CN109840702A - A kind of new projects' collaborative recommendation method based on multi-core integration - Google Patents
A kind of new projects' collaborative recommendation method based on multi-core integration Download PDFInfo
- Publication number
- CN109840702A CN109840702A CN201910070714.2A CN201910070714A CN109840702A CN 109840702 A CN109840702 A CN 109840702A CN 201910070714 A CN201910070714 A CN 201910070714A CN 109840702 A CN109840702 A CN 109840702A
- Authority
- CN
- China
- Prior art keywords
- user
- project
- formula
- similarity
- new projects
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The cold start-up proposed algorithm based on information attribute value that the present invention relates to a kind of, it is therefore an objective to which the shortage of data in new commodity recommendation is solved the problems, such as by the collaborative filtering using multicore Weighted Fusion.The algorithm has determined commodity incidence relation between attribute space in such a way that multicore weights, and recommends new projects to realize to user.Wherein, the Multiple Kernel Learning algorithm is to be weighted summation for each kernel function on the basis of existing kernel function learning algorithm to improve accuracy of the algorithm in complex data environment;The attributes similarity is to make the user calculated to the preference-score of commodity projection with more explanatory by calculating the similitude between commodity on attribute;The weight is optimized by the learning method of stochastic gradient descent.By means of the invention it is possible to learn a kind of item similarity measure for describing user preference according to the attribute information of commodity, to effectively improve the precision of new projects' recommendation.
Description
Technical field
The present invention relates to machine learning field more particularly to a kind of new projects' collaborative recommendation methods based on multi-core integration.
Background technique
With the high speed development of social economy, people can face more and more commodity, information and service, cause user without
From being selected.In order to solve the problems, such as this information overload, recommender system is come into being.What is be most widely used is based on item
Purpose collaborative filtering, core concept are the similar articles for recommending his (she) to like before to user, can not need to use
Personalized recommendation is carried out under the scene of family other information.However, this method will be difficult to if lacking the history evaluation information of project
Effective recommendation is generated, here it is new projects to recommend problem (being also called project cold start-up problem).
The proposed algorithm being commonly cold-started has following three kinds:
(1) using scoring intermediate value, average value, mode etc. fills up vacancy, but the score value of non-scoring item will be complete
Equally, therefore the confidence level of this method is not high;
(2) using the similitude between the method calculating project of neighbour, but most of arest neighbors project searched out is all
It is not commented excessively by active user, therefore, stability is bad;
(3) method for using similarity, the similitude between calculating project, but because between similarity and user preference
It is difficult to reach effective matching, therefore the adaptability of algorithm is not strong.
Summary of the invention
In view of the above problems, the present invention in order to further increase new projects recommendation accuracy, it is proposed that one kind is based on
New projects' collaborative recommendation method of multi-core integration reaches the performance that can effectively promote the cold start-up of project.One kind being based on multicore
New projects' collaborative recommendation method of fusion characterized by comprising
Step 1 establishes data attribute information collection, and the data attribute information collection includes: user's id information, item id letter
The attribute information of breath, score information and project;
Step 2 extracts score information greater than 0 corresponding user's id information and project id information, forms project-user
Group randomly selects any project-user group and is calculated;
Step 3, calculated separately according to the attribute information of project Pearson came similarity between project, cosine similarity and
Jaccard similarity establishes Pearson came core, cosine kernel and Jaccard core respectively;Wherein neighbour's set in Pearson came similarity
It is the intersection of sets collection of the user of two projects, following piecewise function is, to the judgement of n, (n is element in set in the case of 3 kinds
Number),
Step 4, the weighted sum for calculating Pearson came core, cosine kernel and Jaccard core three basic kernel functions, calculation formula is such as
Under:
In formula, simiThe kernel function of corresponding respective type, wiIt is that each kernel function accounts for whole weight respectively, weight and is
1, and each weight is both greater than equal to 0.
Step 5 determines that the neighbours of current project v collect SN, specifically includes
5.1) between calculating project attribute similitude
5.2) all scoring item collection S of active user u comment are found out
5.3) the preceding K project in Item Sets S with current project v attributes similarity is chosen, the neighbours of current project v are constituted
Collect SNvIf calculated less than K by current number, the preceding K for otherwise taking similarity big.
Step 6 calculates prediction scoring of the user u to project v, and calculation formula is as follows:
SN in formulavIndicate the K nearest neighborhoods of item v, wherein used the attribute information of project given to determine
The neighbour of project gathers (j ∈ SNv), and it is formula that user u, which has history scoring sim (v, j) to project j,
In multicore linear combination;rjuIt is score value of the user u on project j;
Step 7 optimizes loss function using the algorithm of stochastic gradient descent, and loss formula is
P (v, u) is prediction scoring in formula, and y (v, u) is true value;Solve loss letter
Number the following steps are included:
7.1) loss function is expressed as to the form of gradient:
7.2) it calculatesValue, calculating process are as follows:
7.3) abbreviation enables
7.4) vector a, b are obtained, brings vector a, b into formulaIt obtainsValue;
Step 8 calculates loss such as formula:
Step 9, the processing that weight w is normalized, it is ensured that the w after each iterationiAdduction is 1, and wiBoth greater than 0;
Step 10, according to formula w(n+1)=wn- α η (0 < α≤0.1) updates w.
Further, further includes: the recommendation of new projects, specific steps are carried out to user are as follows:
The neighbour for 2.a) finding out each new projects gathers { SNv, wherein v is the set of new projects;
It is similar with Jaccard 2.b) to calculate the Pearson came similarity gathered each neighbour of active user, cosine similarity
Degree;
2.c) bring the value of the final weight w of calculating into formulaObtain new projects and neighbour's collection
Weight similitude;
2.d) finally according to formulaIt obtains user to score to the prediction of each new projects, selection
User recommends in new projects corresponding to prediction scoring maximum value.
Further, the Pearson came similarity is used to measure the linear relationship between two vectors, i.e.,
In formula, rx,p, ry,pUser p is respectively indicated to item x, the scoring of item y.Respectively indicate user on an x and item y
Collect the mean value of scoring;I is the user's set of an x and item y to score jointly;
Cosine similarity measures the similitude between vector by the cosine value of angle between two vectors of calculating,
In formula, rxAnd ryRespectively indicate the scoring vector of an x and item y;| | | | indicate that vector field homoemorphism is long;
Jaccard related coefficient is the ratio between two projects between intersection and union, i.e.,
In formula, X and Y respectively indicates the scoring collection of an x and item y.
In conclusion the present invention provides a kind of new projects' collaborative recommendation method based on multi-core integration, passes through kernel function
A nonlinear transformation is carried out, the input space is mapped to high-dimensional feature space, to be carried out linearly in linear space
It calculates, simplifies the complexity of calculating, improve the feasibility of algorithm;It is merged by kernel function, improves algorithm and answered in difference
With adaptability, the raising accuracy rate under environment;By the relationship of attribute information between multi-core integration calculating project, solve to be based on project
Collaborative filtering can not handle the problem of new projects, while provide similarity evaluation criterion between a kind of user of optimization.
The utility model has the advantages that the Multiple Kernel Learning algorithm, is on the basis of existing kernel function learning algorithm, is each core
Function is weighted summation to improve accuracy of the algorithm in complex data environment;The attributes similarity is logical
The similitude calculated between commodity on attribute is crossed, makes the user calculated to the preference-score of commodity projection with more explanatory;Institute
The weight stated is optimized by the learning method of stochastic gradient descent.By means of the invention it is possible to be believed according to the attribute of commodity
Breath learns a kind of item similarity measure for describing user preference, to effectively improve the precision of new projects' recommendation.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, the accompanying drawings in the following description is only some embodiments recorded in the present invention, for this field
For those of ordinary skill, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of new projects' collaborative recommendation method embodiment flow chart based on multi-core integration provided by the invention;
Fig. 2 is a kind of new projects' collaborative recommendation method simplified diagram based on multi-core integration provided by the invention.
Specific embodiment
New projects' collaborative recommendation method embodiment based on multi-core integration that The present invention gives a kind of, in order to lead this technology
The personnel in domain more fully understand the technical solution in the embodiment of the present invention, and enable the above objects, features and advantages of the present invention
It is enough more obvious and easy to understand, technical solution in the present invention is described in further detail with reference to the accompanying drawing:
New projects' collaborative recommendation method embodiment based on multi-core integration that the present invention provides a kind of, as shown in Figure 1, packet
It includes:
S101, step 1 establish data attribute information collection, and the data attribute information collection includes: user's id information, project
The attribute information of id information, score information and project;
S102, step 2 extract score information greater than 0 corresponding user's id information and project id information, form project-
User group randomly selects any project-user group and is calculated;
S103, step 3 calculate separately Pearson came similarity, cosine similarity between project according to the attribute information of project
With Jaccard similarity, Pearson came core, cosine kernel and Jaccard core are established respectively;Wherein neighbour's collection in Pearson came similarity
Conjunction is the intersection of sets collection of the user of two projects, and following piecewise function is in the case of 3 kinds, and to the judgement of n, (n is member in set
The number of element),
S104, step 4, the weighted sum for calculating Pearson came core, cosine kernel and Jaccard core three basic kernel functions, calculate
Formula is as follows:
In formula, simiThe kernel function of corresponding respective type, wiIt is that each kernel function accounts for whole weight respectively, weight and is
1, and each weight is both greater than equal to 0.
S105, step 5 determine that the neighbours of current project v collect SN, specifically include
5.1) between calculating project attribute similitude
5.2) all scoring item collection S of active user u comment are found out
5.3) the preceding K project in Item Sets S with current project v attributes similarity, i.e. neighbours' collection of current project v are chosen
SNv(if being calculated less than K by current number, be more than preceding K for taking similarity big)
S106, step 6 calculate prediction scoring of the user u to project v, and calculation formula is as follows:
SN in formulavIndicate the K nearest neighbours of item v;Sim (v, j) is formulaIn multicore line
Property combination;rjuIt is score value of the user u on project j;
S107, step 7 optimize loss function using the algorithm of stochastic gradient descent, and loss formula is
P (v, u) is prediction scoring in formula, and y (v, u) is true value;Solve loss letter
Number the following steps are included:
7.1) loss function is expressed as to the form of gradient:
7.2) it calculatesValue, calculating process are as follows:
7.3) abbreviation enables
7.4) vector a, b are obtained, brings vector a, b into formulaIt obtainsValue;
S108, step 8, loss such as formula are calculated:
S109, step 9, the processing that weight w is normalized, it is ensured that the w after each iterationiAdduction is 1, and wiIt is all big
In 0;
S1010, step 10, according to formula w(n+1)=wn- α η (0 < α≤0.1) updates w.
Wherein, total the number of iterations is 200000 times, and iterating to calculate interval times every time is 1000 times, and step-length is
0.0002, if the w of current iterationiLess than 0, it is returned to last iteration, and continue to calculate, until penalty values tend to
Stablize, and records the value of final weight w.
Further, further includes: the recommendation of new projects, specific steps are carried out to user are as follows:
The neighbour for 2.a) finding out each new projects gathers { SNv, wherein v is the set of new projects;
It is similar with Jaccard 2.b) to calculate the Pearson came similarity gathered each neighbour of active user, cosine similarity
Degree;
2.c) bring the value of the final weight w of calculating into formulaObtain new projects and neighbour's collection
Weight similitude;
2.d) finally according to formulaIt obtains user to score to the prediction of each new projects, selection
User recommends in new projects corresponding to prediction scoring maximum value.
Further, the Pearson came similarity is used to measure the linear relationship between two vectors, i.e.,
In formula, rx,p, ry,pUser p is respectively indicated to item x, the scoring of item y.Respectively indicate user on an x and item y
Collect the mean value of scoring;I is the user's set of an x and item y to score jointly;
Cosine similarity measures the similitude between vector by the cosine value of angle between two vectors of calculating,
In formula, rxAnd ryRespectively indicate the scoring vector of an x and item y;| | | | indicate that vector field homoemorphism is long;
Jaccard related coefficient is the ratio between two projects between intersection and union, i.e.,
In formula, X and Y respectively indicates the scoring collection of an x and item y.
In conclusion the present invention provides a kind of new projects' collaborative recommendation method based on multi-core integration, passes through kernel function
A nonlinear transformation is carried out, the input space is mapped to high-dimensional feature space, to be carried out linearly in linear space
It calculates, simplifies the complexity of calculating, improve the feasibility of algorithm;It is merged by kernel function, improves algorithm and answered in difference
With adaptability, the raising accuracy rate under environment;By the relationship of attribute information between multi-core integration calculating project, solve to be based on project
Collaborative filtering can not handle the problem of new projects, while provide similarity evaluation criterion between a kind of user of optimization.
Above embodiments are to illustrative and not limiting technical solution of the present invention.Appointing for spirit and scope of the invention is not departed from
What modification or part replacement, are intended to be within the scope of the claims of the invention.
Claims (3)
1. a kind of new projects' collaborative recommendation method based on multi-core integration characterized by comprising
Step 1 establishes data attribute information collection, and the data attribute information collection includes: user's id information, project id information, comments
Divide the attribute information of information and project;
Step 2 extracts score information greater than 0 corresponding user's id information and project id information, forms project-user group, with
Machine is chosen any project-user group and is calculated;
Step 3 calculates separately Pearson came similarity, cosine similarity and Jaccard between project according to the attribute information of project
Similarity establishes Pearson came core, cosine kernel and Jaccard core respectively;Wherein neighbour's set in Pearson came similarity is two
The intersection of sets collection of the user of project, following piecewise function be in the case of 3 kinds, and to the judgement of n, (n is of element in set
Number),
Step 4, the weighted sum for calculating Pearson came core, cosine kernel and Jaccard core three basic kernel functions, calculation formula are as follows:
In formula, simiThe kernel function of corresponding respective type, wiIt is that each kernel function accounts for whole weight respectively, weight and be 1, and
And each weight is both greater than equal to 0.
Step 5 determines that the neighbours of current project v collect SN, specifically includes
5.1) between calculating project attribute similitude
5.2) all scoring item collection S of active user u comment are found out
5.3) the preceding K project in Item Sets S with current project v attributes similarity is chosen, neighbours' collection of current project v is constituted
SNvIf if calculated less than K by current number, the preceding K for otherwise taking similarity big
Step 6 calculates prediction scoring of the user u to project v, and calculation formula is as follows:
SN in formulavIndicate the K nearest neighbours of item v;Sim (v, j) is formulaIn linear group of multicore
It closes;rju(0≤rjuIt≤5) is score value of the user u on project j;
Step 7 optimizes loss function using the algorithm of stochastic gradient descent, loses formula are as follows:
P (v, u) is prediction scoring in formula, and y (v, u) is really to score;Solve loss function
The following steps are included:
7.1) loss function is expressed as to the form of gradient:
7.2) it calculatesValue, calculating process are as follows:
7.3) abbreviation enables
7.4) vector a, b are obtained, brings vector a, b into formulaIt obtainsValue;
Step 8 calculates loss such as formula:
Step 9, the processing that weight w is normalized, it is ensured that the w after each iterationiAdduction is 1, and wiBoth greater than it is equal to 0;
Step 10, according to formula w(n+1)=wn- α η (0 < α≤0.1) updates w.
2. a kind of new projects' collaborative recommendation method based on multi-core integration as described in claim 1, which is characterized in that also wrap
It includes: carrying out the recommendation of new projects, specific steps to user are as follows:
The neighbour for 2.a) finding out each new projects gathers { SNv, wherein v is new projects;
2.b) calculate Pearson came similarity, cosine similarity and Jaccard similarity that active user gathers each neighbour;
2.c) bring the value of the final weight w of calculating into formulaObtain the weighting of new projects and neighbour's collection
Similitude;
2.d) finally according to formulaIt obtains user to score to the prediction of each new projects, selection prediction
User recommends in new projects corresponding to scoring maximum value.
3. a kind of new projects' collaborative recommendation method based on multi-core integration as described in claim 1, which is characterized in that the skin
Your inferior similarity is used to measure the linear relationship between two vectors, it may be assumed that
In formula, rx,p, ry,pUser p is respectively indicated to item x, the scoring of item y.User's collection on an x and item y is respectively indicated to comment
The mean value divided;I is the user's set of an x and item y to score jointly;
Cosine similarity measures the similitude between vector by the cosine value of angle between two vectors of calculating,
In formula, rxAnd ryRespectively indicate the scoring vector of an x and item y;| | | | indicate that vector field homoemorphism is long;
Jaccard related coefficient is the ratio between two projects between intersection and union, it may be assumed that
In formula, X and Y respectively indicates the scoring collection of an x and item y.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910070714.2A CN109840702A (en) | 2019-01-24 | 2019-01-24 | A kind of new projects' collaborative recommendation method based on multi-core integration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910070714.2A CN109840702A (en) | 2019-01-24 | 2019-01-24 | A kind of new projects' collaborative recommendation method based on multi-core integration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109840702A true CN109840702A (en) | 2019-06-04 |
Family
ID=66884108
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910070714.2A Pending CN109840702A (en) | 2019-01-24 | 2019-01-24 | A kind of new projects' collaborative recommendation method based on multi-core integration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109840702A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110309424A (en) * | 2019-07-04 | 2019-10-08 | 东北大学 | A kind of socialization recommended method based on Rough clustering |
CN110472071A (en) * | 2019-07-03 | 2019-11-19 | 中移(杭州)信息技术有限公司 | Multimedia file recommendation method, device, the network equipment and storage medium |
CN111046280A (en) * | 2019-12-02 | 2020-04-21 | 哈尔滨工程大学 | Cross-domain recommendation method for application FM |
CN112381112A (en) * | 2020-10-16 | 2021-02-19 | 华南理工大学 | User identity recognition method and system based on multi-mode item set of user data |
CN116911963A (en) * | 2023-09-14 | 2023-10-20 | 南京龟兔赛跑软件研究院有限公司 | Data-driven pesticide byproduct transaction management method and cloud platform |
-
2019
- 2019-01-24 CN CN201910070714.2A patent/CN109840702A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472071A (en) * | 2019-07-03 | 2019-11-19 | 中移(杭州)信息技术有限公司 | Multimedia file recommendation method, device, the network equipment and storage medium |
CN110309424A (en) * | 2019-07-04 | 2019-10-08 | 东北大学 | A kind of socialization recommended method based on Rough clustering |
CN111046280A (en) * | 2019-12-02 | 2020-04-21 | 哈尔滨工程大学 | Cross-domain recommendation method for application FM |
CN111046280B (en) * | 2019-12-02 | 2023-12-12 | 哈尔滨工程大学 | Cross-domain recommendation method applying FM |
CN112381112A (en) * | 2020-10-16 | 2021-02-19 | 华南理工大学 | User identity recognition method and system based on multi-mode item set of user data |
CN112381112B (en) * | 2020-10-16 | 2023-11-07 | 华南理工大学 | User identity recognition method and system based on multi-mode item set of user data |
CN116911963A (en) * | 2023-09-14 | 2023-10-20 | 南京龟兔赛跑软件研究院有限公司 | Data-driven pesticide byproduct transaction management method and cloud platform |
CN116911963B (en) * | 2023-09-14 | 2023-12-19 | 南京龟兔赛跑软件研究院有限公司 | Data-driven pesticide byproduct transaction management method and cloud platform system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109840702A (en) | A kind of new projects' collaborative recommendation method based on multi-core integration | |
CN103886001A (en) | Personalized commodity recommendation system | |
CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
CN109360057A (en) | Information-pushing method, device, computer equipment and storage medium | |
CN104462383A (en) | Movie recommendation method based on feedback of users' various behaviors | |
CN108460082A (en) | A kind of recommendation method and device, electronic equipment | |
KR101356948B1 (en) | User interest inference method and system in sns using topics on social activities with neighbors | |
CN111125453A (en) | Opinion leader role identification method in social network based on subgraph isomorphism and storage medium | |
CN113592609B (en) | Personalized clothing collocation recommendation method and system utilizing time factors | |
CN111428127B (en) | Personalized event recommendation method and system integrating theme matching and bidirectional preference | |
CN108876536A (en) | Collaborative filtering recommending method based on arest neighbors information | |
CN107145541B (en) | Social network recommendation model construction method based on hypergraph structure | |
CN110134783A (en) | Method, apparatus, equipment and the medium of personalized recommendation | |
CN110598126B (en) | Cross-social network user identity recognition method based on behavior habits | |
CN116244513A (en) | Random group POI recommendation method, system, equipment and storage medium | |
CN110059257B (en) | Project recommendation method based on score correction | |
CN108415928A (en) | A kind of book recommendation method and system based on weighted blend k- nearest neighbor algorithms | |
Ge et al. | A new improved forecasting method integrated fuzzy time series with the exponential smoothing method | |
CN109857928A (en) | User preference prediction technique based on polynary credit evaluation | |
CN109544261A (en) | A kind of intelligent perception motivational techniques based on diffusion and the quality of data | |
CN105989005A (en) | Information pushing method and device | |
Ogundele et al. | SoCaST: exploiting social, categorical and spatio-temporal preferences for personalized event recommendations | |
CN116204723A (en) | Social recommendation method based on dynamic hypergraph representation learning | |
CN104954873A (en) | Intelligent television video customizing method and intelligent television video customizing system | |
CN114417166A (en) | Continuous interest point recommendation method based on behavior sequence and dynamic social influence |
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 |
Application publication date: 20190604 |
|
RJ01 | Rejection of invention patent application after publication |