CN106777123B - A kind of group's recommended method based on two-way tensor resolution model - Google Patents
A kind of group's recommended method based on two-way tensor resolution model Download PDFInfo
- Publication number
- CN106777123B CN106777123B CN201611168532.1A CN201611168532A CN106777123B CN 106777123 B CN106777123 B CN 106777123B CN 201611168532 A CN201611168532 A CN 201611168532A CN 106777123 B CN106777123 B CN 106777123B
- Authority
- CN
- China
- Prior art keywords
- group
- user
- product
- indicate
- parameter
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000002452 interceptive effect Effects 0.000 claims abstract description 13
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000006116 polymerization reaction Methods 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 4
- 239000008186 active pharmaceutical agent Substances 0.000 claims description 3
- 230000000977 initiatory effect Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 7
- 238000006243 chemical reaction Methods 0.000 abstract 1
- 238000012360 testing method Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- NJPPVKZQTLUDBO-UHFFFAOYSA-N novaluron Chemical compound C1=C(Cl)C(OC(F)(F)C(OC(F)(F)F)F)=CC=C1NC(=O)NC(=O)C1=C(F)C=CC=C1F NJPPVKZQTLUDBO-UHFFFAOYSA-N 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 230000009194 climbing Effects 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 241000854350 Enicospilus group Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 235000015170 shellfish Nutrition 0.000 description 1
- 241000894007 species Species 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of group's recommended methods based on two-way tensor resolution model, comprising: 1) defines the interactive relation D for indicating group G, user U and product IS；2) tensor resolution model is constructed；3) conversion solution is carried out to tensor resolution model using Bayes's personalized ordering method, obtains the parameters value in tensor resolution model；4) g-th of group is obtained to group's preference of i-th of productAnd it traverses all commodity and obtains g-th of group to group's preference of all products；5) group preference of g-th of group to all products is subjected to descending sort, and top n product is selected to be pushed to g-th of group as recommended products list.Individual preference is modeled as two-way process by the present invention, can effectively reflect the true forming process of individual preference, improves the precision of group's recommendation, and have preferable robustness.
Description
Technical field
The invention belongs to groups to recommend field, and specifically a kind of group based on two-way tensor resolution model recommends (BTF-
GR) method
Background technique
Social networks early has become a key activities part in social media environment, and user can be certainly in social network-i i-platform
Hair composition group, and the preference for capturing each group is beneficial to us and carries out deep behavioural analysis to user group, into
It and is that target product and service are recommended by group.But group recommends to be different from individual recommendation, because group is usually by having multiplicity
Change user's composition of preference, so group recommends to be not easy to realize.The core missions that group recommends seek to polymerize individual preference with
Generate group's recommendation results, existing aggregation strategy be all the formation of group's preference is modeled as to an one-way process, though
Ran Ji group preference be the polymerization of individual preference as a result, but in true social environment group and individual there are reciprocation,
And existing aggregation strategy can not capture the influence that group generates individual preference, and group is caused to recommend accuracy not high；User's
Diversification is presented in the group property of individual character and group, so influence having differences property of the group to individual preference, but it is existing
Polymerization cannot portray the difference of this influence, so that group's recommendation results are unsatisfactory；And when existing group's recommended method is applied to
When sparse data set, performance is recommended to be decreased obviously, so existing group's recommended method not can be well solved under big data environment
Sparse Problem, do not have preferable robustness.
In recent years, be decomposed into most popular recommended technology, by decompose user and product can be described interact work
With, although tensor resolution is applied to achieve preferable recommendation performance in the recommendations problems such as label recommendations, ad click prediction,
But the time complexity of tensor resolution is high, data scale is limited to, so being not ideally suited under big data environment
Group recommends problem.
Summary of the invention
The present invention proposes a kind of based on two-way tensor resolution model for shortcoming existing for existing group's Generalization bounds
Group's recommended method, to embody the reciprocation of group and individual in the modeling of individual preference, and it is inclined to individual to portray group
The otherness influenced well forms accurate group's preference by polymerizeing accurately individual preference, so that promoting group recommends precision, and
Recommend suitable for group accurate, stable under extensive, sparse data environment.
In order to achieve the above objectives, the technical solution adopted by the present invention are as follows:
A kind of the characteristics of group's recommended method based on two-way tensor resolution model of the invention is to carry out in accordance with the following steps:
Step 1: defining the interactive relation for indicating group G, user U and product I:The friendship
Mutual relation DSIn, G={ G1,...,Gg,...,G|G|Indicate group's set, GgIndicate any g-th of group, 1≤g≤| G |；U
={ U1,...,Uu,...,U|U|Indicate user's set, UuIndicate u-th of user；I={ I1,...,Ii,...,
I|I|Indicate product set, IiIndicate i-th of product,
Step 2: using u-th of user in g-th of group of formula (1) building to the tensor resolution model of i-th of product
In formula (1),Indicate that u-th of user is to the individual preference of i-th of product in g-th of group；Indicate u-th of user to itself individual preferenceInfluence,Indicate u-th of use
Individual preference of g-th of the group belonging to family to u-th of userInfluence, biIndicate the deviation of i-th of product；Table
Show weight of the individual character to u-th of user of u-th of user；Uu,lIndicate first of hidden variable of u-th of user, kuIndicate the
The hidden variable number of u user；Indicate first of hidden variable that i-th of the product interacted is generated with u-th of user；It indicates
Weight of the group property of g-th of group to u-th of user belonging to u-th of user；Gg,mIndicate m-th of g-th of group it is hidden
Variable, kgIndicate the hidden variable number of g-th of group；It indicates m-th that i-th of the product interacted is generated with g-th of group
Hidden variable；
Step 3: using Bayes's personalized ordering method to the tensor resolution modelIt optimizes, obtains
To the tensor resolution modelIn parameters value；
Step 4: obtaining g-th of group to group's preference of i-th of product using formula (2)To obtain g-th group
Group preference of the body to all products:
In formula (2), Δ () is average polymerization function；
Step 4: group preference of g-th of the group to all products is carried out descending sort, and select top n group
Product corresponding to body preference is pushed to g-th of group as recommended products list.
The characteristics of group's recommended method of the present invention, lies also in, and the step 3 is to carry out as follows:
Step 3.1 obtains the tensor resolution model using formula (3)Objective function
In formula (3),Indicate that u-th of user is to i-th in g-th of groupaThe individual preference of a product with to i-thbIt is a
The difference of the individual preference of product；Indicate i-thaA product belongs to the positive feedback set of u-th of user in g-th of group,
The positive feedback collection is combined into all product set interacted with u-th of user in g-th of group；Indicate i-thbIt is a
Product belongs to negative-feedback and the missing value set of u-th of user in g-th of group, and negative-feedback and missing value set are and g-th
All product set that u-th of user did not interact in group；Indicate logistic function；Described in Θ expression
Tensor resolution modelIn parameter sets, and haveλΘIndicate regularization ginseng
Number；
Step 3.2, initiation parameter set Θ and regularization parameter λΘ；
All products in step 3.3, traversal g-th of group in the positive feedback set of u-th of user, and traversing
One is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during each product
Product；
Step 3.4 acquires parameter in the parameter sets Θ using stochastic gradient descent method's
Gradient；In parameterWithWhen for definite value, to the parameterIt is iterated update, Zhi Daoshou respectively
Until holding back, to obtain parameterOptimal value；
All products in step 3.5, traversal g-th of group in the positive feedback set of u-th of user, and traversing
A product is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during each product；
Step 3.6 acquires parameter in the parameter sets Θ using stochastic gradient descent methodWithGradient；?
ParameterWhen for optimal value, to the parameterWithIt is iterated update respectively, until converging to
Only, to obtain parameterWithOptimal value.
Compared with the prior art, the beneficial effects of the present invention are embodied in:
1, the formation of group's preference is modeled as polymerizeing individual preference and group to individual preference generation shadow for the first time by the present invention
Group's preference is modeled as polymerizeing the one-way process of individual preference, this hair by loud two-way process compared to existing group's recommended method
Bright idea about modeling more meets the real scene of group's preference formation, and the present invention not only can capture the friendship between group and individual
Interaction, but also the otherness that group has an impact the individual preference of different user can be portrayed, to effectively increase
Group recommends precision, obtains satisfied group's recommendation results.
2, the present invention solves tensor point by the way that user, group, product three are modeled as the tensor resolution interacted in pairs
High time complexity problem existing for model is solved, while tensor resolution model proposed by the present invention is due to being integrated with group's preference,
The negative effect of Sparse can be effectively reduced, so the invention enables tensor resolution models can be applied to big data environment
Under group recommend problem, tensor resolution model interactive in pairs can obtain higher in linear time complexity in the present invention
Forecast quality.
3, there are a large amount of sparse hidden feedback data under big data environment, directly pass through prediction preference score value solving model
There are relatively large deviations for the individual preference of method prediction, and so as to cause precision and satisfaction decline is recommended, and the present invention seeks model
Solution is converted into sequencing problem, and sort method has good adaptability for sparse hidden feedback data, can be obtained accurately partially
Good sequence.The present invention can get accurately individual ordering of optimization preference, Jin Erju using Bayes's personalized ordering method solving model
It is combined into accurate group's ordering of optimization preference, so effectively improving the accuracy and satisfaction of group's recommendation.
4, the present invention is the personalized weight of user setting, to capture individual preference and group influence to different user
Differentiation effect, the setting of personalized weight is so that the real situation that model is more formed close to individual preference, helps to obtain
Better group recommends precision.
5, true group recommends in environment, and the strategy being usually taken is to provide a recommendation list as long as possible to realize to the greatest extent
It may mostly cover the preference of all users in group.When recommendation list is longer, group's recommended method proposed by the invention is not only
With preferable robustness, and performance is more excellent, so the present invention is suitable for group and recommends environment.
6, large-scale groups are because preference diversity increases, so that the recommendation for large-scale groups is very difficult and not smart
Really, group's recommended method proposed by the present invention all has preferable robustness for the group of different scales.
7, the present invention can be used for the digital products such as the physical products such as household electrical appliances and food, music and film, tourism route and degree
Group's recommender system of the service products such as vacation arrangement, can use in platforms such as the webpages and APP of computer and mobile phone, have a wide range of application
It is general.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 for the present invention mentioned method and conventional method comparison diagram；
Fig. 3 is that data of the invention indicate and solve scheme；
Fig. 4 a is the Average Accuracy mean value comparison diagram of the present invention with benchmark algorithm；
Fig. 4 b is the recall rate comparison diagram that the recommendation list length of the invention with benchmark algorithm is 5；
Fig. 4 c is the average sequence inverse comparison diagram of the present invention with benchmark algorithm；
Fig. 5 a is the Average Accuracy mean value comparison diagram of personalized weight and fixed weight；
Fig. 5 b is the recall rate comparison diagram that the recommendation list length of personalized weight and fixed weight is 5；
Fig. 5 c is the average sequence inverse comparison diagram of personalized weight and fixed weight.
Specific embodiment
Tensor resolution model proposed by the present invention is a kind of two-way pairs of interactive tensor resolution model, i.e. group's preference
Be formed as two-way process, and user, group, product three are modeled as pairs of interactive relation.Group proposed by the present invention pushes away
Method is recommended to establish on the basis of following two points are assumed:
1, each user in a group has a kind of inherent preference for product.Each user is for product
Inherent preference may be subjected to the influence of affiliated group；
2. influence of the group for individual preference is dramatically different between users.The final preference of user is inherent inclined
Good and group influence comprehensive function result.
In the present embodiment, as shown in Figure 1, a kind of group's recommended method based on two-way tensor resolution model, according to following step
It is rapid to carry out:
Step 1: defining the interactive relation for indicating group G, user U and product I:Interactive relation
DSIn, G={ G1,...,Gg,...,G|G|Indicate group's set, GgIndicate any g-th of group, 1≤g≤| G |；U=
{U1,...,Uu,...,U|U|Indicate user's set, UuIndicate u-th of user；I={ I1,...,Ii,...,
I|I|Indicate product set, IiIndicate i-th of product,Traditional group's preference modeling only considers user U and product I
Interaction, and the present invention not only allows for the interaction of user U and product I, while having also contemplated the friendship between group G and user U
Mutually, and for this interactive intensity of different user have differences, group's preference idea about modeling proposed by the present invention with
The difference of traditional group's preference idea about modeling is as shown in Figure 2.
Step 2: by the interactive relation of group G, user U and product I carry out it is interactive in pairs decompose, group G, user U and
The data of product I indicate and pairs of interactive tensor resolution model proposed by the present invention is as shown in figure 3, Fig. 3 also shows this hair
The difference of the tensor resolution model of bright proposition and traditional tensor resolution model recycles formula (1) can construct in g-th of group
Tensor resolution model of u-th of user to i-th of product
In formula (1),Indicate that u-th of user is to the individual preference of i-th of product in g-th of group；Indicate u-th of user to itself individual preferenceInfluence,Indicate u-th of use
Individual preference of g-th of the group belonging to family to u-th of userInfluence, biIndicate the deviation of i-th of product；Table
Show weight of the individual character to u-th of user of u-th of user；Uu,lIndicate first of hidden variable of u-th of user, kuIndicate the
The hidden variable number of u user；Indicate first of hidden variable that i-th of the product interacted is generated with u-th of user；It indicates
Weight of the group property of g-th of group to u-th of user belonging to u-th of user；Gg,mIndicate m-th of g-th of group it is hidden
Variable, kgIndicate the hidden variable number of g-th of group；It indicates m-th that i-th of the product interacted is generated with g-th of group
Hidden variable；
Step 3: using Bayes's personalized ordering method to tensor resolution modelIt optimizes, is opened
Measure decomposition modelIn parameters value, i.e., convert the direct solution of individual preference to the sequence of individual preference；
Step 3.1 solves tensor resolution modelU-th of user in i.e. given g-th of group, acquires u-th of use
The optimized parameter at family, so showing that the learning objective of tensor resolution model is exactly to maximize according to Bayes's personalized ordering method
Posterior probability in formula (2)
In formula (2), Θ indicates tensor resolution modelIn parameter sets, and have Indicate that u-th of user in g-th of group arranges the preference of all products
Sequence；Indicate individual preference likelihood of u-th of the user in g-th of the group obtained by sample set to all products
Function；The priori knowledge of expression parameter.Wherein likelihood functionIt can be expressed as formula (3):
In formula (3),Indicate u-th of user in g-th of group in the ordering of optimization preference of all products, productCome productFront, and the sequence of each pair of product is independently of the sequence of other products pair；It indicates g-th
U-th of user of group is compared to product ibFor productIndividual preference probability, while formula (4) can be expressed as:
In formula (4),Indicate that u-th of user is to i-th in g-th of groupaThe individual preference of a product with to i-thbIt is a
The difference of the individual preference of product；Indicate logistic function.According to this transform mode, and assume parameter
Priori knowledgeObeying mean value is 0, and covariance matrix is ∑ΘGaussian Profile, then in formula (2) posterior probability to number form
Formula can simplify as formula (5):
Tensor resolution modelLearning objective be just converted into minimize formula (6) objective function
In formula (6),Indicate i-thaA product belongs to the positive feedback set of u-th of user in g-th of group, positive and negative
Feedback collection is combined into all product set interacted with u-th of user in g-th of group；Indicate i-thbA product belongs to
The negative-feedback of u-th of user and missing value set in g-th of group, negative-feedback and missing value set are and u in g-th of group
All product set that a user did not interact；λΘIndicate regularization parameter；
Step 3.2, initiation parameter set Θ and regularization parameter λΘ；
All products in g-th step 3.3, traversal of group in the positive feedback set of u-th of user, and it is each in traversal
A product is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during product；
Step 3.4 acquires parameter in parameter sets Θ using stochastic gradient descent methodLadder
Degree；In parameterWithWhen for definite value, to parameterIt is iterated update respectively, until convergence,
To obtain parameterOptimal value；
All products in g-th step 3.5, traversal of group in the positive feedback set of u-th of user, and it is each in traversal
A product is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during product；
Step 3.6 acquires parameter in parameter sets Θ using stochastic gradient descent methodWithGradient；In parameterWhen for optimal value, to parameterWithIt is iterated update respectively, until convergence, to obtain
Obtain parameterWithOptimal value.
Step 4: obtaining g-th of group to group's preference of i-th of product using formula (7)To obtain g-th group
Group preference of the body to all products:
In formula (7), Δ () is average polymerization function；
Step 4: group preference of g-th of group to all products is carried out descending sort, and select top n group inclined
Good corresponding product is pushed to g-th of group as recommended products list.
Experimental demonstration is carried out for the method for the present invention, is specifically included:
1) prepare standard data set
The present invention use CiteULike and Last.fm the two recommend the widely used data set in field as standard
Data set verifies the performance of group's recommended method proposed by the present invention.First data set CiteULike is a scientific researcher
On-line communities website.On CiteULike, label for labelling academic article is can be used in scholar, while can also create and be added
Group shares the article of reference.Testing the CiteULike data set used includes 130321 " group-user-product " ternarys
Group, 11168 articles therein are formed by 1310 user annotations from 584 groups, and average population size is 5.4, from
Preceding 10 articles are selected in each group as test set, remaining data are as training set；Second data set Last.fm be
A on-line communities website towards musomania.On Last.fm, musomanias can mark musician or song, creation or
Person's grade similar with having it is humanoid at group.It tests the Last.fm data set used and contains 317907 " groups-user-production
Product " triple, 1992 musicians therein are formed by 3605 user annotations from 2716 groups, and equal group size is
21.2, select preceding 10 musicians in each group as test set, remaining data are used for training set.
2) evaluation index
Use length of recommended for the recall rate (Rec N) of N, Average Accuracy mean value (MAP) and the average inverse (MRR) that sorts
Evaluation index as this experiment.Recall rate assessment group's recommender system returns to the ability of all Related products, and Average Accuracy is equal
Value and the average sequence accuracy reciprocal for disclosing the Related product in recommendation list.Recommendation list length is the recall rate Rec@N of N
Calculation formula are as follows:
In formula (8), NrelatedIt is the product number in sorted lists and test set while occurred；N is phase in test set
Close the number of product.
Average Accuracy mean value computation formula are as follows:
In formula (9), wherein | G | it is the number of group in test set,It is an indicator variable, if groupRecommendation
The product of ranking n-th also appears in test set in list, then value is 1, remaining situation is all 0.The recommendation of expression
As a result accuracy rate, formula are as follows:
In formula (10), whereinIt is the product number in sorted lists and test set while occurred,It is to recommend column
The number of Related product in table.
Average sequence inverse is that the inverse company of first correct product of sequence multiplies, calculation formula are as follows:
3) it is tested on standard data set
In order to verify the validity of institute's climbing form type of the present invention, we by it is proposed by the present invention be used for group recommendation two-way tensor
It decomposes (BTF-GR) model and 4 kinds of pedestal methods is compared, 4 kinds of pedestal methods are as follows: the collaborative filtering based on user
(UserCF) arest neighbors (UserKNN) algorithm of method-based on user, collaborative filtering (ItemCF) method-based on product
Arest neighbors (ItemKNN) algorithm based on product, matrix decomposition (IMF) method based on hidden feedback, the shellfish based on matrix decomposition
This personalized ordering (BPRMF) method of leaf.It is modeled on CiteULike data set and Last.fm data set with 5 kinds of methods
And recommendation, and recommendation results are compared.Experimental result is as shown in Fig. 4 a, Fig. 4 b, Fig. 4 c.Compared with 4 kinds of pedestal methods, this
Group's recommended method that invention proposes all more preferably is recommended precision on Last.fm and CiteULike data set, and for
Sparse hidden feedback data, performance advantage of the invention are more obvious.
In order to detect influence of the personalized weight for institute's climbing form type of the present invention, we are unified by being arranged for all individuals
Fixed weight construct a receptor model, as control experiment.Fig. 5 a, Fig. 5 b, solid line indicates the present invention in Fig. 5 c
Two-way tensor resolution (BTF-GR) model as a result, dotted line is the result of receptor model.As the result is shown in CiteULike data
On collection and Last.fm data set, two-way tensor resolution (BTF-GR) model is always better than the receptor model with fixed weight.This
A result explanation, user preference and group influence play different effects in the forming process of personal preference, and two-way tensor
This species diversity can be captured by decomposing (BTF-GR) model, preferably recommend precision to obtain.
For the robustness for verifying institute's climbing form type of the present invention, and understand and the robustness comparative situation of 4 kinds of pedestal methods, I
By changing recommendation list length and population size, separately designed 2 experimental groups and verified.Experimental result shows this hair
Bright two-way tensor resolution (BTF-GR) model in recommendation list length and population size all with have preferable robustness,
And precision is recommended to be better than 4 kinds of pedestal methods, while the two-way tensor resolution (BTF-GR) longer, of the invention when recommendation list always
Model performance is more excellent, i.e., the present invention is suitable for group and recommends environment.
Claims (1)
1. a kind of group's recommended method based on two-way tensor resolution model, it is characterized in that carrying out in accordance with the following steps:
Step 1: defining the interactive relation for indicating group G, user U and product I:The interactive relation
In DS, G={ G1,...,Gg,...,G|G|Indicate group's set, GgIndicate any g-th of group, 1≤g≤| G |；U=
{U1,...,Uu,...,U|U|Indicate user's set, UuIndicate u-th of user；1≤u≤|U|；I={ I1,...,Ii,...,
I|I|Indicate product set, IiIndicate i-th of product, 1≤i≤| I |；
Step 2: using u-th of user in g-th of group of formula (1) building to the tensor resolution model of i-th of product
In formula (1),Indicate that u-th of user is to the individual preference of i-th of product in g-th of group；Table
Show u-th of user to itself individual preferenceInfluence,It indicates belonging to u-th of user g-th
Individual preference of the group to u-th of userInfluence, biIndicate the deviation of i-th of product；Indicate u-th of user's
Weight of the individual character to u-th of user；Uu,lIndicate first of hidden variable of u-th of user, kuIndicate the hidden change of u-th of user
Measure number；Indicate first of hidden variable that i-th of the product interacted is generated with u-th of user；Indicate u-th of user institute
Weight of the group property of g-th of the group belonged to u-th of user；Gg,mIndicate m-th of hidden variable of g-th of group, kgIt indicates
The hidden variable number of g-th of group；Indicate m-th of hidden variable that i-th of the product interacted is generated with g-th of group；
Step 3: using Bayes's personalized ordering method to the tensor resolution modelIt optimizes, obtains institute
State tensor resolution modelIn parameters value；
Step 3.1 obtains the tensor resolution model using formula (3)Objective function
In formula (3),Indicate that u-th of user is to i-th in g-th of groupaThe individual preference of a product with to i-thbA product
Individual preference difference；ia∈IguIndicate i-thaA product belongs to the positive feedback set of u-th of user in g-th of group, institute
It states positive feedback collection and is combined into all product set interacted with u-th of user in g-th of group；ib∈I\IguIndicate i-thbA production
Product belong to negative-feedback and the missing value set of u-th of user in g-th of group, and negative-feedback and missing value set are and g-th group
All product set that u-th of user did not interact in body；Indicate logistic function；Θ indicates described
Measure decomposition modelIn parameter sets, and haveλΘIndicate regularization parameter；
Step 3.2, initiation parameter set Θ and regularization parameter λΘ；
All products in step 3.3, traversal g-th of group in the positive feedback set of u-th of user, and it is each in traversal
A product is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during product；
Step 3.4 acquires parameter U in the parameter sets Θ using stochastic gradient descent methodu,l,Gg,m,biLadder
Degree；In parameterWithWhen for definite value, to the parameter Uu,l,Gg,m,biIt is iterated update respectively, until convergence
Until, to obtain parameter Uu,l,Gg,m,biOptimal value；
All products in step 3.5, traversal g-th of group in the positive feedback set of u-th of user, and it is each in traversal
A product is arbitrarily selected from the negative-feedback of u-th of user in g-th of group and missing value set during product；
Step 3.6 acquires parameter in the parameter sets Θ using stochastic gradient descent methodWithGradient；In parameter
Uu,l,Gg,m,biWhen for optimal value, to the parameterWithIt is iterated update respectively, until convergence, from
And obtain parameterWithOptimal value；
Step 4: obtaining g-th of group to group's preference of i-th of product using formula (2)To obtain g-th of group pair
Group's preference of all products:
In formula (2), Δ () is average polymerization function；
Step 4: group preference of g-th of the group to all products is carried out descending sort, and select top n group inclined
Good corresponding product is pushed to g-th of group as recommended products list.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611168532.1A CN106777123B (en) | 2016-12-16 | 2016-12-16 | A kind of group's recommended method based on two-way tensor resolution model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611168532.1A CN106777123B (en) | 2016-12-16 | 2016-12-16 | A kind of group's recommended method based on two-way tensor resolution model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106777123A CN106777123A (en) | 2017-05-31 |
| CN106777123B true CN106777123B (en) | 2019-07-19 |
Family
ID=58892231
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611168532.1A Active CN106777123B (en) | 2016-12-16 | 2016-12-16 | A kind of group's recommended method based on two-way tensor resolution model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106777123B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833117B (en) * | 2017-12-13 | 2020-03-17 | 合肥工业大学 | Bayesian personalized sorting recommendation method considering tag information |
| CN110209905A (en) | 2018-12-20 | 2019-09-06 | 腾讯科技（深圳）有限公司 | Label recommendation method, device and readable medium |
| CN110209946B (en) * | 2019-06-10 | 2021-03-09 | 合肥工业大学 | Social and community-based product recommendation method, system and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103313108A (en) * | 2013-06-14 | 2013-09-18 | 山东科技大学 | Smart TV program recommending method based on context aware |
-
2016
- 2016-12-16 CN CN201611168532.1A patent/CN106777123B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103313108A (en) * | 2013-06-14 | 2013-09-18 | 山东科技大学 | Smart TV program recommending method based on context aware |
Non-Patent Citations (2)
| Title |
|---|
| An Incremental Tensor Factorization Approach for Web Service Recommendation;Wangcai Zhang 等;《2014 IEEE International Conference on Data Mining Workshop》;20141231;第346页-351页 |
| 基于张量分解的个性化标签推荐算法;李贵 等;《计算机科学》;20150228;第42卷(第2期);第267页-273页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106777123A (en) | 2017-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106651519B (en) | Personalized recommendation method and system based on label information | |
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| CN103325061B (en) | A kind of community discovery method and system | |
| CN102982107B (en) | A kind of commending system optimization method merging user, project and context property information | |
| CN110503531A (en) | The dynamic social activity scene recommended method of timing perception | |
| CN106649657A (en) | Recommended system and method with facing social network for context awareness based on tensor decomposition | |
| CN107256241B (en) | Movie recommendation method for improving multi-target genetic algorithm based on grid and difference replacement | |
| CN103632290A (en) | Recommendation probability fusion based hybrid recommendation method | |
| CN112115377A (en) | Graph neural network link prediction recommendation method based on social relationship | |
| CN106777123B (en) | A kind of group's recommended method based on two-way tensor resolution model | |
| CN105302873A (en) | Collaborative filtering optimization method based on condition restricted Boltzmann machine | |
| CN107025311A (en) | A kind of Bayes's personalized recommendation method and device based on k nearest neighbor | |
| CN109214407A (en) | Event detection model, calculates equipment and storage medium at method, apparatus | |
| CN108563690A (en) | A kind of collaborative filtering recommending method based on object-oriented cluster | |
| CN113592609A (en) | Personalized clothing matching recommendation method and system using time factors | |
| CN112380433A (en) | Recommendation meta-learning method for cold-start user | |
| CN110334278A (en) | A kind of web services recommended method based on improvement deep learning | |
| CN109754122A (en) | A kind of Numerical Predicting Method of the BP neural network based on random forest feature extraction | |
| CN110321492A (en) | A kind of item recommendation method and system based on community information | |
| CN115545832A (en) | Commodity search recommendation method and device, equipment and medium thereof | |
| CN113656699B (en) | User feature vector determining method, related equipment and medium | |
| Ji et al. | Large-scale comb-k recommendation | |
| Liu | Music trend prediction based on improved LSTM and random forest algorithm | |
| CN106055715B (en) | A kind of arest neighbors collaborative filtering method expanded based on product item feature | |
| Bhargavi et al. | Comparative study of consumer purchasing and decision pattern analysis using pincer search based data mining method |
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 |