CN106610970A - Collaborative filtering-based content recommendation system and method - Google Patents
Collaborative filtering-based content recommendation system and method Download PDFInfo
- Publication number
- CN106610970A CN106610970A CN201510688543.1A CN201510688543A CN106610970A CN 106610970 A CN106610970 A CN 106610970A CN 201510688543 A CN201510688543 A CN 201510688543A CN 106610970 A CN106610970 A CN 106610970A
- Authority
- CN
- China
- Prior art keywords
- project
- scoring
- algorithm
- similarity
- score
- 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
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
Abstract
The invention discloses a collaborative filtering-based content recommendation system and method. The system comprises a data preprocessing module, an algorithm mixing module, and a result generation module, wherein the algorithm mixing module further comprises an algorithm selection unit, a weighting similarity-based collaborative recommendation algorithm unit, a balance score prediction mechanism-based collaborative recommendation algorithm unit, a score filling-based mixed recommendation algorithm unit, and a collaborative recommendation algorithm unit using score time characteristics; and the algorithm mixing module inputs preprocessed data to the weighting similarity-based collaborative recommendation algorithm unit, the balance score prediction mechanism-based collaborative recommendation algorithm unit, the score filling-based mixed recommendation algorithm unit, and the collaborative recommendation algorithm unit using the score time characteristics, and outputs an algorithm result to the result generation module. According to the system and the method, the sparsity problem and the concept drift problem in the recommendation system are better solved.
Description
Technical field
The present invention relates to proposed algorithm and its system, more particularly, it relates to a kind of be based on collaborative filtering
Content recommendation system and method.
Background technology
With the prevalence of Internet technology, the life of people increasingly be unable to do without network, more and more
People selects to be entertained or done shopping on network.In the face of the user's request of expanding day, and increasingly
How rich in natural resources, fully understand the demand of user, quickly finds oneself needs for user
Resource, become attract user a powerful measure.Based on this demand, personalized recommendation technology is gradually
Gradually it is taken seriously, nowadays comes into the stage of a ripe development.
Personalized recommendation technology, is an importance of user behavior analysis technology, briefly,
It is exactly a process that the resource that he may be interested is found for user.In order to realize the money of personalization
Recommend in source, it is necessary to " understanding " user, " understanding " resource.By dividing subscriber data and a large amount of historical behaviors
Analysis, therefrom draws the interest knowledge of user, then in a reasonable way representing user interest.
Resource is organized simultaneously, chooses reasonable expression way to express resource characteristic.Then using appropriate
Proposed algorithm, matching user interest and resource characteristic completes to recommend.
Information retrieval and information mistake are mostly come from based on the theoretical foundation of the information recommendation method of content
Filter, so-called content-based recommendation method is exactly to browse record to push away to user according to user is past
Recommend the recommendation items that user was not in contact with.Content-based recommendation side is described mainly from two methods
Method:Didactic method and the method based on model.Didactic method is exactly that user by virtue of experience comes
The related computing formula of definition, is then verified further according to the result of calculation and the result of reality of formula,
Then modification formula is or else broken to reach final purpose.And for the method for model is according to conventional
Data as data set, then according to this data set learning a model.
The didactic method applied in general commending system is exactly to be counted using the method for tf-idf
Calculate, the method with there is tf-idf is calculated the higher keyword of weight ratio occurs as retouching in this document
User characteristicses are stated, and using these keywords as the vector for describing user characteristicses;Then further according to quilt
The high keyword of weight in recommendation items is used as the attribute character of recommendation items, then again by this two
The item of vector most close (calculating highest scoring with the vector of user characteristicses) recommends user.In meter
During the similarity of the characteristic vector for calculating the recommended item of user characteristicses vector sum, that what is generally used is cosine
Method, calculates the cosine values of angle between two vectors.
However, commending system is during development and application, various problems are received different degrees of
Affect, particularly sparse sex chromosome mosaicism and concept drift problem become impact and recommend the main of quality to ask
Topic.
The content of the invention
For sparse sex chromosome mosaicism present in prior art and concept drift problem, the purpose of the present invention is
A kind of content recommendation system and method based on collaborative filtering is provided.
For achieving the above object, the present invention is adopted the following technical scheme that:
A kind of content recommendation system based on collaborative filtering, including the data preprocessing module being sequentially connected,
Algorithm mixing module, result-generation module.Wherein, algorithm mixing module further includes algorithms selection
Unit, based on the cooperation recommending algorithm unit of Weighted Similarity, based on balance score in predicting mechanism association
Make proposed algorithm unit, mixing proposed algorithm unit, the utilization scoring time response filled based on scoring
Cooperation recommending algorithm unit.Algorithm mixing module is respectively input into the data of pretreatment based on weighting phase
Like degree cooperation recommending algorithm unit, based on balance score in predicting mechanism cooperation recommending algorithm unit,
Based on the mixing proposed algorithm unit of scoring filling, using the cooperation recommending algorithm list of scoring time response
Unit, the arithmetic result that algorithms selection Unit selection is most matched, and arithmetic result is exported to result generation
Module.
An embodiment of the invention, based on Weighted Similarity cooperation recommending algorithm unit perform with
Lower operation:The calculating project scoring coincidence factor and project-based Weighted Similarity;Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
An embodiment of the invention, the cooperation recommending algorithm unit based on balance score in predicting mechanism
Perform following operation:Calculate project-based similarity;The median and weight of statistical item scoring is flat
Weighing apparatus parameter;Select neighbours' project;Recommendation prediction is carried out based on project scoring.
An embodiment of the invention, is performed following based on the mixing proposed algorithm unit of scoring filling
Operation:Represent recommended project content;Learnt based on content user model;Calculate based on content
Similarity;Score in predicting based on CBF and filling;Calculate the similarity based on scoring;Select neighbours
Project;Recommendation prediction is carried out based on project scoring.
An embodiment of the invention, is performed using the cooperation recommending algorithm unit of scoring time response
Hereinafter operate:Calculating project marking and queuing and time weighting;Time-based weighting phase between calculating project
Like degree;Select neighbours' project;Project-based weighted scoring carries out recommendation prediction.
For achieving the above object, the present invention is adopted the following technical scheme that:
A kind of content recommendation method based on collaborative filtering, comprises the following steps:Data prediction;Will
The data of pretreatment carry out scoring pre- based on the cooperation recommending algorithm of Weighted Similarity, based on balance respectively
The cooperation recommending algorithm of survey mechanism, based on the mixing proposed algorithm of scoring filling, special using the scoring time
The cooperation recommending algorithm of property;The arithmetic result that selection is most matched, and using arithmetic result as content recommendation.
An embodiment of the invention, the cooperation recommending algorithm based on Weighted Similarity includes following step
Suddenly:The calculating project scoring coincidence factor and project-based Weighted Similarity;Select neighbours' project;Base
Recommendation prediction is carried out in project scoring.
An embodiment of the invention, is included based on the cooperation recommending algorithm of balance score in predicting mechanism
Following steps:Calculate project-based similarity;The median and balance of weights ginseng of statistical item scoring
Number;Select neighbours' project;Recommendation prediction is carried out based on project scoring.
An embodiment of the invention, is comprised the following steps based on the mixing proposed algorithm of scoring filling:
Represent recommended project content;Learnt based on content user model;Calculate based on the similarity of content;
Score in predicting based on CBF and filling;Calculate the similarity based on scoring;Select neighbours' project;Base
Recommendation prediction is carried out in project scoring.
An embodiment of the invention, is included following using the cooperation recommending algorithm of scoring time response
Step:Calculating project marking and queuing and time weighting;Time-based Weighted Similarity between calculating project;
Select neighbours' project;Project-based weighted scoring carries out recommendation prediction.
In above-mentioned technical proposal, the content recommendation system based on collaborative filtering and the method for the present invention compared with
Sparse sex chromosome mosaicism and the concept drift in commending system is solved the problems, such as well so that result and actual feelings
The matching degree of condition is higher.
Description of the drawings
Fig. 1 is structural representation of the present invention based on the content recommendation system of collaborative filtering;
Fig. 2 is scoring time weighting attenuation curve schematic diagram.
Specific embodiment
Technical scheme is further illustrated with reference to the accompanying drawings and examples.
With reference to Fig. 1, the present invention discloses a kind of content recommendation system based on collaborative filtering and its corresponding
Method.As shown in figure 1, the system of the present invention includes the data preprocessing module 1 being sequentially connected, calculates
Method mixing module 2, result-generation module 3, and algorithm mixing module 2 further includes algorithms selection
Unit 21, based on the cooperation recommending algorithm unit 22 of Weighted Similarity, based on balance score in predicting mechanism
Cooperation recommending algorithm unit 23, based on scoring filling mixing proposed algorithm unit 24, using scoring
The cooperation recommending algorithm unit 25 of time response.
The data of pretreatment are input into respectively and are based on by the structure according to Fig. 1, algorithm mixing module 2
The cooperation recommending algorithm unit 22 of Weighted Similarity, the cooperation recommending calculation based on balance score in predicting mechanism
Method unit 23, the mixing proposed algorithm unit 24 filled based on scoring, the association for utilizing the time response that scores
Make proposed algorithm unit 25, and algorithms selection unit 21 selects the arithmetic result for most matching, and will calculate
Method result is exported to result-generation module 3.
Below further describing the algorithm performed by above-mentioned unit.
1. the cooperation recommending algorithm unit 22 of Weighted Similarity is based on:
The present invention is directed to sparse sex chromosome mosaicism, is carried out by the Similarity measures of each implementation procedure to recommending
Necessity amendment overcomes and alleviates the impact of openness problem, it is proposed that pushed away based on the cooperation of Weighted Similarity
Algorithm is recommended, refer to adapt to the coincidence factor (Overlap Factor) of disparity items scoring distributed number,
And the amendment and improvement to traditional Similarity Measure is realized by the coincidence factor.
The coincidence factor has measured in quantity public scoring institute's accounting in the global scoring of project between project
Weight, has delineated the coincidence degree of public scoring between project, highlights public scoring registration in similarity
Importance in tolerance, by being acted on project-based similarity, can overlap from scoring
The credibility of angular area point similarity result of calculation, that is, the coincidence degree that scores is higher, then calculated
Similarity can more reflect the true relevance between project, conversely, the credibility of Similarity Measure is then got over
It is low.For any two project t of Similarity Measure between participation projectiAnd tj, two projects are carried out respectively
The user's collection for crossing scoring is combined into Ui={ uc|uc∈U∧rci≠ 0 } and Uj=u | uc∈U∧rcj≠ 0 }, then score weight
Close the factor can form turn to
Formula 2-1
It is only related to the absolute quantities of public scoring different from significance weight, calculated by formula 2-1
The coincidence factor of acquisition is not only proportional to the public scoring quantity of two projects, is also inversely proportional to two projects each
User's scoring quantity, it is ensured that the scoring coincidence factor is adapted to the scoring distributed number of disparity items.
Traditional similarity can be modified by using the coincidence factor, form corresponding weighting included angle cosine phase
Like degree (Weighted COsine SIMilarity, WCOSIM) and weighting Pearson's correlation coefficient
(Weighted Pearson Correlation Coefficient, WPCC), the two is represented by passing in form
The product of system similarity and the factor that overlaps, as shown by the equation.
Formula 2-2
Formula 2-3
It is noted that the coincidence factor proposed by the present invention is all global similarity amendment scheme, it is to CF
In all items between the integrated regulation that carries out of similarity, and the amendment of non local independent similarity, this
Plant complete modification similarity only meaningful when integrally using.On here basis, by adding above-mentioned
Power similarity is integrated in traditional IBCF recommendation process, and the present invention is proposed based on Weighted Similarity
Cooperation recommending algorithm WSBCF (Weighted Similarity-Boosted Collaborative Filtering),
The implementation procedure of WSBCF algorithms can be divided into following 3 step:
(1) the project scoring coincidence factor and project-based Weighted Similarity are calculated
Any project t is calculated according to formula 2-1 according to rating matrixiAnd tjBetween the coincidence factor, and lead to
Cross the Weighted Similarity that formula is calculated between any project.
(2) neighbours' project alternativess
For targeted customer ucArbitrarily do not access projectAccording to step (1)
Weighted Similarity between the project of acquisition, to user ucThe project of access carry out descending sort, and according to
With respect to neighbor choice threshold parameter θ, top- θ some projects are selected as project tiNeighborhood Tci。
(3) project-based score in predicting
For targeted customer ucArbitrarily do not access projectAccording to targeted customer
ucScoring, project tiNeighbours project set TciAnd similarity sim, using equation below to item
Mesh tiCarry out score in predicting
Formula 2-4
WSBCF algorithms employ conventional weight scoring polymerization during score in predicting, according to
The neighbours scored with destination item of targeted customer, with the similarity of neighbours and destination item as power
Weight, by weighted sum the score in predicting to destination item is realized.The total time complexity of WSBCF is
O(mn)+O(mn2)+O (n), this time complexity O (mn to traditional IBCF2) impact it is limited.
2. based on the cooperation recommending algorithm unit 23 for balancing score in predicting mechanism:
In order to reduce the impact of sparse sex chromosome mosaicism to IBCF proposed algorithms from score in predicting angle, can be
A kind of dynamic equilibrium is set up between personalization scoring and global scoring.The present invention establishes a kind of combination
The dynamic equilibrium score in predicting mechanism that propertyization scores with global scoring, and propose one kind and commented based on balance
Divide the cooperation recommending algorithm of forecasting mechanism.
To make the scoring of two classes that respective effect can be played in project-based score in predicting, the present invention
The dynamic equilibrium score in predicting mechanism of proposition is a kind of linear combination with regard to the scoring of two classes.Meanwhile, it is
The dynamic equilibrium of two classes scoring is kept, can be played come both dynamic adjustment by the change of weight
Effect, and the Main Basiss that the distribution character of global score data is adjusted as weight dynamic.For
Any user u of commending systemcAnd its do not access project ti, the dynamic equilibrium scoring of integrated two classes scoring
Prediction can form turn to
Formula 2-5
Wherein,Represent and be directed to destination item tiProject-based personalized scoring, the personalization scoring
Employ the IBCF score in predicting methods described by formula 2-12, giThe overall situation for destination item scores,
And αiFor the balance of weights parameter between personalized scoring and global scoring.
For impact of the sparse sex chromosome mosaicism to score in predicting process, with reference to above-mentioned dynamic equilibrium score in predicting
Mechanism, by carrying out necessary amendment to traditional IBCF score in predicting process, forms a kind of based on flat
Cooperation recommending algorithm IBCFBP (the Item-Based Collaborative Filtering of weighing apparatus score in predicting mechanism
Integrating Balanced Prediction), the implementation procedure of the algorithm mainly includes following 4 step:
(1) project-based Similarity Measure
Based on existing score data, by between project-based similarity calculating method metric terms
Similarity, for any project tiAnd tj, project-based COSIM similarities are represented by
Formula 2-6
Or using project-based PCC similarities, be represented by
Formula 2-7
(2) statistical item scoring median and balance of weights parameter
According to any project tiAcquired score data, counts the global scoring median g of the projecti,
And the dispersibility of project overall situation score data is counted using one of formula 2-5,2-6 or 2-7, with
Represent the balance of weights parameter alpha of balance score in predicting mechanismi。
(3) neighbours' project alternativess
For targeted customer ucArbitrarily do not access project ti, access item purpose is commented according to the user
Point, similarity between the project for obtaining is calculated with reference to step (1), to user ucThe project of access dropped
Sequence sorts, and selects top- θ some projects as project tiNeighbours project set Tci。
(4) project-based score in predicting
For user ucArbitrarily do not access project ti, according to scoring, project t of the useri's
Neighbours project set TciAnd related similarity, using formula to project tiIt is balanced score in predicting.
The total time complexity of IBCFBP is O (n)+O (mn2)+O (n), with traditional cooperation recommending algorithm
Time complexity O (mn2) quite, and the calculating of project median and balance of weights parameter can be from
Line is completed, so the two affects to ignore substantially on the computational efficiency recommended.
3. the mixing proposed algorithm unit 24 filled based on scoring:
The present invention proposes a kind of mixing proposed algorithm HRRF (Hybrid based on scoring filling
Recommendation based on Rating Filling), the algorithm can be according to user's access item purpose
Information content realizes automatization's user modeling, and the user model based on content is used to realize to user
The personalized scoring filling of non-access item purpose, improves the global density of rating matrix, and then based on Jing
The rating matrix that filling is processed, recommends framework to realize the prediction of scoring using IBCF, and HRRF is from scoring
The angle of filling, reduces sparse sex chromosome mosaicism pre- to CF Similarity Measures and scoring by rating matrix
The impact of survey process.
It is the foundation that HRRF realizes scoring filling based on the user model of content, HRRF is based on content
User model serve the effect of Filterbot in itself, but different from Filterbot manual constructions
Mode, the user model of HRRF constructed automatically by machine learning.By the structure of user model
Make and be interpreted as a kind of text categorization task based on machine learning, according to access item purpose content information,
Disaggregated model for different user can automatically be trained by certain machine learning algorithm, and should
Model is described as corresponding user, and HRRF employs Rocchio learning algorithms to realize based on interior
The user modeling of appearance, it is also possible to which use is trained according to actual recommendation environmental selection other machines learning algorithm
Family model.
The overall execution process of HRRF mixing proposed algorithms can be described as following 7 steps:
(1) recommended project content representation
Because the user model of HRRF is built upon in the content basis of recommended, first
Need to be indicated the content of information object, HRRF employs traditional VSM models and represents recommendation
The content of object, if commending system includes the project set being made up of n recommended project
T={ ts| 1≤s≤n }, the feature space for described project content is X={ x1,x2,...,xd, recommended
tiCharacteristic vector beWherein characteristic componentRepresent feature xjFor
Project tiWeight.IfRepresent feature xjIn project tiThe word frequency occurred in content, and dfjRepresent
Comprising feature xjProject number, then feature weightIt is represented by
Formula 2-8
(2) study based on content user model
The user model of HRRF has been used and recommended project identical feature space X={ x1,x2,...,xd,
If the user set U={ u of commending systemc| 1≤c≤m }, user uiRepresented based on the user model of content
ForAccording to the existing positive example scoring item set T of user+With negative example scoring item
Mesh set T-, the threshold value of classification of scoring selects the scoring average of the targeted customer, according to Rocchio
Practise the feature weight of Algorithm for Training user modelIt is represented by
Formula 2-9
Because Rocchio learning algorithms are a kind of batch-mode algorithm (Batch Learning), so user
Model must be periodically updated, but the renewal of model can be completed offline, will not be to commending system
Online calculating performance make too big impact.
(3) Similarity Measure based on content
According to the user model based on content that step (2) is obtained, for commending system in any use
Family ui, the user and its item that arbitrarily do not score are measured based on the similarity calculating method of content by tradition
Mesh tjSimilarity in terms of content, it is for instance possible to use the COSIM Similarity Measures of formula 2-10
Form.
Formula 2-10
(4) score in predicting based on CBF and filling
In order to improve the global density of rating matrix, according to step (3) obtain user model with do not score
Similarity between project, score in predicting is carried out using traditional CBF to the project that do not access of user, and is made
With the corresponding rating matrix position of prediction scoring filling, user-project scoring complete matrix is formed.In scoring
During codomain scope is for the commending system of [min, max], for user uiAnd it does not arbitrarily access project
tj, then predict that score value is
r′ij=min+s (i, j) | max-min | formulas 2-11
(5) Similarity Measure based on scoring
By the process of step (1)-(4), original sparse user rating matrix has obtained filling process,
Subsequent step will be realized recommending using IBCF algorithms, for any project tiAnd tj, it is based between the two
The COSIM similarities of project are
Formula 2-12
Note, whether needs had accessed project t according to useri, determine to select based on the filling of content
Score value r 'ciOr original score value rciThe scoring r that comes in representation formula 2-12 "ci, i.e.,
Formula 2-13
(6) neighbours' project alternativess
For destination item ti, according to similarity between the project that step (5) is obtained, to every other project
Descending sort is carried out, and selects k most like project to constitute project tiNeighborhood Ti。
(7) project-based score in predicting
For targeted customer uiArbitrarily do not access project tj, the project is carried out by IBCF methods
Score in predicting.In view of the filling generated by CBF is scored to score less than true in credibility,
During score in predicting, similarity is reduced reducing scoring Filling power in score in predicting by appropriate
Effect, the calculating of score in predicting is represented by
Formula 2-14
TIBCF algorithms only introduce time weighting during Similarity Measure and score in predicting, from meter
From the point of view of in calculation amount, the addition of time weighting is limited to the overall calculation complexity effect of recommendation.
4. using the cooperation recommending algorithm unit 25 of scoring time response:
Current proposed algorithm is not set up perceiving the dynamic mechanism of user interest change, the use for being formed
Family interest model is a kind of static models, with the generation of concept drift, the recommendation quality of commending system
By the unstable of performance, when particularly user interest is undergone mutation, the accuracy of recommendation will drastically deteriorate.
Therefore, drifting problem when invention is directed to, invention proposes a kind of cooperation using scoring time response and pushes away
Recommend algorithm TIBCF (Temporal Item-Based Collaborative Filtering, TIBCF).
TIBCF algorithms proposed by the present invention are needed according to scoring during concept drift is overcome
The generation time makes a distinction to the importance for scoring.Therefore, TIBCF algorithms enter to traditional rating matrix
Gone it is necessary perfect, except record score value in addition to, be also recorded for score generation time information,
User in TIBCF algorithms-project rating matrix is represented by:
R (m × n)={ < rij,dij> | (ui∈U)∧(tj∈T)∧(0≤rij≤q)∧(dij=NULL ∨ dij∈DateTime)}
User uiFor project tjScore information be expressed as two tuples in rating matrix R (m × n)
< rij,dij>, rijAnd dijScore value and scoring time are corresponded to respectively.In order to pre- in Similarity Measure and scoring
The importance of scoring is distinguished in survey from time angle, TIBCF algorithms introduce the concept of time weighting,
Corresponding time weighting is given for the scoring that different time is produced.For user uiIn time dijTo project
tjProduced scoring rij, TIBCF according to the decaying exponential function form calculus of the formula scorings when
Between weight w (i, j).
Formula 2-15
Wherein, RiFor user uiHistory scoring ordered set, RK function representations scoring rijIn RiIn
Position Number.With reference to the method for AWS in document, TIBCF arranges the half-life of scoring weight
(Half-Life Span) be λ/log (| Ri|), it was both related to the scoring quantity of user, but with attenuation parameter λ
It is related.The length of half-life, the selection of λ and the scoring of concrete application can be adjusted by attenuation parameter λ
Distribution is relevant, can be obtained by testing, and also dynamic can adjust according to the change of commending system accuracy
It is whole.For the user with 500 scorings, Fig. 2 illustrates λ=300 and λ=500 news commentary fraction weight
The example of attenuation curve.
After the time weightings obtained with difference scoring are calculated according to formula, TIBCF is in traditional IBCF
The weight is all integrated with the Similarity Measure and score in predicting of algorithm, so as to realize accordingly being based on respectively
The Weighted Similarity of time is calculated and weighted scoring prediction, by carrying out to two critical process of IBCF
Time-based amendment, can to greatest extent reduce impact of the concept drift problem to recommending.TIBCF
The implementation procedure of proposed algorithm can be divided into following 4 steps:
(1) project marking and queuing and time weighting are calculated
For any userThe generation time scored according to the user is right by closely to remote
The sequence that its all history scoring is carried out temporally generates history scoring ordered set Ri, calculate user ui
Any history scoring rijTime weighting w (i, j).
(2) time-based Weighted Similarity is calculated between project
Obtained after the time weighting of all scorings by step (1), to system in any two project tiWith
tj, according to formula 2-15 or 2-16 time-based weighting Pearson's correlation coefficient between the two is calculated
More than (Temporal Pearson Correlation Coefficient, TPCC) or time-based weighting angle
String similarity (Temporal COsine SIMilarity, TCOSIM).
Formula 2-16
Formula 2-17
(3) neighbours' project alternativess
For targeted customer ucArbitrarily do not access project ti, according to similar between the project that step (2) is obtained
Degree, to user ucAll projects that accessed carry out descending sort, and select top- θ some projects to make
For project tiNeighbours project set Tci。
(4) project-based weighted scoring prediction
For targeted customer ucArbitrarily do not access project ti, according to user ucScoring, ti's
Neighbours project set Tci, scoring time weighting w and respective weight similarity sim, by equation below
To project tiCarry out score in predicting
Formula 2-18
TIBCF algorithms only introduce time weighting during Similarity Measure and score in predicting, from meter
From the point of view of in calculation amount, the addition of time weighting is limited to the overall calculation complexity effect recommended, still can be with
Keep and tradition IBCF algorithm O (mn2) suitable calculating time complexity level.
Those of ordinary skill in the art is it should be appreciated that the embodiment of the above is intended merely to
The bright present invention, and be not used as limitation of the invention, as long as in the spirit of the present invention
Interior, the change, modification to embodiment described above all will fall in the range of claims of the present invention.
Claims (10)
1. a kind of content recommendation system based on collaborative filtering, it is characterised in that include:
Data preprocessing module, algorithm mixing module, the result-generation module being sequentially connected;
Wherein, algorithm mixing module further includes algorithms selection unit, the association based on Weighted Similarity
Make proposed algorithm unit, based on balance score in predicting mechanism cooperation recommending algorithm unit, based on scoring
The mixing proposed algorithm unit of filling, the cooperation recommending algorithm unit for utilizing scoring time response;
The data of pretreatment are input into respectively and are pushed away based on the cooperation of Weighted Similarity by the algorithm mixing module
Recommend algorithm unit, fill based on the cooperation recommending algorithm unit of balance score in predicting mechanism, based on scoring
Mixing proposed algorithm unit, using scoring time response cooperation recommending algorithm unit, the algorithm
Select unit selects the arithmetic result for most matching, and the arithmetic result is exported to result-generation module.
2. the content recommendation system of collaborative filtering is based on as claimed in claim 1, it is characterised in that
The cooperation recommending algorithm unit based on Weighted Similarity performs following operation:
The calculating project scoring coincidence factor and project-based Weighted Similarity;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
3. the content recommendation system of collaborative filtering is based on as claimed in claim 1, it is characterised in that
The cooperation recommending algorithm unit based on balance score in predicting mechanism performs following operation:
Calculate project-based similarity;
The median and balance of weights parameter of statistical item scoring;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
4. the content recommendation system of collaborative filtering is based on as claimed in claim 1, it is characterised in that
The mixing proposed algorithm unit based on scoring filling performs following operation:
Represent recommended project content;
Learnt based on content user model;
Calculate based on the similarity of content;
Score in predicting based on CBF and filling;
Calculate the similarity based on scoring;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
5. the content recommendation system of collaborative filtering is based on as claimed in claim 1, it is characterised in that
The cooperation recommending algorithm unit using scoring time response performs following operation:
Calculating project marking and queuing and time weighting;
Time-based Weighted Similarity between calculating project;
Select neighbours' project;
Project-based weighted scoring carries out recommendation prediction.
6. a kind of content recommendation method based on collaborative filtering, it is characterised in that comprise the following steps:
Data prediction;
The data of pretreatment are carried out respectively based on the cooperation recommending algorithm of Weighted Similarity, based on balance
The cooperation recommending algorithm of score in predicting mechanism, mixing proposed algorithm, the utilization scoring filled based on scoring
The cooperation recommending algorithm of time response;
The arithmetic result that selection is most matched, and using the arithmetic result as content recommendation.
7. the content recommendation method of collaborative filtering is based on as claimed in claim 6, it is characterised in that
Comprised the following steps based on the cooperation recommending algorithm of Weighted Similarity:
The calculating project scoring coincidence factor and project-based Weighted Similarity;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
8. the content recommendation method of collaborative filtering is based on as claimed in claim 6, it is characterised in that
The cooperation recommending algorithm based on balance score in predicting mechanism is comprised the following steps:
Calculate project-based similarity;
The median and balance of weights parameter of statistical item scoring;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
9. the content recommendation method of collaborative filtering is based on as claimed in claim 6, it is characterised in that
The mixing proposed algorithm based on scoring filling is comprised the following steps:
Represent recommended project content;
Learnt based on content user model;
Calculate based on the similarity of content;
Score in predicting based on CBF and filling;
Calculate the similarity based on scoring;
Select neighbours' project;
Recommendation prediction is carried out based on project scoring.
10. the content recommendation method of collaborative filtering is based on as claimed in claim 6, it is characterised in that
The cooperation recommending algorithm using scoring time response is comprised the following steps:
Calculating project marking and queuing and time weighting;
Time-based Weighted Similarity between calculating project;
Select neighbours' project;
Project-based weighted scoring carries out recommendation prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510688543.1A CN106610970A (en) | 2015-10-21 | 2015-10-21 | Collaborative filtering-based content recommendation system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510688543.1A CN106610970A (en) | 2015-10-21 | 2015-10-21 | Collaborative filtering-based content recommendation system and method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106610970A true CN106610970A (en) | 2017-05-03 |
Family
ID=58611410
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510688543.1A Pending CN106610970A (en) | 2015-10-21 | 2015-10-21 | Collaborative filtering-based content recommendation system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106610970A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107230002A (en) * | 2017-06-09 | 2017-10-03 | 天津大学 | Project-based rating matrix prediction algorithm |
CN107491557A (en) * | 2017-09-06 | 2017-12-19 | 徐州医科大学 | A kind of TopN collaborative filtering recommending methods based on difference privacy |
CN108228745A (en) * | 2017-12-19 | 2018-06-29 | 北京荣之联科技股份有限公司 | A kind of proposed algorithm and device based on collaborative filtering optimization |
CN108810640A (en) * | 2018-06-15 | 2018-11-13 | 重庆知遨科技有限公司 | A kind of recommendation method of TV programme |
CN108875045A (en) * | 2018-06-28 | 2018-11-23 | 第四范式(北京)技术有限公司 | The method and its system of machine-learning process are executed for text classification |
CN109933717A (en) * | 2019-01-17 | 2019-06-25 | 华南理工大学 | A kind of academic conference recommender system based on mixing proposed algorithm |
CN110580311A (en) * | 2019-06-21 | 2019-12-17 | 东华大学 | Dynamic time perception collaborative filtering method |
CN111339434A (en) * | 2018-12-03 | 2020-06-26 | 阿里巴巴集团控股有限公司 | Information recommendation method and device, electronic equipment and computer storage medium |
CN111400591A (en) * | 2020-03-11 | 2020-07-10 | 腾讯科技(北京)有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN112748941A (en) * | 2020-08-06 | 2021-05-04 | 腾讯科技(深圳)有限公司 | Feedback information-based target application program updating method and device |
CN113256387A (en) * | 2021-06-15 | 2021-08-13 | 杭州魔点科技有限公司 | Personalized recommendation method for item, electronic equipment and storage medium |
US11727312B2 (en) | 2019-09-03 | 2023-08-15 | International Business Machines Corporation | Generating personalized recommendations to address a target problem |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663128A (en) * | 2012-04-24 | 2012-09-12 | 南京师范大学 | Recommending system of large-scale collaborative filtering |
CN102880691A (en) * | 2012-09-19 | 2013-01-16 | 北京航空航天大学深圳研究院 | User closeness-based mixed recommending system and method |
CN103578007A (en) * | 2012-07-20 | 2014-02-12 | 三星电子(中国)研发中心 | Mixed recommendation system and method for intelligent device |
CN103617540A (en) * | 2013-10-17 | 2014-03-05 | 浙江大学 | E-commerce recommendation method of tracking user interest changes |
CN105354720A (en) * | 2014-09-18 | 2016-02-24 | 中南大学 | Visual clustering based consumption place mixed recommendation method |
-
2015
- 2015-10-21 CN CN201510688543.1A patent/CN106610970A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663128A (en) * | 2012-04-24 | 2012-09-12 | 南京师范大学 | Recommending system of large-scale collaborative filtering |
CN103578007A (en) * | 2012-07-20 | 2014-02-12 | 三星电子(中国)研发中心 | Mixed recommendation system and method for intelligent device |
CN102880691A (en) * | 2012-09-19 | 2013-01-16 | 北京航空航天大学深圳研究院 | User closeness-based mixed recommending system and method |
CN103617540A (en) * | 2013-10-17 | 2014-03-05 | 浙江大学 | E-commerce recommendation method of tracking user interest changes |
CN105354720A (en) * | 2014-09-18 | 2016-02-24 | 中南大学 | Visual clustering based consumption place mixed recommendation method |
Non-Patent Citations (1)
Title |
---|
任磊: ""推荐系统关键技术研究"", 《中国博士学位论文全文数据库•信息科技辑》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107230002A (en) * | 2017-06-09 | 2017-10-03 | 天津大学 | Project-based rating matrix prediction algorithm |
CN107491557A (en) * | 2017-09-06 | 2017-12-19 | 徐州医科大学 | A kind of TopN collaborative filtering recommending methods based on difference privacy |
CN108228745A (en) * | 2017-12-19 | 2018-06-29 | 北京荣之联科技股份有限公司 | A kind of proposed algorithm and device based on collaborative filtering optimization |
CN108810640B (en) * | 2018-06-15 | 2020-12-18 | 浙江广业软件科技有限公司 | Television program recommendation method |
CN108810640A (en) * | 2018-06-15 | 2018-11-13 | 重庆知遨科技有限公司 | A kind of recommendation method of TV programme |
CN108875045A (en) * | 2018-06-28 | 2018-11-23 | 第四范式(北京)技术有限公司 | The method and its system of machine-learning process are executed for text classification |
CN111339434A (en) * | 2018-12-03 | 2020-06-26 | 阿里巴巴集团控股有限公司 | Information recommendation method and device, electronic equipment and computer storage medium |
CN111339434B (en) * | 2018-12-03 | 2023-04-28 | 阿里巴巴集团控股有限公司 | Information recommendation method and device, electronic equipment and computer storage medium |
CN109933717A (en) * | 2019-01-17 | 2019-06-25 | 华南理工大学 | A kind of academic conference recommender system based on mixing proposed algorithm |
CN109933717B (en) * | 2019-01-17 | 2021-05-14 | 华南理工大学 | Academic conference recommendation system based on hybrid recommendation algorithm |
CN110580311A (en) * | 2019-06-21 | 2019-12-17 | 东华大学 | Dynamic time perception collaborative filtering method |
CN110580311B (en) * | 2019-06-21 | 2023-08-01 | 东华大学 | Dynamic time perception collaborative filtering method |
US11727312B2 (en) | 2019-09-03 | 2023-08-15 | International Business Machines Corporation | Generating personalized recommendations to address a target problem |
CN111400591A (en) * | 2020-03-11 | 2020-07-10 | 腾讯科技(北京)有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN111400591B (en) * | 2020-03-11 | 2023-04-07 | 深圳市雅阅科技有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN112748941A (en) * | 2020-08-06 | 2021-05-04 | 腾讯科技(深圳)有限公司 | Feedback information-based target application program updating method and device |
CN112748941B (en) * | 2020-08-06 | 2023-12-12 | 腾讯科技(深圳)有限公司 | Method and device for updating target application program based on feedback information |
CN113256387A (en) * | 2021-06-15 | 2021-08-13 | 杭州魔点科技有限公司 | Personalized recommendation method for item, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106610970A (en) | Collaborative filtering-based content recommendation system and method | |
Reddy et al. | Content-based movie recommendation system using genre correlation | |
CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
CN105224699B (en) | News recommendation method and device | |
CN111444395B (en) | Method, system and equipment for obtaining relation expression between entities and advertisement recall system | |
CN104063481A (en) | Film individuation recommendation method based on user real-time interest vectors | |
CN106503025A (en) | Method and system is recommended in a kind of application | |
CN113569129A (en) | Click rate prediction model processing method, content recommendation method, device and equipment | |
CN103559303A (en) | Evaluation and selection method for data mining algorithm | |
CN109816015B (en) | Recommendation method and system based on material data | |
CN106846029B (en) | Collaborative filtering recommendation algorithm based on genetic algorithm and novel similarity calculation strategy | |
CN107016566A (en) | User model construction method based on body | |
Du et al. | Pcard: Personalized restaurants recommendation from card payment transaction records | |
Fu et al. | Collaborative filtering recommendation algorithm towards intelligent community | |
CN116244513A (en) | Random group POI recommendation method, system, equipment and storage medium | |
CN107305559A (en) | Method and apparatus are recommended in one kind application | |
CN110264277A (en) | Data processing method and device, medium and the calculating equipment executed by calculating equipment | |
CN105279699A (en) | Recommendation method combining multi-class untrust relation based on collaborative filtering | |
CN116957128A (en) | Service index prediction method, device, equipment and storage medium | |
CN112464106B (en) | Object recommendation method and device | |
Li et al. | Collaborative filtering recommendation algorithm based on user characteristics and user interests | |
CN113657766A (en) | Tourist attraction joy index metering method based on tourist multi-metadata | |
CN111523034A (en) | Application processing method, device, equipment and medium | |
CN110825967A (en) | Recommendation list re-ranking method for improving diversity of recommendation system | |
CN110766488A (en) | Method and device for automatically determining theme scene |
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: 20170503 |
|
WD01 | Invention patent application deemed withdrawn after publication |