CN107491557A - A kind of TopN collaborative filtering recommending methods based on difference privacy - Google Patents
A kind of TopN collaborative filtering recommending methods based on difference privacy Download PDFInfo
- Publication number
- CN107491557A CN107491557A CN201710795637.8A CN201710795637A CN107491557A CN 107491557 A CN107491557 A CN 107491557A CN 201710795637 A CN201710795637 A CN 201710795637A CN 107491557 A CN107491557 A CN 107491557A
- Authority
- CN
- China
- Prior art keywords
- user
- topn
- project
- average score
- difference
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000001914 filtration Methods 0.000 title claims abstract description 17
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 238000011084 recovery Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of TopN collaborative filtering recommending methods based on difference privacy, it is related to difference secret protection technology and Collaborative Recommendation technical field.This method is first to original input data Rm*nLaplace noises are reasonably added, user, which scores, after being disturbed recordsThen the similarity between targeted customer and other users is calculated using cosine similarity.On the one hand addition Laplce's noise is recorded by being scored to original user, it is met ε difference secret protection models, ensure that it has higher availability while ensureing and issuing data set personal secrets, be effectively protected the original score data of user.On the other hand; it is improved from proposed algorithm itself; solves the problems, such as the privacy of existing TopN recommended models leakage user; prevent attacker from speculating that it browses record so as to obtain the personal like of user according to the recommendation list of targeted customer; effectively user can precisely be recommended while privacy of user is protected.
Description
Technical field
It is specifically a kind of based on difference privacy the present invention relates to difference secret protection technology and Collaborative Recommendation technical field
TopN collaborative filtering recommending methods.
Background technology
Proposed algorithm is that a kind of historical data to user carries out Knowledge Discovery, and by these knowledge for predicting user couple
The method of related object fancy grade.The application direction of proposed algorithm mainly includes:Predict scoring and utilization of the user to article
TopN recommends personalized recommendation list to user.Wherein, it is that domestic and foreign scholars are universally recognized that TopN, which recommends, has actual meaning
The recommended models of justice.And in numerous proposed algorithms, collaborative filtering is the most commonly used proposed algorithm, based on cooperateing with
The TopN commending systems of filter algorithm are commonly used for major website so as to improve Consumer's Experience.However, collecting substantial amounts of user's letter
While breath carries out data analysis with research, there is also the risk of privacy leakage for individual sensitive information.For example, attacker's root
Speculate that it browses record so as to obtain the personal like of user, or by knowing personal background according to the recommendation list of targeted customer
The research of knowledge can deduce privacy information of user etc..
At present, the method for secret protection based on data distortion (such as k- is anonymous, l- diversity and t- close to model)
The attack under specific background knowledge can be tackled, ganging up against for a variety of background knowledges can not be adapted to, there is significant limitation.This
Outside, the attacker of commending system often possesses the prerequisite for understanding the substantial amounts of historical behavior record of user, that is, has substantial amounts of
Background knowledge, thus using the method based on data distortion effective secret protection can not be carried out to recommendation process.Difference privacy
It is a kind of secret protection model that there is strict mathematical to derive being widely used.The model to analysis mainly by tying
Noise twisting data is added in fruit to reach the purpose of secret protection, and may insure the accuracy of data analysis.Its advantage is
Ignore attacker and possess how many background knowledge, extremely agree with the demand of commending system secret protection.It is it is contemplated that difference is hidden
Private is applied to TopN commending systems, prevents the sensitive information leakage of user in TopN recommendation process.
Scholar notices Privacy Protection in commending system at present but its achievement in research is limited.One kind has
The Collaborative Filtering Recommendation Algorithm of secret protection is proposed that its secret protection is only that the initial data of collection is carried out at random first
Disturbance.Hereafter, there is scholar that difference privacy is applied into commending system, difference privacy is carried out to item-to-item covariance matrixes
Processing, it was demonstrated that it is feasible to carry out difference secret protection to recommendation process.On this basis, the personalized recommendation based on agency
System CRESDUP is suggested, and more has scholar to apply to Privacy Preservation Mechanism in the collaborative filtering recommending based on social networks.So
And the existing document about difference secret protection does not relate to the secret protection of TopN commending systems also.
The content of the invention
The shortcomings that in order to overcome above-mentioned prior art, the present invention provide a kind of TopN collaborative filterings based on difference privacy and pushed away
Method is recommended, is improved in terms of data prediction and proposed algorithm two, essence is realized while user privacy information is protected
Accurate recommendation.
The present invention is realized with following technical scheme:A kind of TopN collaborative filtering recommending methods based on difference privacy,
It is specific as follows including data pre-processing unit, score in predicting unit and TopN recommendation units:
A, data pre-processing unit:User's rating matrix is established by the original scoring record of user, the overall situation after obtaining plus making an uproar
Average score, user's average score and project average score;
B, score in predicting unit:Entered using the global average score after making an uproar, user's average score and project average score is added
The recovery processing of row score in predicting and data;
C, TopN recommendation units:Similarity between user is calculated according to cosine similarity, found and targeted customer's interest
Similar user, recommended candidate list is generated using collaborative filtering, then it is disturbed using Laplce's mechanism, finally
The TopN project that selection targeted customer is most interested in is carried out to be recommended.
Preferably, data pre-processing unit comprises the following steps that:
1) original user rating matrix R first, is established according to original user scoring recordm*n, wherein m is number of users, and n is
Project sum;
2) global average score G then, is calculated, and noise Lap (△ f/ ε are added to it1), global susceptibility △ f=
rmax-rmin, wherein rminAnd rmaxRespectively minimum scoring and highest scoring, ε1For privacy budget, obtain protecting with difference privacy
The project average score DPG of shield;
3) next, the number ICnt for first counting the user of participation project i scorings is used for calculating with difference secret protection
Project average score IAvg [i];
4) if the project average score IAvg [i] with difference secret protection is less than rminOr more than rmaxThen return and perform
Step 3), otherwise directly perform step 5);
5) user's average score UAvg [u] with difference privacy is similarly calculated with step 3), and is controlled in [rmin,
rmax] in the range of.
Preferably, score in predicting unit comprises the following steps that:
1) project average score IAvg [i] and user's average score UAvg [u] meter obtained using data pre-processing unit
Calculate prediction scoring
2) prediction is scoredCarry out recovery processing;
3) score in predicting unit is repeated, until each score data has difference secret protection, the use after being disturbed
Family rating matrix
Preferably, TopN recommendation units comprise the following steps that:
1) basisEstablish item_user to table look-up, and the similarity moment between user is calculated using cosine similarity
Battle array Wui, find other K users similar with target interest, then select this K user seen and N that targeted customer has not seen
Individual project generates recommended candidate list, and uses Laplce's mechanism random disturbances candidate list, and obtaining one has N number of recommendation
Bulleted list Nu;
2) repeat step 1) acquire the recommendation list N with secret protection until each user.
The beneficial effects of the invention are as follows:On the one hand addition Laplce's noise is recorded by being scored to original user, makes it
Meet ε-difference secret protection model, ensure that it has higher can use while ensureing and issuing data set personal secrets
Property, it is effectively protected the original score data of user;On the other hand, it is improved, solves existing from proposed algorithm itself
TopN recommended models reveal the problem of privacy of user, prevent attacker from speculating that it browses note according to the recommendation list of targeted customer
Record effectively can precisely be recommended user so as to obtain the personal like of user while privacy of user is protected.
Brief description of the drawings
Fig. 1 is FB(flow block) of the present invention;
Fig. 2 is the recommendation results by data prediction;
Fig. 3 is N, the timings of K mono-, whether there is the Precision that difference processing TopN recommends and compares;
Fig. 4 is N, and regularly, different types of DP-TopN algorithms Precision compares K mono-.
Embodiment
As shown in figure 1, a kind of TopN collaborative filtering recommending methods based on difference privacy, are used for this method below
ε-difference concept of privacy and the related notion for adding mechanism of making an uproar.
Define 1 ε-difference privacy definition:Given two datasets D and D' at most differ a record, and Range (A) is to appoint
One random data mining algorithm A of meaning span, Pr [Es] represent event EsThe risk that is disclosed of privacy, if algorithm A exists
Any output result O (O ∈ Range (A)) on data set D and D' meets following inequality, then A meets ε-difference privacy:
Pr[A(D)∈O]≤eε×Pr[A(D')∈O]
Wherein, ε represents privacy cost parameter, and ε is smaller, and secret protection degree is higher.If it can be seen that certain from defining 1
Individual data processing algorithm meets ε-difference privacy definition, then the algorithm can effectively protect privacy of user.
It is the major technique for realizing difference secret protection to add noise.Conventional noise addition mechanism is suitable for numeric type
Laplce's noise addition mechanism of data and the index mechanism suitable for nonumeric type data.Set forth herein DP-TopN push away
Recommending algorithm will use Laplce's mechanism to carry out adding pretreatment of making an uproar to initial data, and proposed algorithm is accordingly improved, really
Protect whole recommendation and meet that difference privacy defines.
2 are defined for any one function f:D→RdIf algorithm A output result meets following inequality, A meets
ε-difference privacy.
A (D)=f (D)+<Lap1(△f/ε),...,Lapd(△f/ε)>
Wherein, △ f are global susceptibilitys, determine to add noise content size, Lapi(△ f/ ε) (1≤i≤d) is mutually solely
Vertical Laplace variable, △ f are bigger, and required noise is bigger.
Define 3 global susceptibility definition:Function records in different data sets from two only one and obtains output result
Maximum differential.
Wherein | | | |pNorm is represented, the global susceptibility of function is determined in itself by function.Some functions have less
Global susceptibility, if the global susceptibility of counting function is 1, average, the global susceptibility of the function such as median it is then larger.
Define the definition of 4 index mechanism:Give a scoring functions u:(D × O) → R, if algorithm A meets following equalities,
Then A meets ε-difference privacy.
Wherein, △ u are scoring functions u (D, r) global sensitiveness, and marking is higher, and the probability for being chosen output is bigger.
Based on a kind of defined above, DP-TopN proposed algorithms based on difference secret protection model of present invention proposition, including
Data pre-processing unit, score in predicting unit and TopN recommendation units.
Data pre-processing unit:User's rating matrix is established by the original scoring record of user, the overall situation after asking plus making an uproar is averaged
Scoring, user's average score and project average score, are comprised the following steps that:
1) original user rating matrix R first, is established according to original user scoring recordm*n, wherein m is number of users, and n is
Project sum;
2) global average score G then, is calculated, and noise Lap (△ f/ ε are added to it1), global susceptibility △ f=
rmax-rmin, wherein rminAnd rmaxRespectively minimum scoring and highest scoring, ε1For privacy budget, obtain protecting with difference privacy
The project average score DPG of shield;
3) next, the number ICnt for first counting the user of participation project i scorings is used for calculating with difference secret protection
Project average score IAvg [i];
4) if the project average score IAvg [i] with difference secret protection is less than rminOr more than rmaxThen return and perform
Step 3), otherwise directly perform step 5);
5) user's average score UAvg [u] with difference privacy is similarly calculated with step 3), and is controlled in [rmin,
rmax] in the range of.
Score in predicting unit:Carried out using the global average score after making an uproar, user's average score and project average score is added
The recovery processing of score in predicting and data, is comprised the following steps that:
1) project average score IAvg [i] and user's average score UAvg [u] meter obtained using data pre-processing unit
Calculate prediction scoring
2) prediction is scoredCarry out recovery processing;
3) score in predicting unit is repeated, until each score data has difference secret protection, the use after being disturbed
Family rating matrix
TopN recommendation units:Similarity between user is calculated according to cosine similarity, found and targeted customer's interest phase
As user, using collaborative filtering generate recommended candidate list, then it is disturbed using Laplce's mechanism, it is most laggard
Row is chosen the TopN project that targeted customer is most interested in and recommended, and comprises the following steps that:
1) basisEstablish item_user to table look-up, and the similarity moment between user is calculated using cosine similarity
Battle array Wui, find other K users similar with target interest, then select this K user seen and N that targeted customer has not seen
Individual project generates recommended candidate list, and uses Laplce's mechanism random disturbances candidate list, and obtaining one has N number of recommendation
Bulleted list Nu;
2) repeat step 1) acquire the recommendation list N with secret protection until each user.
Abbreviation DP-TopN (Differentialprivacy TopN) algorithm of the present invention, from above step, DP-
The data pre-processing unit of TopN algorithms, step 1), original user rating matrix R is established according to original user scoring recordm*n,
The storage and calculating of data, step 2)~5 can be facilitated), the statistic information of correlation is calculated, and Laplce is added to it and made an uproar
Sound, control in [rmin,rmax] in the range of, it is ensured that the availability and privacy of data;Utilize in score in predicting unit and obtain
To statistic carry out score in predicting and carrying out recovery processing to it, further increase the availability of data.Recommend in TopN
Unit, user's similarity matrix W is calculated using cosine similarityui, other K users similar with targeted customer's interest are found,
Then select this K user to see and the generation recommended candidate list of N number of project that targeted customer has not seen, and use Laplce
Mechanism random disturbances candidate list, obtaining one has N number of recommended project list Nu.It follows that this method meets that difference is hidden
Private model, TopN commending system privacy leakage problems can be efficiently solved, be protected while ensureing issue data set personal secrets
Having demonstrate,proved it has higher availability.In addition, the parameter ε in DP-TopN algorithms represents privacy budget, by adjusting ε, Ke Yi
Reach balance between protection privacy of user and accurate recommendation.
Experimental verification and interpretation of result
Experimental data set and environment:In order to assess DP-TopN algorithms, experiment uses and comes from http://
Www.grouplens.org/node/ True Data collection MovieLens.The data set includes tri- numbers of 100K, 1M and 10M altogether
According to collection, select size to be tested for 1M data set herein, and be 10 parts by data set random division, take 9 parts every time as instruction
Practice integrate, 1 part be test set progress 10- folding cross validation.Experimental situation is:Intel Core i5-4210U CPU 2.40GHz
Processor, 8GB internal memories, Windows10 operating systems, emulation experiment is carried out using python language.Following obtained experimental data
It is the average value that its result is taken on the basis of 10 experiments of operation.
Analysis of experimental results
Fig. 2 is accuracy rate curve map (the parameter ε by data prediction and without data prediction1=ε2=0.25, βm
=15, βp=20).There it can be seen that data prediction serves the effect that protection is specifically scored, enchancement factor (base is excluded
This does not influence the degree of accuracy of recommendation results) find:The accuracy ratio that User-based TopN based on user recommend is based on project
Item-based TopN accuracy it is high.
Fig. 3 is K=60, during N=10, the accuracy rate Precision of DP-TopN proposed algorithms and two kinds of basic recommendation algorithms
Contrast.For Mostpopular algorithms to select most popular N number of project every time, user-based TopN algorithms are to select time in figure
The top n project of list is selected, this paper DP-TopN algorithms select N number of project by difference secret protection.As seen from the figure, when
ε3The obvious high accuracy rate with MostPopular algorithms of accuracy rate that >=5, DP-TopN recommend.And as 5≤ε3When≤15, DP-
TopN accuracy rate then falls between can be in the equalization point between protection privacy of user and effectively recommendation.Because ε3More
Small, secret protection degree is bigger, and the noise of addition is also bigger, and the availability of data is lower.
Fig. 4 gives fixed N=10, during K=60, DP-TopN algorithms based on user (user-based) and based on item
The accuracy rate Precision comparison diagrams of the DP-TopN algorithms of mesh (item-based).As shown in Figure 2, recommendation process is without difference
During privacy processing, the accuracy rate that project-based TopN recommends is recommended higher than the TopN based on user.And as shown in Figure 4, when pushing away
Process is recommended after the processing of difference privacy, project-based TopN recommendation results accuracy rate is still higher than the TopN based on user and pushed away
Result is recommended, recommendation effect is also more excellent.Therefore, project-based DP-TopN algorithms are recommended herein to be used for improving Consumer's Experience
Degree.
In summary, DP-TopN meets ε-difference secret protection model, can effectively protect the original scoring number of user
According to.Solve the problems, such as the privacy of existing TopN recommended models leakage user, prevent attacker from being arranged according to the recommendation of targeted customer
Table speculates that it browses record so as to obtain the personal like of user, can still can be to user while privacy of user is protected
Efficiently and accurately recommended.
Claims (4)
- A kind of 1. TopN collaborative filtering recommending methods based on difference privacy, it is characterised in that:Including data pre-processing unit, comment Divide predicting unit and TopN recommendation units, it is specific as follows:A, data pre-processing unit:User's rating matrix is established by the original scoring record of user, the overall situation after obtaining plus making an uproar is averaged Scoring, user's average score and project average score;B, score in predicting unit:Commented using the global average score after making an uproar, user's average score and project average score is added Prediction and the recovery of data is divided to handle;C, TopN recommendation units:Similarity between user is calculated according to cosine similarity, found similar to targeted customer's interest User, using collaborative filtering generate recommended candidate list, then it is disturbed using Laplce's mechanism, finally carried out The TopN project that targeted customer is most interested in is chosen to be recommended.
- A kind of 2. TopN collaborative filtering recommending methods based on difference privacy according to claim 1, it is characterised in that:Number Data preprocess unit comprises the following steps that:1) original user rating matrix R first, is established according to original user scoring recordm*n, wherein m is number of users, and n is project Sum;2) global average score G then, is calculated, and noise Lap (△ f/ ε are added to it1), global susceptibility △ f=rmax-rmin, Wherein rminAnd rmaxRespectively minimum scoring and highest scoring, ε1For privacy budget, the project with difference secret protection is obtained Average score DPG;3) next, the number ICnt for first counting the user of participation project i scorings is used for calculating the project with difference secret protection Average score IAvg [i];4) if the project average score IAvg [i] with difference secret protection is less than rminOr more than rmaxThen return and perform step 3) step 5), is otherwise directly performed;5) user's average score UAvg [u] with difference privacy is similarly calculated with step 3), and is controlled in [rmin,rmax] model In enclosing.
- A kind of 3. TopN collaborative filtering recommending methods based on difference privacy according to claim 2, it is characterised in that:Comment Point predicting unit comprises the following steps that:1) the project average score IAvg [i] and user's average score UAvg [u] obtained using data pre-processing unit calculates pre- Test and appraisal point2) prediction is scoredCarry out recovery processing;3) score in predicting unit is repeated, until each score data has difference secret protection, the user after being disturbed comments Sub-matrix
- A kind of 4. TopN collaborative filtering recommending methods based on difference privacy according to claim 3, it is characterised in that: TopN recommendation units comprise the following steps that:1) basisEstablish item_user to table look-up, and the similarity matrix W between user is calculated using cosine similarityui, Find other K users similar with target interest, then select this K user seen and N number of item that targeted customer has not seen Mesh generates recommended candidate list, and uses Laplce's mechanism random disturbances candidate list, and obtaining one has N number of recommended project List Nu;2) repeat step 1) acquire the recommendation list N with secret protection until each user.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710795637.8A CN107491557A (en) | 2017-09-06 | 2017-09-06 | A kind of TopN collaborative filtering recommending methods based on difference privacy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710795637.8A CN107491557A (en) | 2017-09-06 | 2017-09-06 | A kind of TopN collaborative filtering recommending methods based on difference privacy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107491557A true CN107491557A (en) | 2017-12-19 |
Family
ID=60652293
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710795637.8A Pending CN107491557A (en) | 2017-09-06 | 2017-09-06 | A kind of TopN collaborative filtering recommending methods based on difference privacy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107491557A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108681938A (en) * | 2018-05-07 | 2018-10-19 | 常熟市第人民医院 | A kind of collaborative filtering Top-N methods based on difference privacy and user activity |
CN109195098A (en) * | 2018-06-20 | 2019-01-11 | 苏州大学 | A kind of shared collaborative filtering method based on difference privacy |
CN109977324A (en) * | 2019-03-28 | 2019-07-05 | 南京邮电大学 | A kind of point of interest method for digging and system |
CN111259260A (en) * | 2020-03-30 | 2020-06-09 | 九江学院 | Privacy protection method in personalized recommendation based on sorting classification |
CN112214793A (en) * | 2020-09-30 | 2021-01-12 | 南京邮电大学 | Random walk model recommendation method based on fusion of differential privacy |
CN112487473A (en) * | 2020-11-11 | 2021-03-12 | 东南大学 | Differential privacy protection method for collusion inference attack in collaborative filtering |
CN113204793A (en) * | 2021-06-09 | 2021-08-03 | 辽宁工程技术大学 | Recommendation method based on personalized differential privacy protection |
CN113360945A (en) * | 2021-06-29 | 2021-09-07 | 招商局金融科技有限公司 | Noise adding method, device, equipment and medium based on differential privacy |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110064221A1 (en) * | 2009-09-11 | 2011-03-17 | Microsoft Corporation | Differential privacy preserving recommendation |
EP2629248A1 (en) * | 2012-02-15 | 2013-08-21 | Thomson Licensing | Method of creating content recommendations based on user ratings of content with improved user privacy |
CN104156388A (en) * | 2014-06-26 | 2014-11-19 | 西安邮电大学 | Collaborative filtering recommendation method based on trustful privacy maintenance in personalized search |
CN104537025A (en) * | 2014-12-19 | 2015-04-22 | 北京邮电大学 | Frequent sequence mining method |
CN106557654A (en) * | 2016-11-16 | 2017-04-05 | 中山大学 | A kind of collaborative filtering based on difference privacy technology |
CN106610970A (en) * | 2015-10-21 | 2017-05-03 | 上海文广互动电视有限公司 | Collaborative filtering-based content recommendation system and method |
CN106909607A (en) * | 2017-01-06 | 2017-06-30 | 南京邮电大学 | A kind of collaborative filtering group recommending method based on random perturbation technology |
-
2017
- 2017-09-06 CN CN201710795637.8A patent/CN107491557A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110064221A1 (en) * | 2009-09-11 | 2011-03-17 | Microsoft Corporation | Differential privacy preserving recommendation |
EP2629248A1 (en) * | 2012-02-15 | 2013-08-21 | Thomson Licensing | Method of creating content recommendations based on user ratings of content with improved user privacy |
CN104156388A (en) * | 2014-06-26 | 2014-11-19 | 西安邮电大学 | Collaborative filtering recommendation method based on trustful privacy maintenance in personalized search |
CN104537025A (en) * | 2014-12-19 | 2015-04-22 | 北京邮电大学 | Frequent sequence mining method |
CN106610970A (en) * | 2015-10-21 | 2017-05-03 | 上海文广互动电视有限公司 | Collaborative filtering-based content recommendation system and method |
CN106557654A (en) * | 2016-11-16 | 2017-04-05 | 中山大学 | A kind of collaborative filtering based on difference privacy technology |
CN106909607A (en) * | 2017-01-06 | 2017-06-30 | 南京邮电大学 | A kind of collaborative filtering group recommending method based on random perturbation technology |
Non-Patent Citations (1)
Title |
---|
鲜征征 等: "差分隐私保护在推荐系统中的应用研究", 《计算机应用研究》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108681938A (en) * | 2018-05-07 | 2018-10-19 | 常熟市第人民医院 | A kind of collaborative filtering Top-N methods based on difference privacy and user activity |
CN109195098A (en) * | 2018-06-20 | 2019-01-11 | 苏州大学 | A kind of shared collaborative filtering method based on difference privacy |
CN109195098B (en) * | 2018-06-20 | 2020-11-03 | 苏州大学 | Shared collaborative filtering method based on differential privacy |
CN109977324B (en) * | 2019-03-28 | 2022-09-16 | 南京邮电大学 | Interest point mining method and system |
CN109977324A (en) * | 2019-03-28 | 2019-07-05 | 南京邮电大学 | A kind of point of interest method for digging and system |
CN111259260A (en) * | 2020-03-30 | 2020-06-09 | 九江学院 | Privacy protection method in personalized recommendation based on sorting classification |
CN111259260B (en) * | 2020-03-30 | 2023-06-02 | 九江学院 | Privacy protection method in personalized recommendation based on sorting classification |
CN112214793A (en) * | 2020-09-30 | 2021-01-12 | 南京邮电大学 | Random walk model recommendation method based on fusion of differential privacy |
CN112487473A (en) * | 2020-11-11 | 2021-03-12 | 东南大学 | Differential privacy protection method for collusion inference attack in collaborative filtering |
CN112487473B (en) * | 2020-11-11 | 2024-06-07 | 东南大学 | Differential privacy protection method for collusion inference attack in collaborative filtering |
CN113204793A (en) * | 2021-06-09 | 2021-08-03 | 辽宁工程技术大学 | Recommendation method based on personalized differential privacy protection |
CN113360945A (en) * | 2021-06-29 | 2021-09-07 | 招商局金融科技有限公司 | Noise adding method, device, equipment and medium based on differential privacy |
CN113360945B (en) * | 2021-06-29 | 2023-04-07 | 招商局金融科技有限公司 | Noise adding method, device, equipment and medium based on differential privacy |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107491557A (en) | A kind of TopN collaborative filtering recommending methods based on difference privacy | |
Becker et al. | Weights and importance in composite indicators: mind the gap | |
Pinto et al. | Using early view patterns to predict the popularity of youtube videos | |
Zhang et al. | Latent factor transition for dynamic collaborative filtering | |
CN108920503A (en) | A kind of micro- video personalized recommendation algorithm based on social networks degree of belief | |
Shang | Forecasting intraday S&P 500 index returns: A functional time series approach | |
CN106126549A (en) | A kind of community's trust recommendation method decomposed based on probability matrix and system thereof | |
CN113688252B (en) | Safe cross-domain recommendation method based on multi-feature collaborative knowledge graph and blockchain | |
CN103500228A (en) | Similarity measuring method improved through collaborative filtering recommendation algorithm | |
CN108470052B (en) | Anti-trust attack recommendation algorithm based on matrix completion | |
CN105760649A (en) | Big-data-oriented creditability measuring method | |
CN111125517A (en) | Implicit matrix decomposition recommendation method based on differential privacy and time perception | |
Gao et al. | Item anomaly detection based on dynamic partition for time series in recommender systems | |
Inuzuka et al. | Recommendation system based on prediction of user preference changes | |
CN110598126A (en) | Cross-social network user identity recognition method based on behavior habits | |
Ji et al. | Portfolio diversification strategy via tail‐dependence clustering and ARMA‐GARCH Vine Copula approach | |
Yang et al. | A differential privacy framework for collaborative filtering | |
Aliaj et al. | Nowcasting inflation with Lasso‐regularized vector autoregressions and mixed frequency data | |
Iacus et al. | How to exploit big data from social networks: a subjective well-being indicator via Twitter | |
CN105447087A (en) | Video recommendation method and apparatus | |
CN110727867A (en) | Semantic entity recommendation method based on fuzzy mechanism | |
Su et al. | An item-based music recommender system using music content similarity | |
CN113204793A (en) | Recommendation method based on personalized differential privacy protection | |
Wiedermann | Decisions concerning the direction of effects in linear regression models using fourth central moments | |
Zhang et al. | A Robust Collaborative Recommendation Algorithm Based on Least Median Squares Estimator. |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171219 |
|
WD01 | Invention patent application deemed withdrawn after publication |