value that table 1 the present invention obtains

The part that the present invention does not set forth in detail belongs to techniques well known.Above embodiment is only in order to technical scheme of the present invention to be described but not be limited in the scope of embodiment; To those skilled in the art; As long as various variations claim limit and the spirit and scope of the present invention confirmed in; These variations are conspicuous, and all utilize innovation and creation that the present invention conceives all at the row of protection.

Claims (3)

1. the dynamic recommend method of a commending system of optimizing based on training set, it is characterized in that: these method concrete steps are following:

Step (1) is set up preliminary recommended models: according to original user's score data, utilize based on the modeling method in the normalized matrix factorization recommended models and generate initial recommended models;

Step (2) AdaBoost training: the foundation of utilizing the middle recommended models that generates of step (1) to judge as primary classification makes up grader; The recommendation that calculates based on recommended models and the classification of the difference condition decision data between the raw value; Utilize AdaBoost algorithm study original training sample, and take turns the new grader of end back generation at each;

Step (3) screening error sample: each is taken turns training and all need filter out difficult sample in the training process that utilizes the AdaBoost algorithm; The division of difficult sample can adopt the otherness between predicted value and the actual value to judge in the method, promptly judges awkward sample during greater than a certain threshold value when this species diversity; After the AdaBoost training that warp is too much taken turns, the data that repeatedly are chosen awkward sample can be used as the error sample removal, are used for the required training data set of next iteration thereby construct;

Step (4) reconstruct recommended models: the training data to obtain in the step (3) is the basis, in conjunction with the AdaBoost training data, regenerates recommended models.

Step (5) produces recommendation results: as input, the recommended models of utilizing step (4) to obtain calculates recommendation results and returns to the user with the user characteristics vector.

2. the dynamic recommend method of a kind of commending system of optimizing based on training set according to claim 1 is characterized in that: to the AdaBoost cluster training of raw data set, specific as follows in said (2) part:

1. step revises normalized matrix factorization recommended models, no longer original score data set T is divided into two sub-set T ₁, T ₂, wherein, data set T ₁Be used to learn data set T ₂Be used for judging that study stops, but all data among the data set T all learnt, set the AdaBoost training iteration wheel number I, whenever take turns the number of times R of study, the error range errPermission of permission, and initialization feature vector set;

2. step utilizes normalized matrix factorization recommended models learning training data R time in the iteration of the first round, on the set of eigenvectors that training obtains in the calculation training data respective user to the estimated value of the scoring of project

And obtain itself and actual value r _{U, i}Absolute error, i.e. absolute error

The AbsE value that 3. step calculates in 2. when step is judged the awkward sample of secondary data during greater than errPermission, travels through the total errCount that whole training datas obtains difficult sample ⁿAnd go out the error rate ε of sample by computes ⁿ, wherein | T| representes the number of samples in the training set;

${\mathrm{\&epsiv;}}^n=\frac{{\mathrm{ErrCount}}^n}{\left|T\right|}$ (1) formula;

In the formula: ε ⁿExpression sample error rate, | T| representes the number of samples in the training set, errCount ⁿThe whole training data of expression traversal obtains the sum of difficult sample;

Step is 4. according to the error rate ε that calculates in the step (3) ⁿThe adjustment study number of times of training sample in the next round iteration is specially: when the AbsE of training sample data value during less than errPermission, the study number of times of this sample in the next round iteration is trainTime ^N+1=trainTime ⁿ* ε ⁿ, wherein if trainTime ^N+1Get 1 for＜1, when the AbsE of training sample data value determined that it is difficult sample in (2) formula during greater than errPermission, the study number of times of this sample in the next round iteration did ${\mathrm{TrainTime}}^{n+1}=\frac{{\mathrm{TrainTime}}^n}{{\mathrm{\&epsiv;}}^n},$ Available as shown in the formula expression:

${\mathrm{TrainTime}}^{n+1}=\left\{\begin{array}{cc}\frac{{\mathrm{TrainTime}}^n}{{\mathrm{\&epsiv;}}^n},& (\mathrm{AbsE}\&GreaterEqual;\mathrm{ErrPermission})\\ \mathrm{Max}({\mathrm{TrainTime}}^n*{\mathrm{\&epsiv;}}^n,1),& (\mathrm{AbsE}<\mathrm{ErrPermission})\end{array}\right.$ (2) formula;

In the formula: trainTime ⁿThe study number of times of sample in n next round iteration, trainTime ^N+1The study number of times of sample in n+1 next round iteration, ε ⁿThe error rate of calculating in expression step (3), the absolute error that the substep of AbsE value representation step (2) 2. calculates, errPermission representes the error range that allows;

5. step reduces errPermission with fixed proportion declineRate after, the iteration of a beginning new round, and the study number of times of each sample carries out according to calculating in the step (4) in the iteration that this is taken turns.

3. the dynamic recommend method of a kind of commending system of optimizing based on training set according to claim 1 and 2 is characterized in that: the method for the screening error sample in the said step (3), specific as follows:

Steps A, in step (2), utilizing in the training process of AdaBoost algorithm each to take turns training all need to filter out difficult sample go forward side by side row labels and statistics;

After the AdaBoost iteration training that step B, warp are too much taken turns, training data is traveled through the number of times that each sample of statistics is judged into difficult sample;

Step C, according to clearance delRate, remove from training set judging into the higher sample of difficult sample number of times, thereby obtain new training set.

Images