CN108984551A - A kind of recommended method and system based on the multi-class soft cluster of joint - Google Patents
A kind of recommended method and system based on the multi-class soft cluster of joint Download PDFInfo
- Publication number
- CN108984551A CN108984551A CN201710400994.XA CN201710400994A CN108984551A CN 108984551 A CN108984551 A CN 108984551A CN 201710400994 A CN201710400994 A CN 201710400994A CN 108984551 A CN108984551 A CN 108984551A
- Authority
- CN
- China
- Prior art keywords
- matrix
- user
- article
- cluster
- prediction
- 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 38
- 239000011159 matrix material Substances 0.000 claims abstract description 204
- 230000002452 interceptive effect Effects 0.000 claims abstract description 25
- 230000006870 function Effects 0.000 claims description 43
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 6
- 235000013399 edible fruits Nutrition 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000002349 favourable effect Effects 0.000 abstract description 4
- 238000001914 filtration Methods 0.000 description 11
- 238000000354 decomposition reaction Methods 0.000 description 9
- 238000011156 evaluation Methods 0.000 description 5
- 230000006399 behavior Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Development Economics (AREA)
- Physics & Mathematics (AREA)
- Finance (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of recommended methods and system based on the multi-class soft cluster of joint, wherein the recommended method includes: to obtain user-article interactive information, according to the user-article interactive information building rating matrix and classification matrix;Multi-class soft clustering processing is carried out to the rating matrix and the classification matrix, obtains multi-class soft cluster result;The prediction of user preferences degree is carried out to the multi-class soft cluster result using weighting Non-negative Matrix Factorization, obtains prediction result;Recommend the prediction highest article of score to user according to the prediction result.In embodiments of the present invention, can according to user to the favorable rating of article to score in predicting is carried out, article is recommended to user according to score in predicting, prediction accuracy is higher.
Description
Technical field
The present invention relates to data analysis technique field more particularly to a kind of recommended methods based on the multi-class soft cluster of joint
And system.
Background technique
Personalized recommendation is applied to the every aspect of our lives at present, it can be from a large amount of article, article, film, sound
The interested part of user user is filtered out in happy, network etc..Recommender system more popular at present includes Amazon etc. each
Electric business platform, music recommender system, movie system.One good recommender system can be from for recommender system owner and user
In be benefited.
Can generally it be divided into according to the different recommender systems of recommended method following several:
1. being based on the recommendation of content (Content-based)
2. being based on the recommendation of collaborative filtering (Collaborative Filtering-Based)
Mixed type 3. (Hybrid) recommender system.
What content-based recommendation algorithm was completed is original collaborative filtering task.They are gone using data processing technique
The neighborhood of a user is established, then usually used is that the use for predicting not score is gone in the weighted sum scored
Family-article pair.The most important link of content-based recommendation algorithm is exactly the calculating of similarity.Wherein more famous place
Reason method includes Pearson correlation coefficient, vector similarity and their some expansions.
It is also most successful proposed algorithm that recommender system based on collaborative filtering, which is most widely used at present,.Be based on content
Recommender system unlike, it is not necessarily to handle the attribute of user and article, it is only necessary to know user-article interactive information.
User-article interactive information can be explicit, be also possible to implicit.Explicit information, including such as user is to article
Marking etc., hiding information can be the behavior of user, such as purchase, number of clicks, label etc..The letter of two kinds of forms
Ceasing generally may be stored in a big but very sparse data matrix, wherein that row representative is user, arrange representative is
Article.In fact, most proposed algorithms based on collaborative filtering is operated to this matrix.
Due to privacy of user protection, item property are various etc., based entirely on content recommendation using less, it is more at present
Number proposed algorithms are studied toward the direction of the recommendation based on collaborative filtering, are also had and are much combined based on content and based on cooperateing with
Filter the algorithm of both direction.Recommendation based on collaborative filtering, and multiple subclass can be divided into, it is based on user (User- respectively
Based recommendation), be based on commodity (Item-Based) recommendation, based on social networks (Social-Based) recommendation and
Recommendation etc. based on model (Model-based).In classification of the above based on the recommender system of collaborative filtering, it is based on
The recommendation of model refers to using the existing data of system, study one model of building, and then is recommended using the model, such as
It can be matrix decomposition, be also possible to using model conversations such as Bayes classifier, decision tree, neural networks be classification problem,
Pretreated result is either carried out to data based on clustering technique.
The classical recommendation based on collaborative filtering, there is two main problems to be solved:
1. Deta sparseness.In actual life application, the quantity of user and commodity is very huge, the quotient that user's evaluation is crossed
Product quantity be for whole commodity amount it is considerably less, the quantity for buying commentator in the user of identical commodity is also seldom, comments
Sub-matrix is extremely sparse.If doing just with rating matrix based on user or based on the collaborative filtering recommending of commodity, prediction
Recommendation effect can be poor.
It include the cold start-up of user and commodity 2. being cold-started.New user does not have historical behavior data, so can not learn
The hobby of the user simultaneously carries out personalized recommendation.New commodity is also difficult to pass through due to no relevant user's score data
The mode of collaborative filtering is recommended.
3. the dynamic expansion of data.In the recommender system of reality, rating matrix will not be unalterable, and new user is to existing
The evaluation of article, existing subscriber are likely to out the evaluation of new article or existing subscriber to the New Appraisement of existing article
It is existing.
The problem of for Sparse, that current some research work are mainly taken is the spy that matrix decomposition carries out vector
Sign is extracted, or is pre-processed to data using clustering technique.User is cold-started, what is generally taken is to utilize user
Information, is perhaps inquired by guided bone or using global popular recommendation process, after user produces behavioral data again into
Personalized recommendation of the row based on collaborative filtering.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, and it is soft poly- based on multi-class joint that the present invention provides one kind
The recommended method and system of class, can according to user to the favorable rating of article to carrying out score in predicting, according to score in predicting to
User recommends article, and prediction accuracy is higher.
In order to solve the above-mentioned technical problem, the embodiment of the invention provides a kind of recommendations based on the multi-class soft cluster of joint
Method, the recommended method include:
User-article interactive information is obtained, according to the user-article interactive information building rating matrix and classification square
Battle array;
Multi-class soft clustering processing is carried out to the rating matrix and the classification matrix, obtains multi-class soft cluster knot
Fruit;
The prediction of user preferences degree is carried out to the multi-class soft cluster result using weighting Non-negative Matrix Factorization, obtains prediction
As a result;
Recommend the prediction highest article of score to user according to the prediction result.
It is preferably, described according to the user-article interactive information building rating matrix and classification matrix, comprising:
User-article relationship, user-user relationship, article-article is obtained according to the user-article interactive information to close
System;
The scoring is constructed according to the user-article relationship, the user-user relationship, the article-article relationship
Matrix and the classification matrix;
The classification matrix includes user's classification matrix and taxonomy of goods matrix.
It is preferably, described that multi-class soft clustering processing is carried out to the rating matrix and the classification matrix, comprising:
According to the shared low order space matrix of the rating matrix and classification matrix building;
Multi-class cluster iterative calculation processing is carried out to the low order space matrix using objective function is minimized, obtains institute
State objective function iterative value;
It is compared using the objective function iterative value with iteration threshold, if the objective function iterative value is less than iteration
Threshold value then stops iteration, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;
Each row of the cluster low order space matrix is normalized, multi-class soft cluster result is obtained.
Preferably, described pre- to the multi-class soft cluster result progress user preferences degree using weighting Non-negative Matrix Factorization
It surveys, comprising:
Subclass matrix is carried out to the multi-class soft cluster result to mark off, and obtains subclass matrix;
Non-negative Matrix Factorization prediction processing is carried out to the subclass matrix, obtains subclass Matrix prediction result;
Enter to calculate to the subclass Matrix prediction result using the weighted sum, obtains prediction result.
It is preferably, described that the prediction highest article of score is recommended to user according to the prediction result, comprising:
The prediction result that will acquire carries out prediction score from low sequence is got, and obtains ranking results;
Highest preceding 10 articles that will sort recommend user.
In addition, the embodiment of the invention also provides a kind of recommender system based on the multi-class soft cluster of joint, the recommendation
System includes:
Matrix constructs module: for obtaining user-article interactive information, being constructed according to the user-article interactive information
Rating matrix and classification matrix;
Cluster module: it for carrying out multi-class soft clustering processing to the rating matrix and the classification matrix, obtains more
The soft cluster result of classification;
Prediction module: for carrying out user preferences degree to the multi-class soft cluster result using weighting Non-negative Matrix Factorization
Prediction obtains prediction result;
Recommending module: for recommending the prediction highest article of score to user according to the prediction result.
Preferably, the matrix building module includes:
Relation acquisition unit: for obtaining user-article relationship, user-user according to the user-article interactive information
Relationship, article-article relationship;
Matrix construction unit: for according to the user-article relationship, the user-user relationship, the article-object
Product relationship constructs the rating matrix and the classification matrix;
The classification matrix includes user's classification matrix and taxonomy of goods matrix.
Preferably, the cluster module includes:
Second matrix construction unit: for the low order space shared according to the rating matrix and classification matrix building
Matrix;
Cluster iteration unit: it changes for carrying out multi-class cluster to the low order space matrix using minimum objective function
For calculation processing, the objective function iterative value is obtained;
Judging unit: for being compared using the objective function iterative value with iteration threshold, if the objective function
Iterative value is less than iteration threshold, then stops iteration, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;
Normalization unit: it is normalized for each row to the cluster low order space matrix, obtains multiclass
Not soft cluster result.
Preferably, the prediction module includes:
Matrix division unit: it marks off for carrying out subclass matrix to the multi-class soft cluster result, obtains subclass
Matrix;
Subclass Matrix prediction unit: for carrying out Non-negative Matrix Factorization prediction processing to the subclass matrix, subclass is obtained
Matrix prediction result;
Weight calculation unit: it for entering to calculate to the subclass Matrix prediction result using the weighted sum, obtains
Prediction result.
Preferably, the recommending module includes:
Sequencing unit: the prediction result for will acquire carries out prediction score from low sequence is got, and obtains sequence knot
Fruit;
Recommendation unit: for that will sort, highest preceding 10 articles recommend user.
In embodiments of the present invention, by using multi-class soft cluster, user and taxonomy of goods are overlapped to multiple
In subclass, user and article share identical Cluster space, carry out multi-class soft cluster, the discovery of actually one interest domain
Process assigns to the article of a certain type and the user for liking the type in one classification, for example by electronic product and likes electricity
The user of sub- product is referred to a subclass, and daily necessity and the user for liking daily necessity are attributed to a subclass;Meanwhile one
A client can reside in multiple subclasses, has both liked daily class article, has also liked electronic product;The subclass of generation can be very well
The interest domain at earth's surface requisition family;By these subclasses, submatrix is generated from original matrix, and matrix decomposition algorithm is applied to
On the one hand these submatrixs greatly reduce the degree of rarefication of matrix, improve matrix decomposition estimated performance;On the other hand, due to making
It is the scoring of the user of same interest domain to predict, reliability is higher than the scoring for considering all users, because of purchase
The evaluation of the uninterested article of user, for user-interest domain article score in predicting has certain interference;It can basis
Multi-class soft cluster, to score in predicting is carried out, recommends article to user according to score in predicting to the favorable rating of user and article,
Prediction accuracy is higher.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it is clear that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the method flow schematic diagram of the recommended method based on the multi-class soft cluster of joint in the embodiment of the present invention;
Fig. 2 is the system structure composition signal of the recommender system based on the multi-class soft cluster of joint in the embodiment of the present invention
Figure.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts all other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the method flow schematic diagram of the recommended method based on the multi-class soft cluster of joint in the embodiment of the present invention,
As shown in Figure 1, the recommended method includes:
S11: obtaining user-article interactive information, according to the user-article interactive information building rating matrix and classification
Matrix;
S12: multi-class soft clustering processing is carried out to the rating matrix and the classification matrix, obtains multi-class soft cluster
As a result;
S13: the prediction of user preferences degree is carried out to the multi-class soft cluster result using weighting Non-negative Matrix Factorization, is obtained
Prediction result;
S14: the prediction highest article of score is recommended to user according to the prediction result.
S11 is described further:
User-article relationship, user-user relationship, article-article is obtained according to the user-article interactive information to close
System;The rating matrix is constructed according to the user-article relationship, the user-user relationship, the article-article relationship
With the classification matrix;The classification matrix includes user's classification matrix and taxonomy of goods matrix.
In user-article interactive information, there is three kinds of different types of internal relations: user-article relationship, uses
Family-customer relationship, article-article relationship.
Assuming that having n user and m article, furthermore we only have user-article rating matrix at known information, wherein TijThat represent is scoring of the user i to article j, uiThat represent is i-th of user, yjWhat is represented is j-th
Article.
Our target is simultaneously to assign to user and article in c subclass, and wherein user/article can appear in multiple
In subclass.
The result of MCoC can use a classification matrixIt indicates, wherein PijWhat is represented is an element
(user or article) is to the indicated value of j-th of subclass, Pij∈[0,1].If Pij> 0, which just represents this i-th of element, belongs to j-th
Subclass, Pij=0 is not belonging to;PijSize then represent the associated weight for belonging to the subclass, wherein all weights of every a line
The sum of size is 1.If the subclass quantity where fixed each element, for example be k (1 < k < c), then every a line just has k a non-
Zero.It is classical joint clustering problem if k=1, each object exists only in one of class.Joint cluster result square
Battle array can be indicated with following form:
WhereinIt is user's classification matrix,It is taxonomy of goods matrix.
S12 is described further:
According to the shared low order space matrix of the rating matrix and classification matrix building;Using minimum target letter
It is several that multi-class cluster iterative calculation processing is carried out to the low order space matrix, obtain the objective function iterative value;Using institute
It states objective function iterative value to be compared with iteration threshold, if the objective function iterative value is less than iteration threshold, stopping changes
In generation, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;To each of the cluster low order space matrix
Row is normalized, and obtains multi-class soft cluster result.
It is the shared low order space matrix of building first, will considers these three relationships simultaneously, and propose the expression of loss function,
Loss function minimization problem is converted by clustering problem:
1) user-article relationship
As soon as if a user is made that high scoring to article, then may more appear in the same subclass simultaneously
In;In order to make these elements with strong tie put together, following loss function is proposed to user-article relationship:
Wherein qiIt is the i-th row of Q, rjIt is the jth row of R,For the degree diagonal matrix of user,It is the degree diagonal matrix of article
This loss function is very readily comprehensible, because it is known that only user-Item Information, minimize this damage
Losing function means to take the high user-article pair to score, in matrix of consequence P, the instruction of the instruction vector sum article j of user i
Vector must be very close.
2) user-user relationship
What this step was done is to be modeled using rating matrix T to user-user relationship.Firstly the need of calculating two two users
Between similarity, Euclidean distance, the equidistant calculation method of Pearson correlation coefficient may be used herein;It uses
The similar loss function calculation method of user-article above, has:
WhereinThis loss function means the high user of two similarities, has in matrix of consequence
There is more like instruction vector.
3) article-article relationship
What this step was done is to be modeled using rating matrix T to article-article relationship, way and user-use above
Family relationship modeling is similar;Firstly the need of the similarity calculated between article two-by-two, there is loss function:
WhereinThis loss function means the high article of two similarities, has in matrix of consequence
There is more like instruction vector.
4) objective function
In summary three loss functions obtain the loss function for solving classification matrix P:
∈ (P)=∈ (Q, R)+∈ (Q)+∈ (R)
s.t.
|Pi|=k, i=1 ..., (m+n)
Parameter c is the quantity of all subclasses oneself, and k is that each user or article allow existing subclass quantity (1≤k
≤ c), | | it is constraint base, represents the quantity of the nonzero value of a vector.
Due to these constraints, it is difficult to resolve certainly very much so that minimizing objective function, so taking an approximate method
It goes to obtain an approximate solution;Way is similar with spectral clustering, is divided into two stages
A) all users and article are mapped to a shared low order space, construct shared low order space matrix:
This step is the r dimension approximate expression of P to be obtained;Loss function abbreviation is obtained first:
Wherein LQFor the degree diagonal matrix of userRemaining weight matrix between 0 and user
The difference of W, LRFor the degree diagonal matrix of articleThe difference of remaining weight matrix W between 0 and user;
Matrix S:
The sum of Tr () calculating matrix diagonal line;The approximate solution of matrix of consequence is compressed to a r dimension space, and item will be constrained
Part loosens, and is converted into and minimizes following objective function:
By solving the r minimal eigenvalue of MX=λ X, X=[x is obtained1..., xr]。
B) to categorical clusters iteration:
Cluster can use two kinds of cluster modes, be hard cluster and soft cluster respectively;In embodiments of the present invention, use is soft
Cluster, an object may alternatively appear in multiple classifications, therefore fuzzy c-means is selected to be clustered;Way be minimize with
Lower objective function:
Wherein PijIt is the relationship of (user or the commodity) and subclass j of object i in matrix of consequence P, vjIt is the class center of the subclass,
D () is distance function, and l is the parameter of blurring degree;Iteration updates P and V:
Calculating target function after each iteration, stops iteration when the value that objective function improves is less than a threshold value;
After stopping iteration, to every a line of P, only retain maximum k element, and be normalized, guarantee every a line and be 1;
In in embodiments of the present invention, iteration threshold can be 0.5, and specific threshold value can be formulated according to the demand of user, in we
It is not strictly required in the embodiment of face.
S13 is described further:
Subclass matrix is carried out to the multi-class soft cluster result to mark off, and obtains subclass matrix;To the subclass square
Battle array carries out Non-negative Matrix Factorization prediction processing, obtains subclass Matrix prediction result;Using the weighted sum to the subclass square
Battle array prediction result enters to calculate, and obtains prediction result.
By the subclass marked off above, some small matrixes are obtained from original rating matrix.In each submatrix
It is weighted NMF matrix (nonnegative matrix) decomposition.
One simple improvement is made to Weighted N MF matrix;In general, the initial value of NMF is frequently with random initializtion side
Method;However, it was found that there is such phenomenons: user A scoring is most lower in having scoring, and user B scoring is most higher
In the case of, random initializtion carry out WNMF result be often A other scoring higher than B other scoring, reason mainly with
Machine initialization is similar for the angle of moment of a vector to the interest vector of all user/articles, therefore works as user i counterpart
When product j has lower scoring, matrix decomposition tends to the corresponding user vector of orthogonalization and article vector.Therefore, initial
When changing two matrixes of matrix decomposition, consider that user and article have the average score of scoring.
After being predicted in each submatrix, final prediction score is calculated using the method for weighted sum:
Wherein, Pr (ui, yj, k) and what is represented is that user i scores to the prediction of article j in subclass k.
S14 is described further:
The prediction result that will acquire carries out prediction score from low sequence is got, and obtains ranking results;To sort highest
Preceding 10 articles recommend user.
In embodiments of the present invention, marking and queuing is carried out using from high sort method on earth, then according to ranking results
Sequence is recommended into user near 10 preceding articles;Here sort method can be to be varied, can be according to user's
Hobby selects different sortords.
Fig. 2 is the system structure composition signal of the recommender system based on the multi-class soft cluster of joint in the embodiment of the present invention
Figure, as shown in Fig. 2, the recommender system includes:
Matrix constructs module 11: for obtaining user-article interactive information, according to the user-article interactive information structure
Build rating matrix and classification matrix;
Cluster module 12: it for carrying out multi-class soft clustering processing to the rating matrix and the classification matrix, obtains
Multi-class soft cluster result;
Prediction module 13: for carrying out user preferences to the multi-class soft cluster result using weighting Non-negative Matrix Factorization
Degree prediction, obtains prediction result;
Recommending module 14: for recommending the prediction highest article of score to user according to the prediction result.
Preferably, the matrix building module 11 includes:
Relation acquisition unit: for obtaining user-article relationship, user-user according to the user-article interactive information
Relationship, article-article relationship;
Matrix construction unit: for according to the user-article relationship, the user-user relationship, the article-object
Product relationship constructs the rating matrix and the classification matrix;
The classification matrix includes user's classification matrix and taxonomy of goods matrix.
Preferably, the cluster module 12 includes:
Second matrix construction unit: for the low order space shared according to the rating matrix and classification matrix building
Matrix;
Cluster iteration unit: it changes for carrying out multi-class cluster to the low order space matrix using minimum objective function
For calculation processing, the objective function iterative value is obtained;
Judging unit: for being compared using the objective function iterative value with iteration threshold, if the objective function
Iterative value is less than iteration threshold, then stops iteration, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;
Normalization unit: it is normalized for each row to the cluster low order space matrix, obtains multiclass
Not soft cluster result.
Preferably, the prediction module 13 includes:
Matrix division unit: it marks off for carrying out subclass matrix to the multi-class soft cluster result, obtains subclass
Matrix;
Subclass Matrix prediction unit: for carrying out Non-negative Matrix Factorization prediction processing to the subclass matrix, subclass is obtained
Matrix prediction result;
Weight calculation unit: it for entering to calculate to the subclass Matrix prediction result using the weighted sum, obtains
Prediction result.
Preferably, the recommending module 14 includes:
Sequencing unit: the prediction result for will acquire carries out prediction score from low sequence is got, and obtains sequence knot
Fruit;
Recommendation unit: for that will sort, highest preceding 10 articles recommend user.
Specifically, the working principle of the system related functions module of the embodiment of the present invention can be found in the correlation of embodiment of the method
Description, which is not described herein again.
In embodiments of the present invention, by using multi-class soft cluster, user and taxonomy of goods are overlapped to multiple
In subclass, user and article share identical Cluster space, carry out multi-class soft cluster, the discovery of actually one interest domain
Process assigns to the article of a certain type and the user for liking the type in one classification, for example by electronic product and likes electricity
The user of sub- product is referred to a subclass, and daily necessity and the user for liking daily necessity are attributed to a subclass;Meanwhile one
A client can reside in multiple subclasses, has both liked daily class article, has also liked electronic product;The subclass of generation can be very well
The interest domain at earth's surface requisition family;By these subclasses, submatrix is generated from original matrix, and matrix decomposition algorithm is applied to
On the one hand these submatrixs greatly reduce the degree of rarefication of matrix, improve matrix decomposition estimated performance;On the other hand, due to making
It is the scoring of the user of same interest domain to predict, reliability is higher than the scoring for considering all users, because of purchase
The evaluation of the uninterested article of user, for user-interest domain article score in predicting has certain interference;It can basis
Multi-class soft cluster, to score in predicting is carried out, recommends article to user according to score in predicting to the favorable rating of user and article,
Prediction accuracy is higher.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
In addition, being provided for the embodiments of the invention a kind of recommended method based on the soft cluster of multi-class joint above and being
System is described in detail, herein should use a specific example illustrates the principle and implementation of the invention, with
The explanation of upper embodiment is merely used to help understand method and its core concept of the invention;Meanwhile for the general of this field
Technical staff, according to the thought of the present invention, there will be changes in the specific implementation manner and application range, in conclusion
The contents of this specification are not to be construed as limiting the invention.
Claims (10)
1. a kind of recommended method based on the soft cluster of multi-class joint, which is characterized in that the recommended method includes:
User-article interactive information is obtained, according to the user-article interactive information building rating matrix and classification matrix;
Multi-class soft clustering processing is carried out to the rating matrix and the classification matrix, obtains multi-class soft cluster result;
The prediction of user preferences degree is carried out to the multi-class soft cluster result using weighting Non-negative Matrix Factorization, obtains prediction knot
Fruit;
Recommend the prediction highest article of score to user according to the prediction result.
2. the recommended method according to claim 1 based on the multi-class soft cluster of joint, which is characterized in that described according to institute
State user-article interactive information building rating matrix and classification matrix, comprising:
User-article relationship, user-user relationship, article-article relationship are obtained according to the user-article interactive information;
The rating matrix is constructed according to the user-article relationship, the user-user relationship, the article-article relationship
With the classification matrix;
The classification matrix includes user's classification matrix and taxonomy of goods matrix.
3. the recommended method according to claim 1 based on the multi-class soft cluster of joint, which is characterized in that described to described
Rating matrix and the classification matrix carry out multi-class soft clustering processing, comprising:
According to the shared low order space matrix of the rating matrix and classification matrix building;
Multi-class cluster iterative calculation processing is carried out to the low order space matrix using objective function is minimized, obtains the mesh
Scalar functions iterative value;
It is compared using the objective function iterative value with iteration threshold, if the objective function iterative value is less than iteration threshold
Value then stops iteration, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;
Each row of the cluster low order space matrix is normalized, multi-class soft cluster result is obtained.
4. the recommended method according to claim 1 based on the multi-class soft cluster of joint, which is characterized in that described use adds
It weighs Non-negative Matrix Factorization and the prediction of user preferences degree is carried out to the multi-class soft cluster result, comprising:
Subclass matrix is carried out to the multi-class soft cluster result to mark off, and obtains subclass matrix;
Non-negative Matrix Factorization prediction processing is carried out to the subclass matrix, obtains subclass Matrix prediction result;
Enter to calculate to the subclass Matrix prediction result using the weighted sum, obtains prediction result.
5. the recommended method according to claim 1 based on the multi-class soft cluster of joint, which is characterized in that described according to institute
It states prediction result and recommends the prediction highest article of score to user, comprising:
The prediction result that will acquire carries out prediction score from low sequence is got, and obtains ranking results;
Highest preceding 10 articles that will sort recommend user.
6. a kind of recommender system based on the soft cluster of multi-class joint, which is characterized in that the recommender system includes:
Matrix constructs module: for obtaining user-article interactive information, according to the user-article interactive information building scoring
Matrix and classification matrix;
Cluster module: it for carrying out multi-class soft clustering processing to the rating matrix and the classification matrix, obtains multi-class
Soft cluster result;
Prediction module: pre- for carrying out user preferences degree to the multi-class soft cluster result using weighting Non-negative Matrix Factorization
It surveys, obtains prediction result;
Recommending module: for recommending the prediction highest article of score to user according to the prediction result.
7. the recommender system according to claim 6 based on the multi-class soft cluster of joint, which is characterized in that the matrix structure
Modeling block includes:
Relation acquisition unit: for obtaining user-article relationship according to the user-article interactive information, user-user closes
System, article-article relationship;
Matrix construction unit: for being closed according to the user-article relationship, the user-user relationship, the article-article
System constructs the rating matrix and the classification matrix;
The classification matrix includes user's classification matrix and taxonomy of goods matrix.
8. the recommender system according to claim 6 based on the multi-class soft cluster of joint, which is characterized in that the cluster mould
Block includes:
Second matrix construction unit: for the low order spatial moment shared according to the rating matrix and classification matrix building
Battle array;
Cluster iteration unit: by carrying out based on multi-class cluster iteration using minimum objective function to the low order space matrix
Calculation processing, obtains the objective function iterative value;
Judging unit: for being compared using the objective function iterative value with iteration threshold, if the objective function iteration
Value is less than iteration threshold, then stops iteration, obtains cluster low order space matrix;Conversely, then continuing to iterate to calculate;
Normalization unit: it is normalized, obtains multi-class soft for each row to the cluster low order space matrix
Cluster result.
9. the recommender system according to claim 6 based on the multi-class soft cluster of joint, which is characterized in that the prediction mould
Block includes:
Matrix division unit: it marks off for carrying out subclass matrix to the multi-class soft cluster result, obtains subclass matrix;
Subclass Matrix prediction unit: for carrying out Non-negative Matrix Factorization prediction processing to the subclass matrix, subclass matrix is obtained
Prediction result;
Weight calculation unit: for entering to calculate to the subclass Matrix prediction result using the weighted sum, prediction is obtained
As a result.
10. the recommender system according to claim 6 based on the multi-class soft cluster of joint, which is characterized in that the recommendation
Module includes:
Sequencing unit: the prediction result for will acquire carries out prediction score from low sequence is got, and obtains ranking results;
Recommendation unit: for that will sort, highest preceding 10 articles recommend user.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710400994.XA CN108984551A (en) | 2017-05-31 | 2017-05-31 | A kind of recommended method and system based on the multi-class soft cluster of joint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710400994.XA CN108984551A (en) | 2017-05-31 | 2017-05-31 | A kind of recommended method and system based on the multi-class soft cluster of joint |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108984551A true CN108984551A (en) | 2018-12-11 |
Family
ID=64501422
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710400994.XA Pending CN108984551A (en) | 2017-05-31 | 2017-05-31 | A kind of recommended method and system based on the multi-class soft cluster of joint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108984551A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110930259A (en) * | 2019-11-15 | 2020-03-27 | 安徽海汇金融投资集团有限公司 | Creditor right recommendation method and system based on mixed strategy |
CN111859155A (en) * | 2020-08-04 | 2020-10-30 | 深圳前海微众银行股份有限公司 | Item recommendation method, equipment and computer-readable storage medium |
CN112365325A (en) * | 2020-12-11 | 2021-02-12 | 南京魏东电子商务有限公司 | Intelligent advertisement recommendation method based on feature recognition of big data platform |
CN112955883A (en) * | 2018-12-29 | 2021-06-11 | 深圳市欢太科技有限公司 | Application recommendation method and device, server and computer-readable storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110295846A1 (en) * | 2010-05-28 | 2011-12-01 | Apple Inc. | Predicting content and context performance based on performance history of users |
CN103514496A (en) * | 2012-06-21 | 2014-01-15 | 腾讯科技(深圳)有限公司 | Method and system for processing recommended target software |
CN106202480A (en) * | 2016-07-19 | 2016-12-07 | 淮阴工学院 | A kind of network behavior based on K means and LDA bi-directional verification custom clustering method |
CN106296425A (en) * | 2016-08-03 | 2017-01-04 | 哈尔滨工业大学深圳研究生院 | Attributed graph clustering method based on Weight associating Non-negative Matrix Factorization and system |
CN106446947A (en) * | 2016-09-22 | 2017-02-22 | 华南理工大学 | High-dimension data soft and hard clustering integration method based on random subspace |
-
2017
- 2017-05-31 CN CN201710400994.XA patent/CN108984551A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110295846A1 (en) * | 2010-05-28 | 2011-12-01 | Apple Inc. | Predicting content and context performance based on performance history of users |
CN103514496A (en) * | 2012-06-21 | 2014-01-15 | 腾讯科技(深圳)有限公司 | Method and system for processing recommended target software |
CN106202480A (en) * | 2016-07-19 | 2016-12-07 | 淮阴工学院 | A kind of network behavior based on K means and LDA bi-directional verification custom clustering method |
CN106296425A (en) * | 2016-08-03 | 2017-01-04 | 哈尔滨工业大学深圳研究生院 | Attributed graph clustering method based on Weight associating Non-negative Matrix Factorization and system |
CN106446947A (en) * | 2016-09-22 | 2017-02-22 | 华南理工大学 | High-dimension data soft and hard clustering integration method based on random subspace |
Non-Patent Citations (1)
Title |
---|
吴湖等: "两阶段联合聚类协同过滤算法", 《软件学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112955883A (en) * | 2018-12-29 | 2021-06-11 | 深圳市欢太科技有限公司 | Application recommendation method and device, server and computer-readable storage medium |
CN110930259A (en) * | 2019-11-15 | 2020-03-27 | 安徽海汇金融投资集团有限公司 | Creditor right recommendation method and system based on mixed strategy |
CN111859155A (en) * | 2020-08-04 | 2020-10-30 | 深圳前海微众银行股份有限公司 | Item recommendation method, equipment and computer-readable storage medium |
CN112365325A (en) * | 2020-12-11 | 2021-02-12 | 南京魏东电子商务有限公司 | Intelligent advertisement recommendation method based on feature recognition of big data platform |
CN112365325B (en) * | 2020-12-11 | 2021-09-24 | 东方财富信息股份有限公司 | Intelligent advertisement recommendation method based on feature recognition of big data platform |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yu et al. | A cross-domain collaborative filtering algorithm with expanding user and item features via the latent factor space of auxiliary domains | |
CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
CN110674407B (en) | Hybrid recommendation method based on graph convolution neural network | |
CN104731962B (en) | Friend recommendation method and system based on similar corporations in a kind of social networks | |
Bhagat et al. | Node classification in social networks | |
US7774227B2 (en) | Method and system utilizing online analytical processing (OLAP) for making predictions about business locations | |
CN107545471B (en) | Big data intelligent recommendation method based on Gaussian mixture | |
Mustafa et al. | Collaborative filtering: Techniques and applications | |
EP2377080A1 (en) | Machine optimization devices, methods, and systems | |
CN105787767A (en) | Method and system for obtaining advertisement click-through rate pre-estimation model | |
CN108984551A (en) | A kind of recommended method and system based on the multi-class soft cluster of joint | |
CN109840833A (en) | Bayes's collaborative filtering recommending method | |
Sánchez et al. | Efficient algorithms for a robust modularity-driven clustering of attributed graphs | |
Singh et al. | Generalized correlation coefficients of intuitionistic fuzzy sets with application to MAGDM and clustering analysis | |
Deodhar et al. | A framework for simultaneous co-clustering and learning from complex data | |
de Aguiar Neto et al. | Pre-processing approaches for collaborative filtering based on hierarchical clustering | |
Meena et al. | Group recommender systems–an evolutionary approach based on multi-expert system for consensus | |
Shang et al. | Demographic inference via knowledge transfer in cross-domain recommender systems | |
Hazrati et al. | Entity representation for pairwise collaborative ranking using restricted Boltzmann machine | |
CN114925268A (en) | Recommendation method and system based on graph neural network, electronic device and computer readable medium | |
Yin et al. | A survey of learning-based methods for cold-start, social recommendation, and data sparsity in e-commerce recommendation systems | |
He et al. | SRSH: a social recommender system based on Hadoop | |
Idrissi et al. | A new hybrid-enhanced recommender system for mitigating cold start issues | |
Ifada et al. | Do-rank: DCG optimization for learning-to-rank in tag-based item recommendation systems | |
Bourhim et al. | Towards a Novel Graph-based collaborative filtering approach for recommendation systems |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181211 |