CN110321490A - Recommended method, device, equipment and computer readable storage medium - Google Patents
Recommended method, device, equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN110321490A CN110321490A CN201910628953.5A CN201910628953A CN110321490A CN 110321490 A CN110321490 A CN 110321490A CN 201910628953 A CN201910628953 A CN 201910628953A CN 110321490 A CN110321490 A CN 110321490A
- Authority
- CN
- China
- Prior art keywords
- user
- recommended
- score
- scoring
- 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.)
- Granted
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the present invention provides a kind of recommended method, device, equipment and computer readable storage medium.This method comprises: the attribute tensor of the multiple objects that do not score of attribute tensor sum based on multiple objects that scored of the first user in recommender system, calculates the first user and scores the first prediction of each object that do not score；To multiple objects to be recommended of the first determining user, calculates the first user and score the second prediction of each object to be recommended；Based on the first prediction scoring and the second prediction scoring, at least one target recommended is determined, in multiple do not score object and multiple objects to be recommended to be recommended.The embodiment of the present invention can be improved recommendation diversity.
Description
Technical field
The present embodiments relate to Internet technical field more particularly to a kind of recommended method, device, equipment and computers
Readable storage medium storing program for executing.
Background technique
With the rapid development of internet technology, online information is also in explosive growth.User is in the selection for facing magnanimity
When often have no way of doing it, the proposition of recommender system can help user quickly to navigate to its interested target.On the one hand it pushes away
The system of recommending can effectively be promoted user face magnanimity selection when experience, on the other hand, recommender system can with helpdesk into
Row user's water conservancy diversion, and attract and carry out more users.
Currently, existing recommended method mainly has method based on content, based on the method for collaborative filtering and mixing side
Method.Method based on content is the content information according to article itself, if the text of article describes, attribute tags etc., and then base
Text similarity is calculated in these content informations, so that similarity between article is obtained, it finally will be similar with user's history article
Article recommends user.But it is similarity of the article in the one-dimensional space for the calculating of similarity based on the method for content, from
And cause to recommend diversity lower.
Summary of the invention
The embodiment of the present invention provides a kind of recommended method, device, equipment and computer readable storage medium, is recommended with improving
Diversity.
In a first aspect, the embodiment of the present invention provides a kind of recommended method, comprising: based in recommender system the first user it is more
The attribute tensor of the multiple objects that do not score of the attribute tensor sum of a object that scored calculates the first user to the of the object that do not score
One prediction scoring；To multiple objects to be recommended of the first determining user, calculates the first user and treat the second pre- of recommended
Assessment point；Based on the first prediction scoring and the second prediction scoring, determined in multiple do not score object and multiple objects to be recommended
At least one target recommended, to be recommended.
Second aspect, the embodiment of the present invention provide a kind of recommendation apparatus, comprising: the first computing module, for based on recommendation
The attribute tensor of the multiple objects that do not score of the attribute tensor sum of multiple objects that scored of first user in system calculates first and uses
It scores the first prediction of the object that do not score at family；Second computing module, for the multiple to be recommended right of the first determining user
As calculating the second prediction scoring that the first user treats recommended；Determining module, for based on the first prediction scoring and second
Prediction scoring, determines at least one target recommended, in multiple do not score object and multiple objects to be recommended to be pushed away
It recommends.
The third aspect, the embodiment of the present invention provide a kind of recommendation apparatus, comprising: memory；
Processor；And
Computer program；
Wherein, computer program stores in memory, and is configured as being executed by processor to realize such as first aspect
Method.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, are stored thereon with computer program,
The method that computer program is executed by processor to realize first aspect.
Recommended method, device, equipment and computer readable storage medium provided in an embodiment of the present invention, by based on recommendation
The attribute tensor of the multiple objects that do not score of the attribute tensor sum of multiple objects that scored of first user in system calculates first and uses
It scores the first prediction of the object that do not score at family；To multiple objects to be recommended of the first determining user, the first user couple is calculated
Second prediction scoring of object to be recommended；Based on the first prediction scoring and the second prediction scoring, in multiple objects and more of not scoring
At least one target recommended is determined in a object to be recommended, to be recommended.Due to object and the object that do not score of having scored
It is expressed as attribute tensor respectively, so that the calculating of the first prediction scoring considers in recommender system between the attribute of each object
Association, therefore, can alleviate data sparsity problem, realize the diversification of recommendation results.
Detailed description of the invention
Fig. 1 is recommended method flow chart provided in an embodiment of the present invention；
Fig. 2 be another embodiment of the present invention provides recommended method flow chart；
Fig. 3 be another embodiment of the present invention provides recommended method flow chart；
Fig. 4 is the node relationships figure of user provided in an embodiment of the present invention, object and label；
Fig. 5 is user provided in an embodiment of the present invention, object-label pair node relationships figure；
Fig. 6 is schematic diagram provided in an embodiment of the present invention；
Fig. 7 be another embodiment of the present invention provides recommended method flow chart；
Fig. 8 is the structural schematic diagram of recommendation apparatus provided in an embodiment of the present invention；
Fig. 9 is the structural schematic diagram of recommendation apparatus provided in an embodiment of the present invention.
Through the above attached drawings, it has been shown that the specific embodiment of the disclosure will be hereinafter described in more detail.These attached drawings
It is not intended to limit the scope of this disclosure concept by any means with verbal description, but is by referring to specific embodiments
Those skilled in the art illustrate the concept of the disclosure.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
In the prior art, the method based on collaborative filtering is generally that other similar use of interest therewith are found for user
Family, by their interested commending contents to original subscriber.It does not need the attribute and content information of processing user and article.Mixing
Method is to combine polymorphic type recommended method in certain method, or the result that a variety of recommended methods obtain is merged,
Or a variety of methods are fused into a combination frame.And there is following two in the algorithm based on collaborative filtering: 1,
Data sparsity problem is difficult to obtain its interest according to its historical record merely that is, for sluggish user；2, efficiency is asked
Topic, i.e., most of collaborative filtering method is easy to the problem of facing large-scale calculations and higher dimensional matrix storage at present.And it mixes
It is tactful most of be it is predefined, the strategy of mixing is all more stiff, cannot carry out the adjustment of adaptability according to the actual situation.
Recommended method provided in an embodiment of the present invention, it is intended to solve at least one technical problem as above of the prior art.
How to be solved with technical solution of the specifically embodiment to technical solution of the present invention and the application below above-mentioned
Technical problem is described in detail.These specific embodiments can be combined with each other below, for the same or similar concept
Or process may repeat no more in certain embodiments.Below in conjunction with attached drawing, the embodiment of the present invention is described.
Fig. 1 is recommended method flow chart provided in an embodiment of the present invention.The embodiment of the present invention is as above for the prior art
Technical problem provides recommended method, and specific step is as follows for this method:
Step 101, the attribute tensor sum based on multiple objects that scored of the first user in recommender system is multiple does not score
The attribute tensor of object calculates the first user and scores the first prediction of each object that do not score.
Specifically, recommender system includes multiple users and multiple objects, recommender system can be according to the Characteristic of Interest of user
And buying behavior, to the interested object of user recommended user, to realize personalized recommendation.Wherein, each user can be right
Multiple objects are answered, object can be information, be also possible to article, such as a film, a piece of music, a book, a commodity
Etc..First user refers to the user of pending recommendation, can be any one user in recommender system.User can be with to object
It scores, for example, a certain user scores to a film, a piece of music, a book or a commodity.Optionally, it uses
Family can be user to the film, music, book or commodity to the scoring of a film, a piece of music, a book or a commodity
Direct marking, be also possible to platform and commented according to what evaluation information of the user to the film, music, book or commodity was calculated
Point, the present invention is not specifically limited in this embodiment.
Optionally, every an object has multiple attributes.For example, the attribute of a certain portion's film may include movie name, lead
It drills, performer, the attribute informations such as type.The attribute of certain a piece of music may include the attributes such as musical designation, singer's title, style letter
Breath.The attribute of a certain part commodity may include the attribute informations such as product name, commodity price, merchandise classification.In short, attribute is constituted
Expression to certain an object, the present invention will not enumerate herein.
In embodiments of the present invention, the attribute tensor of object of having scored refers to and has commented the first user in recommender system
The attribute of the object divided is expressed using tensor.Assuming that the object that scored is X, then the object X that scored can be expressed as a N
Rank tensor,Each dimension of the N rank tensor, i.e., a category of every corresponding object X that scored of single order
Property, IiIndicate the set sizes of all values of ith attribute.
Do not score object attribute tensor it is similar with the attribute tensor for the object that scored, for details, reference can be made to the objects that scored
The introduction of attribute tensor, details are not described herein.
In the embodiment of the present invention, the object that do not score can be used as the object to be recommended of the first user, and recommender system can be counted
It calculates the first user to score to the first of each object that do not score the prediction, the interest journey for the object that do not score predicting the first user to this
Degree.
Step 102, multiple objects to be recommended to the first determining user calculate the first user to each object to be recommended
Second prediction scoring.
Specifically, can be through collaborative filtering method multiple objects to be recommended of the first user determined.For example, first
The label that the object of interest and/or the first user for first finding the first user mark the object of interest, and then pass through the first user
Object of interest and/or the first user label that the object of interest is marked find with the first user have same object and/or
The second user of the same label of same target finally finds the object of interest of second user again, then by the emerging of second user
To be recommended object of the interesting object as the first user calculates the first user and scores the second prediction of each object to be recommended, come
Predict the first user to the level of interest of the object to be recommended.
It should be noted that step 101 and step 102 do not limit specifically it is successive execute sequence, can be and first carry out step
Rapid 101, then step 102 is executed, it is also possible to first carry out step 102, then execute step 101, can also be and be performed simultaneously step
101 and step 102.
Step 103, based on the first prediction scoring and the second prediction scoring, in multiple objects and multiple to be recommended right of not scoring
As at least one target recommended of middle determination, to be recommended.
Optionally, based on the first prediction scoring and the second prediction scoring, in multiple objects and multiple to be recommended right of not scoring
As at least one target recommended of middle determination, forms recommendation list and recommend the first user.
The embodiment of the present invention is more by the attribute tensor sum based on multiple objects that scored of the first user in recommender system
The attribute tensor of a object that do not score calculates the first user and scores the first prediction of each object that do not score；To determining
Multiple objects to be recommended of one user calculate the first user and score the second prediction of each object to be recommended；It is pre- based on first
Assessment divides and the second prediction scoring, and the recommendation pair of at least one target is determined in multiple do not score object and multiple objects to be recommended
As to be recommended.Due to having scored object and the object that do not score is expressed as attribute tensor respectively, so that the first prediction scoring
Therefore data sparsity problem can be alleviated by calculating the association considered in recommender system between the attribute of each object, realize
The diversification of recommendation results.
Fig. 2 be another embodiment of the present invention provides recommended method flow chart.On the basis of the above embodiments, this implementation
The recommended method that example provides specifically comprises the following steps:
Step 201, the respective attribute tensor sum of the object of scoring based on the first user in recommender system do not score object
Attribute tensor, calculate the similarity that has scored between object and the object that do not score.
Optionally, the respective attribute tensor sum of the object that scored based on the first user in recommender system does not score object
Attribute tensor calculates the similarity that has scored between object and the object that do not score, comprising: object and does not comment to will score respectively
Divide object to carry out the attribute tensor that tensor is expressed and carries out tensor resolution respectively；According to scored object and the object that do not score it is each
From tensor resolution as a result, calculating the similarity that has scored between object and the object that do not score.Specifically, respectively to will score
Object and the object that do not score carry out the attribute tensor that tensor is expressed carry out respectively tensor resolution refer to scored object and
The respective attribute tensor of the object that do not score carries out tensor resolution respectively, wherein the attribute of scored object and the object that do not score
Amount is to carry out tensor to each attribute of scored object and the object that do not score respectively to express to obtain.Specifically, according to having scored
Object and the respective tensor resolution of the object that do not score refer to as a result, calculating the similarity to have scored between object and the object that do not score
According to the attribute tensor sum for the object that scored do not score the attribute tensor of object tensor resolution as a result, calculate scored object and
The similarity not scored between object.In a specific embodiment, the scoring based on the first user in recommender system
The respective attribute tensor sum of object does not score the attribute tensor of object, and calculating has been scored similar between object and the object that do not score
Degree, comprising: respectively to the attribute tensor sum for the object that scored do not score object attribute tensor carry out tensor resolution, obtain multiple
The multiple second feature tensors of fisrt feature tensor sum, an attribute of the corresponding object that scored of each fisrt feature tensor, each
One attribute of the corresponding object that do not score of second feature tensor；According to the multiple second feature tensors of multiple fisrt feature tensor sums,
Calculate the similarity to have scored between object and the object that do not score.
Specifically, object of assuming to have scored is X1, the object that do not score is X2, then the object X1 that scored can be expressed as one
N rank tensor,Each dimension of the N rank tensor, i.e., every single order correspondence have scored object X1's
One attribute, I1iIndicate scored object X1 ith attribute all values set sizes.Similarly, do not score object X2
Also it can be expressed as a N rank tensor,Each dimension of the N rank tensor, i.e., every single order pair
Should not scored an attribute of object X2, I2iIndicate do not score object X2 ith attribute all values set sizes.
Below by the object that scored for the tensor resolution of X1 for be illustrated.For example, using the method for tensor resolution
By scored object be X1 attribute tensor resolution at a core tensor identical with attribute tensor dimensionWith the factor matrix in each dimensionCarry out the approximate expression of mode-i matrix multiplication, table
It is specific as follows up to formula:
Wherein, core tensorRepresent the relationship between the implicit factor of each attribute for the object X1 that scored, each dimension
On factor matrix A(i)Represent the relationship between the feature and the implicit factor on the attribute.For example, cinema network can be expressed as one
A tetradic,Wherein each dimension respectively indicates movie name, director, performer, type.By tensor
After decomposition, core tensor is d × d × d × d tensor, and d is smaller constant, and each factor matrix be d ×
InThe matrix of size.Assuming that A(1)Movie name subspace factor matrix is represented, thenIndicate the feature of all films
Tensor is an I1× d × d × d size tensor, each film can be indicated with d × d × d characteristic tensor.
Optionally, for the object v that scorediAnd vj, attribute tensor is respectively XiAnd Xj, then scored object viAnd vjIt
Between similarity are as follows:
In formula (2),In l, m, n respectively represent attribute tensor XiValue in three dimensions.
Further, phase between any two can be calculated using all objects in above-mentioned calculation method ergodic data library
Like degree, the similarity matrix of these objects is obtained.
Step 202, based on the similarity and the first user to have scored between object and the object that do not score to having scored pair
The scoring of elephant calculates the first user and scores the first prediction of the object that do not score.
Specifically, the first prediction scoring can be calculated using following formula:
In formula (3), r (ui,vk) represent user uiTo certain an object vkScoring, sim (vi,vk) represent object viAnd object
vkBetween similarity, can be calculated using formula (2).
For example, user u1To v1、u2And v3Three objects are given a mark respectively, and score is respectively r1, r2, r3, object v4
It does not give a mark, v4With v1、u2And v3Similarity be respectively s1、s2、s3, then can be using above-mentioned formula (2) prediction user u1 to right
As the score of v4, the prediction for obtaining v4 is scored at (s1*r1+s2*r2+s3*r3)/(s1+s2+s3)。
The object of not giving a mark of the first user is carried out in advance likewise, can be calculated using formula (2) and formula (3)
Assessment point, all users can use prediction matrix R to the scoring or prediction scoring of all objects in recommender systemt(ui,vj) come
It indicates, every a line of prediction matrix represents some user in recommender system, and each column represent a certain user to some
The marking or prediction scoring of object.
Each attribute of object is constituted tensor by the embodiment of the present invention, and abstract characteristics of objects is extracted using tensor resolution method
Tensor can capture the connection between the different each attributes of object, and since abstract object is special compared with traditional method
It is all smaller to levy each dimension of tensor, therefore only needs a small amount of memory space.
Fig. 3 be another embodiment of the present invention provides recommended method flow chart.Fig. 4 is use provided in an embodiment of the present invention
The node relationships figure at family, object and label.Fig. 5 is user provided in an embodiment of the present invention, object-label pair node relationships
Figure.On the basis of the above embodiments, recommended method provided in this embodiment specifically comprises the following steps:
Step 301, the multiple incidence relations for determining object to be recommended, incidence relation be the first user, the first user it is emerging
Interesting object and/or the label marked to object of interest have the second of same interest object and/or same label with the first user
Incidence relation between user and the object of interest of second user.
It is assumed that the first user is u1, the first user u1Object of interest be v1, the first user u1To interest pair
As v1The label beaten is t1.According to object of interest v1With label t1Find multiple second user u2、u3And u4And second user u2
Object of interest v2And v3, second user u3Object of interest vn, second user u4Object of interest v3.In the present embodiment, emerging
Interesting object can be the object that user selected, the object or used object that browsed, can also be pair bought
As.It should be understood that being also possible to only look for by label that the object of interest of the first user or the first user were beaten in Fig. 4
To these shared object of interest or the second user of label, the embodiment of the present invention is herein only with the object of interest of the first user
It is illustrated with for label.Label in Fig. 4 be in order to find more relevant second users because some first users and
Article that may be uncommon between second user, but have common label, therefore, more second are found by label and is used
Behind family, and more relative articles can be excavated down.As shown in figure 4, the neighbor node of each first user includes article
And label.For example, the first user scores to " Notre Dame de Paris " read in bean cotyledon reading, while tagged such as " wave
It is unrestrained ".So " Notre Dame de Paris " is exactly the adjacent object of the first user, and " romance " is exactly the adjacent label of the first user.
Specifically, being all corresponding with one and the object of the first user-association or one for each first user
Label, a second user, one of second user can have an incidence relation between object or label, these associations are closed
System is got up using line and curve connection, available node relationships schematic diagram as shown in Figure 4.It can be seen that u1、v1、u2And v3Between
Line can be referred to as v3An incidence relation or a paths, and for v3For, there is also u1、t1、u4And v3This
One paths or incidence relation, in practical applications, each object to be recommended can have multiple associations to close by son like this
System or mulitpath.In this way, for v3For, there are two incidence relation or two paths for tool in Fig. 4.
Connection between first user and second user can also pass through an object mark by object or label
Label pair, i.e. object label corresponding with the object, object tag is to for < vl,tm>.Such as shown in Fig. 5 < v1,t1>.For
For Fig. 5, u1、<v1,t1>、u2And v3As v3An incidence relation or a paths, likewise, u1、<v1,t1>、u3With
v3It also is v3An incidence relation or a paths, in this way, for v3For, there are two incidence relation or two for tool in Fig. 5
Paths.
Step 302, using the object of interest of second user as object to be recommended, calculate object to be recommended and closed in each association
Positive rate under system.
Step 303 adds up positive rate of the object to be recommended under multiple incidence relations, obtains the second prediction scoring.
As shown in figure 4, for v3For, u1、v1、u2And v3Between path have a scoring, u1、t1、u4And v3Between
Path have a scoring, the scoring on this two paths is added up to get to u1To v3Second prediction scoring.
Likewise, as shown in figure 5, for v3For, u1、<v1,t1>、u2And v3Between path have a scoring, u1、
<v1,t1>、u3And v3Between path also have a scoring, the scoring on this two paths is added up to get to u1It is right
v3Second prediction scoring.
Optionally, positive rate of the object to be recommended under each incidence relation is preset according to the first parameter preset, second
The front scoring quantity and negative scoring quantity of parameter and each object to be recommended are calculated.
Optionally, the first parameter preset and the second parameter preset are the front scorings according to objects all in recommender system
Probability average determines.
Optionally, the probability of the front scoring of every an object is according to the number for being higher than the object average score in recommender system
The ratio of amount and all scoring quantity of the object determines.
In the embodiment of the present invention, scoring p of the object to be recommended on an incidence relation or a pathsvAccording to as follows
Formula is calculated:
In formula, a is the first parameter preset, and b is the second parameter preset,For object to be recommended front scoring number, i.e.,
Higher than the scoring quantity of the object average score to be recommended,For the negative scoring number of object to be recommended, that is, being lower than should be wait push away
Recommend the scoring quantity of object average score.
The determination of first parameter preset and the second parameter preset can be determined by the following method: assuming that be recommended
The front probability θ of object vvDistribution obey beta distribution θ~Beta (a, b), then for each object to be recommended, take its
Then the average value of all scorings in recommender system calculates the ratio that general comment dosis refracta is occupied higher than the scoring quantity of the average value
Weight, as the front probability θ of the articlev.The front probability of objects all in recommender system is averaged to obtainIn turn
A and b are determined according to the front probability average of all objects, evenEqual to front probability average.Such as institute
The front probability average for having article is 0.6, then a takes 600, b to take 400.Certainly, a and b can also take other numerical value,
Such as a takes 900, b to take 600.For the value of a and b, those skilled in the art can set according to actual needs, as long as making
It meets?.
The practical application of above-mentioned formula is illustrated below by citing:
If the average value of all scorings of article v3 is 3.5, the scoring quantity in all scorings lower than 3.5 is 20, high
In 3.5 scoring quantity be 15, thenIt is available to be recommended by the formula (4)
Scoring of the object on a certain paths is scored according to the path of object to be recommended and is superimposed, it can be deduced that is all not score in the past
Object to be recommended score.For user uiFrom article vlIt sets out to user ujObject v is arrived againmA paths ρ (ui,vl,
uj,vm), it is scored at S ρ (ui,vl,uj,vm)=pvm.As shown in figure 4, u1To v1To u2V is arrived again2Path score beu1It arrives
t1To u4V is arrived again3Path score be pv3。
Optionally, to multiple objects to be recommended of the first determining user, the first user treats recommended the is calculated
Two prediction scorings, comprising: indicate the interaction of users all in recommender system and all objects using the first matrix；By it is multiple to
Positive rate of the recommended under each incidence relation is indicated using the second matrix；According to the first matrix, the transposition of the first matrix
Matrix and the second matrix, are calculated third matrix, and third matrix includes the second pre- assessment that the first user treats recommended
Point.Specifically, may determine that whether the quantity of multiple objects to be recommended of the first determining user reaches threshold value, if reaching threshold
Value, then calculate the positive rate of all objects in recommender system, to form the second matrix；If not up to threshold value calculates determination
The first user multiple objects to be recommended positive rate, thus formed the second matrix, in this case, remaining in recommender system
The positive rate of object is 0.
In the present embodiment, as shown in fig. 6, in recommender system all users and all objects, can use first
Matrix carrys out A expression, and the row of first matrix represents all users in recommender system, and column represent all objects in recommender system
And/or the label of all objects.In first matrix, there is the position interacted (namely to beat user and object and/or label
Position that is excessive or marking label) it is set as 1, not interactive is set as 0.Further, it is also possible to by owning in recommender system
User and the interaction of all objects indicate that the row of the second matrix represents all users in recommender system using the second matrix P,
Column represent positive rate of every an object under each incidence relation in recommender system.In this way, then can directly pass through A × AT×P
A third matrix is calculated, the row of the third matrix represents all users in recommender system, and column represent a certain user couple
The marking or the second prediction scoring of one object.
For example, it is assumed that having user u1, u2, u3 and object v1, v2, v3, user u1 and object v1, v3 have interaction, user
U2 and object v2 has interaction, and user u3 and object v2, v3 have interaction, and adjacency matrix is expressed asMatrix A
Every a line respectively represent user u1, u2, u3, each column respectively represent object v1, v2, v3, then the transposition of A isIt calculates firstDiagonal line is revised as 0, then is obtainedRepresent whether all users in recommender system have incidence relation between any two, wherein 1
Representing has same interest object and/or same label between the row and corresponding two users of the column, 0 represents the row and the column
There is no same interest object and/or same label between corresponding two users；Assuming that Interactive matrix is expressed asResult and Interactive matrix phase then according to matrix multiple principle, after diagonal line is all revised as 0
Multiply, obtains prediction matrixOptionally, in order to consistent with the first prediction scoring value,
Element each in prediction matrix can also be mapped in scoring section.The specific method of mapping is exemplified below: assuming that scoring model
Enclosing is 0~5, then maximum value in R matrix is mapped to 5, R matrix each single item correction value and is calculated by following formula: correction value/initial
The first row first row=5/1*1/2=5/2 in maximum value, such as R in value=5/R.Finally obtain revised prediction matrixEach element in finally obtained prediction matrix be in recommender system each user to each right
The marking of elephant or the second prediction scoring.
It should be noted that all users are referred to and all included in recommender system are pushed away using this in the embodiment of the present invention
The user for recommending the user of system or registering in recommender system, all objects refer to all and user for including in recommender system
There may be the objects of interactive relation.For example, recommender system be purchase system, then all users refer in the purchase system into
The user of row registration or the user for using the purchase system, all objects refer to commodity all in the purchase system, including
The commodity interacted occurred with user, and the commodity interacted never occurred with any user.
The embodiment of the present invention utilizes adjacent node information (object of interest and/or mark including the first user of the first user
Label) and recommender system in object to be recommended positive rate realization collaborative filtering.Compared to traditional method, what the second prediction was scored
Calculating can realize therefore parallel computation can effectively improve computational efficiency by the operation of three matrixes.
Fig. 7 be another embodiment of the present invention provides recommended method flow chart.On the basis of the above embodiments, this implementation
The recommended method that example provides specifically comprises the following steps:
Step 701, based on first prediction scoring and second prediction scoring, calculate the first user to it is each do not score object and
The integrated forecasting of each object to be recommended is scored.
Optionally, based on first prediction scoring and second prediction scoring, calculate the first user to it is each do not score object and
The integrated forecasting of each object to be recommended is scored, comprising: determines the first the activity of the user and the freshness for the object that do not score；Root
According to the freshness of the first the activity of the user and the object that do not score, the weight of the first prediction scoring is determined；According to the first pre- assessment
The weight divided determines the weight of the second prediction scoring；According to the weight of the first prediction scoring, the first prediction scoring, the second pre- assessment
The weight divided and the second prediction scoring determine that the first user treats the integrated forecasting scoring of recommended.Wherein, first user
Liveness is to be determined according to the first user to the level of interaction of object in recommender system and the first adjustment parameter；Do not score object
Freshness is to be determined according to object is not scored by the degree of concern of user in recommender system and the second adjustment parameter；First pre- assessment
The weight divided and the sum of weight of the second prediction scoring are 1.
Specifically, the first the activity of the user can reflect out the first user to the interaction journey of object in recommender system
Degree.First the activity of the user and the first user are positively correlated the level of interaction for the object that scored, i.e., if the first user is pushing away
The historical record recommended in system is more, then represents that the first the activity of the user is higher, and vice versa.First the activity of the user can
To be expressed using following formula:
In formula (5), Nr(u) user u is indicated in the historical record number in recommender system between object, and α is the first adjusting
Parameter can be set based on experience value.
Specifically, the freshness for the object that do not score represent this do not score object by users all in recommender system concern journey
Degree.Do not score object freshness and the object that do not score it is negatively correlated by all degree of concern in recommender system, i.e., if do not commented
Divide object more by interactive record in recommender system, freshness is lower, and vice versa.The freshness of object of not scoring can
To be expressed using following formula:
In formula, Nr(v) for expression object v by interactive number in recommender system, β is the second adjustment parameter, can basis
Empirical value is set.
Further, according to above-mentioned formula (5) and formula (6), then the expression formula that available integrated forecasting is scored:
R (u, v)=w (u, v) Rt(u,v)+(1-w(u,v))·Rc(u,v) (7)
In formula, Rt(u, v) is that the first user scores to the first prediction of the object that do not score, Rc(u, v) is that the first user treats
Second prediction scoring of recommended, w (u, v) is the weight of the first prediction scoring, and 1-w (u, v) second predicts the weight of scoring,
The weight of first prediction scoring and the sum of the weight of the second prediction scoring are 1.
In embodiments of the present invention, the expression formula of the weight w (u, v) of the first prediction scoring is as follows:
It is fresh with the raising of the first the activity of the user or object to be recommended it can be seen from above-mentioned formula (8)
The weight of the reduction of degree, the first prediction scoring gradually decreases.
Step 702 determines that at least one target pushes away according to integrated forecasting scoring in do not score object and object to be recommended
Recommend object.
Specifically, be to calculate the first user to score to each integrated forecasting for not scoring object and each object to be recommended,
And according to integrated forecasting scoring to all do not score object and all object progress descending arrangements to be recommended, and take in the top
Preset quantity object recommendation gives the first user.
The embodiment of the present invention according to the freshness of the first the activity of the user and the object that do not score by being adaptively adjusted
First prediction scoring and the second prediction scoring can fully utilize the calculation method of the first prediction scoring and the second prediction scoring
Advantage.Compared to previous predefined mixed strategy, the method for the embodiment of the present invention it is more flexible with it is reasonable.
Fig. 8 is the structural schematic diagram of recommendation apparatus provided in an embodiment of the present invention.Recommendation dress provided in an embodiment of the present invention
The process flow that the offer of recommended method embodiment can be provided is set, as shown in figure 8, recommendation apparatus 80 includes: the first computing module
81, the second computing module 82 and determining module 83；Wherein, the first computing module 81, for based on the first user in recommender system
Multiple objects that scored the multiple objects that do not score of attribute tensor sum attribute tensor, calculate the first user do not score each
First prediction scoring of object；Second computing module 82 calculates for multiple objects to be recommended to the first determining user
One user scores to the second prediction of each object to be recommended；Determining module 83, for pre- based on the first prediction scoring and second
Assessment point determines at least one target recommended, in multiple do not score object and multiple objects to be recommended to be recommended.
Optionally, the attribute tensor sum of the first computing module 81 object that scored of first user in based on recommender system
Do not score the attribute tensor of object, calculates the first user and scores the first prediction of the object that do not score, is specifically used for: based on having commented
The attribute tensor for dividing the attribute tensor sum of object not score object, calculating have been scored similar between object and the object that do not score
Degree；Scoring based on the similarity and the first user to have scored between object and the object that do not score to the object that scored calculates
First user scores to the first prediction of the object that do not score.
Optionally, the first computing module 81 does not score the attribute of object in the attribute tensor sum based on the object that scored
Amount calculates when having scored similarity between object and the object that do not score, is specifically used for: object and not commenting to will score respectively
Divide object to carry out the attribute tensor that tensor is expressed and carries out tensor resolution respectively；According to scored object and the object that do not score it is each
From tensor resolution as a result, calculating the similarity that has scored between object and the object that do not score.
Optionally, multiple objects to be recommended of the first determining user of 82 pairs of the second computing module calculate the first user couple
When the second prediction scoring of each object to be recommended, it is specifically used for: determines multiple incidence relations of object to be recommended, incidence relation
It is that the first user, the object of interest of the first user and/or the label marked to object of interest and the first user have same interest
Incidence relation between the object of interest of the second user and second user of object and/or same label；By second user
Object of interest calculates positive rate of the object to be recommended under each incidence relation as object to be recommended；Object to be recommended is existed
Positive rate under multiple incidence relations is cumulative, obtains the second prediction scoring.
Optionally, multiple objects to be recommended of the first determining user of 82 pairs of the second computing module calculate the first user couple
When the second prediction scoring of each object to be recommended, it is specifically used for: by the interaction of users and all objects all in recommender system
It is indicated using the first matrix；Positive rate of multiple objects to be recommended under each incidence relation is indicated using the second matrix；Root
According to the first matrix, the transposed matrix of the first matrix and the second matrix, third matrix is calculated, third matrix includes the first user
Treat the second prediction scoring of recommended.
Optionally, positive rate of the object to be recommended under each incidence relation is preset according to the first parameter preset, second
The front scoring quantity and negative scoring quantity of parameter and each object to be recommended are calculated.
Optionally, the first parameter preset and the second parameter preset are the front scorings according to objects all in recommender system
Probability average determines.
Optionally, the probability of the front scoring of every an object is according to the number for being higher than the object average score in recommender system
The ratio of amount and all scoring quantity of the object determines.
Optionally, determining module 83 is based on the first prediction scoring and the second prediction scoring, in multiple objects and more of not scoring
When determining at least one target recommended in a object to be recommended, it is specifically used for: based on the first prediction scoring and the second prediction
Scoring calculates the first user and scores each integrated forecasting for not scoring object and each object to be recommended；According to integrated forecasting
Scoring determines at least one target recommended in do not score object and object to be recommended.
Optionally, determining module 83 be based on first prediction scoring and second prediction scoring, calculate the first user to it is each not
The integrated forecasting scoring of scoring object and each object to be recommended, comprising: determine the first the activity of the user and the object that do not score
Freshness；According to the freshness of the first the activity of the user and the object that do not score, determine the first prediction scoring weight and
The weight of second prediction scoring；According to the weight of the first prediction scoring, the weight and the of the first prediction scoring, the second prediction scoring
Two prediction scorings determine that the first user treats the integrated forecasting scoring of recommended.
Optionally, the first the activity of the user is according to the first user to the level of interaction of object in recommender system and first
Adjustment parameter determines；The freshness of object of not scoring is according to the object that do not score by the degree of concern of user in recommender system and the
Two adjustment parameters determine.
The recommendation apparatus of embodiment illustrated in fig. 8 can be used for executing the technical solution of above method embodiment, realization principle
Similar with technical effect, details are not described herein again.
Fig. 9 is the structural schematic diagram of recommendation apparatus provided in an embodiment of the present invention.The recommendation apparatus can execute recommendation side
The process flow that method embodiment provides, as shown in figure 9, recommendation apparatus 90 includes: memory 91, processor 92, computer program
With communication interface 93；Wherein, computer program is stored in memory 91, and is configured as executing recommended method by processor 92
The process flow that embodiment provides.
The recommendation apparatus of embodiment illustrated in fig. 9 can be used for executing the technical solution of above method embodiment, realization principle
Similar with technical effect, details are not described herein again.
In addition, the embodiment of the present invention also provides a kind of computer readable storage medium, it is stored thereon with computer program, is counted
Calculation machine program is executed by processor the recommended method to realize above-described embodiment.
In several embodiments provided by the present invention, it should be understood that disclosed device and method can pass through it
Its mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of unit, only
A kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or components can combine or
Person is desirably integrated into another system, or some features can be ignored or not executed.Another point, shown or discussed is mutual
Between coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or communication link of device or unit
It connects, can be electrical property, mechanical or other forms.
Unit may or may not be physically separated as illustrated by the separation member, shown as a unit
Component may or may not be physical unit, it can and it is in one place, or may be distributed over multiple networks
On unit.It can some or all of the units may be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
The above-mentioned integrated unit being realized in the form of SFU software functional unit can store and computer-readable deposit at one
In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are used so that a computer
It is each that equipment (can be personal computer, server or the network equipment etc.) or processor (processor) execute the present invention
The part steps of embodiment method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (Read-Only
Memory, ROM), random access memory (Random Access Memory, RAM), magnetic or disk etc. is various to deposit
Store up the medium of program code.
Those skilled in the art can be understood that, for convenience and simplicity of description, only with above-mentioned each functional module
Division progress for example, in practical application, can according to need and above-mentioned function distribution is complete by different functional modules
At the internal structure of device being divided into different functional modules, to complete all or part of the functions described above.On
The specific work process for stating the device of description, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (18)
1. a kind of recommended method characterized by comprising
The attribute of the multiple objects that do not score of attribute tensor sum based on multiple objects that scored of the first user in recommender system
Amount calculates first user and scores the first prediction of the object that do not score；
To multiple objects to be recommended of determining first user, first user is calculated to the of the object to be recommended
Two prediction scorings；
Based on the first prediction scoring and the second prediction scoring, in multiple objects and multiple described wait push away of not scoring
It recommends and determines at least one target recommended in object, to be recommended.
2. the method according to claim 1, wherein described commented based on the multiple of the first user in recommender system
Divide the attribute tensor of the multiple objects that do not score of attribute tensor sum of object, calculates first user to the object that do not score
First prediction scoring, comprising:
Based on multiple scored object and the multiple respective attribute tensors of the object that do not score of the first user in recommender system, calculate
The similarity to have scored between object and the object that do not score；
It has been commented based on the similarity to have scored between object and the object that do not score and first user described
Divide the scoring of object, calculates first user and score the first prediction of the object that do not score.
3. according to the method described in claim 2, it is characterized in that, described commented based on the multiple of the first user in recommender system
Divide object and multiple respective attribute tensors of the object that do not score, calculates described has scored between object and the object that do not score
Similarity, comprising:
Object has been scored and the object that do not score carries out the attribute tensor expressed of tensor and carries out respectively to by described respectively
Tensor resolution；
According to it is described scored object and it is described do not score the respective tensor resolution of object as a result, calculate it is described scored object and
Similarity between the object that do not score.
4. method according to claim 1-3, which is characterized in that described couple of determining first user's is more
A object to be recommended calculates first user and scores the second prediction of the object to be recommended, comprising:
Determine that multiple incidence relations of the object to be recommended, the incidence relation are first user, first user
Object of interest and/or to the label of object of interest mark, with first user with same interest object and/or phase
With the incidence relation between the second user of label and the object of interest of the second user；
Using the object of interest of the second user as the object to be recommended, the object to be recommended is calculated in each pass
Positive rate under connection relationship；
Positive rate of the object to be recommended under multiple incidence relations is added up, the second prediction scoring is obtained.
5. method according to claim 1-3, which is characterized in that described couple of determining first user's is more
A object to be recommended calculates first user and scores the second prediction of the object to be recommended, comprising:
The interaction of users all in the recommender system and all objects is indicated using the first matrix；
Positive rate of the multiple object to be recommended under each incidence relation is indicated using the second matrix；
According to first matrix, the transposed matrix and second matrix of first matrix, third matrix, institute is calculated
Stating third matrix includes that first user scores to the second prediction of the object to be recommended.
6. method according to claim 4 or 5, which is characterized in that the object to be recommended is under each incidence relation
Positive rate be according to the first parameter preset, the second parameter preset and each object to be recommended front scoring quantity and
Negative scoring quantity is calculated.
7. according to the method described in claim 6, it is characterized in that, first parameter preset and second parameter preset are
It is determined according to the probability average of the front scoring of objects all in the recommender system.
8. method according to claim 1-7, which is characterized in that described based on the first prediction scoring and institute
The second prediction scoring is stated, the recommendation pair of at least one target is determined in the multiple do not score object and multiple objects to be recommended
As, comprising:
Based on the first prediction scoring and the second prediction scoring, calculates first user and described do not score pair to each
As the integrated forecasting scoring with each object to be recommended；
Determine that at least one target pushes away in do not score object and the object to be recommended according to integrated forecasting scoring
Recommend object.
9. according to the method described in claim 8, it is characterized in that, described pre- based on the first prediction scoring and described second
Assessment point calculates first user and comments each integrated forecasting for not scoring object and each object to be recommended
Point, comprising:
Determine the freshness of first the activity of the user and the object that do not score；
According to the freshness of first the activity of the user and the object that do not score, the power of the first prediction scoring is determined
Weight and the weight of the second prediction scoring；
According to the weight of the first prediction scoring, the weight and described of the first prediction scoring, the second prediction scoring
Second prediction scoring determines first user respectively to the comprehensive of each do not score object and each object to be recommended
Close prediction scoring.
10. according to the method described in claim 9, it is characterized in that, first the activity of the user is according to described first
User determines the level of interaction of object in the recommender system and the first adjustment parameter；
The freshness for not scoring object be according to it is described do not score object by user in the recommender system degree of concern
It is determined with the second adjustment parameter.
11. a kind of recommendation apparatus characterized by comprising
First computing module, it is multiple not for the attribute tensor sum based on multiple objects that scored of the first user in recommender system
The attribute tensor of scoring object calculates first user and scores the first prediction of the object that do not score；
Second computing module calculates first user couple for multiple objects to be recommended to determining first user
Second prediction scoring of the object to be recommended；
Determining module, for based on the first prediction scoring and the second prediction scoring, in multiple objects and more of not scoring
At least one target recommended is determined in a object to be recommended, to be recommended.
12. device according to claim 11, which is characterized in that first computing module is based in recommender system the
The attribute tensor sum of the object that scored of one user does not score the attribute tensor of object, calculates first user and does not comment described
The the first prediction scoring for dividing object, is specifically used for:
The object of scoring and the respective attribute tensor of the object that do not score based on the first user in recommender system calculate described commented
Divide the similarity between object and the object that do not score；
It has been commented based on the similarity to have scored between object and the object that do not score and first user described
Divide the scoring of object, calculates first user and score the first prediction of the object that do not score.
13. device according to claim 11 or 12, which is characterized in that second computing module is to determining described
Multiple objects to be recommended of one user, when calculating second prediction scoring of first user to each object to be recommended,
It is specifically used for:
Determine that multiple incidence relations of the object to be recommended, the incidence relation are first user, first user
Object of interest and/or to the label of object of interest mark, with first user with same interest object and/or phase
With the incidence relation between the second user of label and the object of interest of the second user；
Using the object of interest of the second user as the object to be recommended, the object to be recommended is calculated in each pass
Positive rate under connection relationship；
Positive rate of the object to be recommended under multiple incidence relations is added up, the second prediction scoring is obtained.
14. device according to claim 11 or 12, which is characterized in that second computing module is to determining described
Multiple objects to be recommended of one user, when calculating second prediction scoring of first user to each object to be recommended,
It is specifically used for:
The interaction of users all in the recommender system and all objects is indicated using the first matrix；
Positive rate of multiple objects to be recommended under each incidence relation is indicated using the second matrix；
According to first matrix, the transposed matrix and second matrix of first matrix, third matrix, institute is calculated
Stating third matrix includes that first user scores to the second prediction of the object to be recommended.
15. the described in any item devices of 1-14 according to claim 1, which is characterized in that the determining module is based on described first
Prediction scoring and the second prediction scoring, determine at least one target in multiple do not score object and multiple objects to be recommended
When recommended, it is specifically used for:
Based on the first prediction scoring and the second prediction scoring, calculates first user and described do not score pair to each
As the integrated forecasting scoring with each object to be recommended；
Determine that at least one target pushes away in do not score object and the object to be recommended according to integrated forecasting scoring
Recommend object.
16. device according to claim 15, which is characterized in that the determining module be based on it is described first prediction scoring and
The second prediction scoring calculates first user to the comprehensive of each do not score object and each object to be recommended
Close prediction scoring, comprising:
Determine the freshness of first the activity of the user and the object that do not score；
According to the freshness of first the activity of the user and the object that do not score, the first prediction scoring is determined respectively
Weight and it is described second prediction scoring weight；
According to the weight of the first prediction scoring, the weight and described of the first prediction scoring, the second prediction scoring
Second prediction scoring obtains first user respectively to the comprehensive of each do not score object and each object to be recommended
Close prediction scoring.
17. a kind of recommendation apparatus characterized by comprising
Memory；
Processor；And
Computer program；
Wherein, the computer program stores in the memory, and is configured as being executed by the processor to realize such as
Any method in claim 1-10.
18. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
Such as claim 1-10 described in any item methods are realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910628953.5A CN110321490B (en) | 2019-07-12 | 2019-07-12 | Recommendation method, device, equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910628953.5A CN110321490B (en) | 2019-07-12 | 2019-07-12 | Recommendation method, device, equipment and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110321490A true CN110321490A (en) | 2019-10-11 |
| CN110321490B CN110321490B (en) | 2020-05-22 |
Family
ID=68121968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910628953.5A Active CN110321490B (en) | 2019-07-12 | 2019-07-12 | Recommendation method, device, equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321490B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112948701A (en) * | 2021-04-16 | 2021-06-11 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
| CN113112148A (en) * | 2021-04-09 | 2021-07-13 | 北京邮电大学 | Evaluation method for evaluation result of recommendation system model and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101685458A (en) * | 2008-09-27 | 2010-03-31 | 华为技术有限公司 | Recommendation method and system based on collaborative filtering |
| CN102982466A (en) * | 2012-07-17 | 2013-03-20 | 华东师范大学 | Graded forecasting method based on user liveness |
| CN105025091A (en) * | 2015-06-26 | 2015-11-04 | 南京邮电大学 | Shop recommendation method based on position of mobile user |
| CN106649657A (en) * | 2016-12-13 | 2017-05-10 | 重庆邮电大学 | Recommended system and method with facing social network for context awareness based on tensor decomposition |
| CN106897911A (en) * | 2017-01-10 | 2017-06-27 | 南京邮电大学 | A kind of self adaptation personalized recommendation method based on user and article |
-
2019
- 2019-07-12 CN CN201910628953.5A patent/CN110321490B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101685458A (en) * | 2008-09-27 | 2010-03-31 | 华为技术有限公司 | Recommendation method and system based on collaborative filtering |
| CN102982466A (en) * | 2012-07-17 | 2013-03-20 | 华东师范大学 | Graded forecasting method based on user liveness |
| CN105025091A (en) * | 2015-06-26 | 2015-11-04 | 南京邮电大学 | Shop recommendation method based on position of mobile user |
| CN106649657A (en) * | 2016-12-13 | 2017-05-10 | 重庆邮电大学 | Recommended system and method with facing social network for context awareness based on tensor decomposition |
| CN106897911A (en) * | 2017-01-10 | 2017-06-27 | 南京邮电大学 | A kind of self adaptation personalized recommendation method based on user and article |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113112148A (en) * | 2021-04-09 | 2021-07-13 | 北京邮电大学 | Evaluation method for evaluation result of recommendation system model and electronic equipment |
| CN113112148B (en) * | 2021-04-09 | 2022-08-05 | 北京邮电大学 | Evaluation method for evaluation result of recommendation system model and electronic equipment |
| CN112948701A (en) * | 2021-04-16 | 2021-06-11 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
| CN112948701B (en) * | 2021-04-16 | 2023-10-20 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110321490B (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105095267B (en) | Recommendation method and device for user participation items | |
| CN103309866B (en) | The method and apparatus for generating recommendation results | |
| CN110532479A (en) | A kind of information recommendation method, device and equipment | |
| US9082086B2 (en) | Adaptively learning a similarity model | |
| CN109840730B (en) | Method and device for data prediction | |
| CN104331411B (en) | The method and apparatus of recommended project | |
| CN107248095A (en) | Recommend method and device | |
| CN105354202B (en) | Data push method and device | |
| CN106997347A (en) | Information recommendation method and server | |
| CN110097433A (en) | Recommended method, device, equipment and storage medium based on attention mechanism | |
| CN109242612A (en) | A kind of method and apparatus of Products Show | |
| JP6976207B2 (en) | Information processing equipment, information processing methods, and programs | |
| CN112633973A (en) | Commodity recommendation method and related equipment thereof | |
| US20120296776A1 (en) | Adaptive interactive search | |
| CN108205775A (en) | The recommendation method, apparatus and client of a kind of business object | |
| JP6405343B2 (en) | Information processing apparatus, information processing method, and program | |
| CN111611499B (en) | Collaborative filtering method, collaborative filtering device and collaborative filtering system | |
| CN109064293A (en) | Method of Commodity Recommendation, device, computer equipment and storage medium | |
| CN110263821A (en) | Transaction feature generates the generation method and device of the training of model, transaction feature | |
| CN109410001A (en) | A kind of Method of Commodity Recommendation, system, electronic equipment and storage medium | |
| CN113643103A (en) | Product recommendation method, device, equipment and storage medium based on user similarity | |
| Di Bartolomeo et al. | There is more to streamgraphs than movies: Better aesthetics via ordering and lassoing | |
| CN111695024A (en) | Object evaluation value prediction method and system, and recommendation method and system | |
| CN110321490A (en) | Recommended method, device, equipment and computer readable storage medium | |
| CN110473073A (en) | The method and device that linear weighted function is recommended |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |