CN110334286A - A kind of personalized recommendation method based on trusting relationship - Google Patents
A kind of personalized recommendation method based on trusting relationship Download PDFInfo
- Publication number
- CN110334286A CN110334286A CN201910633177.8A CN201910633177A CN110334286A CN 110334286 A CN110334286 A CN 110334286A CN 201910633177 A CN201910633177 A CN 201910633177A CN 110334286 A CN110334286 A CN 110334286A
- Authority
- CN
- China
- Prior art keywords
- user
- trust
- value
- matrix
- trusting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 3
- 230000007423 decrease Effects 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 abstract description 5
- 230000000694 effects Effects 0.000 abstract description 4
- 230000006399 behavior Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000000644 propagated effect Effects 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
- G06F16/9536—Search customisation based on social or collaborative filtering
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of personalized recommendation methods based on trusting relationship.It is related to recommended method field, improved faith mechanism is dissolved into Collaborative Filtering Recommendation Algorithm by this method by establishing the collaborative filtering model based on trust.Trust value is calculated by global trusting and local trust and is obtained, and wherein belief propagation mechanism is utilized in local trust, calculates the direct trust value and indirect trust values of user;Global trusting is calculated by the way of trusting digraph, then merges trust value with scoring similarity, solves the problems, such as Deta sparseness, Malicious recommendation.By gradient descent algorithm, user characteristics vector sum item feature vector is calculated, prediction score value is generated, further improves the precision of recommender system.The present invention is under conditions of meeting recommendation, in conjunction with users to trust relationship and user's similitude, so that recommendation effect greatly improves.
Description
Technical field
The present invention relates to recommended method field more particularly to a kind of personalized recommendation methods based on trusting relationship.
Background technique
Become recently as the various aspects of internet, the fast development of Internet of Things (IoT) and e-commerce, people's life
The problem of obtaining and increasingly facilitate, but also bringing information overload, this makes people be difficult letter needed for obtaining from vast resources
Breath.In order to allow user is more acurrate to get required information, recommender system is come into being, and recommender system mainly goes through user
History behavior is analyzed, to get the preference of user, is then recommended for user, and the content recommended is personalized
's.By such method, the information for being supplied to user was not only useful but also efficient.Personalized recommendation method is such as personal by analysis
The information of the historical behavior of feature and user etc provides a user may interested service.Personalized recommendation research has become
The hot spot of data mining and field of social network.
Collaborative filtering is most mature, widest personalized recommendation algorithm.Traditional Collaborative Filtering Recommendation Algorithm is based on using
The attribute data at family, historical scores data and social label etc., but most of social network informations for all seldom considering user.Such as
What is preferably to improve a part for recommending quality indispensable using the social relationship information of user.
In real life, people often tend to receive the recommendation of acquaintance or trustee.User is logical to the assessment of project
It often can greatly influence the selection of his friend.Due to the trust between friend, the recommendation in conjunction with trust is more acurrate.Many researchs
Personnel are it has been shown that the social use trusted can more accurately excavate the interest of user and improve the quality of proposed algorithm.Knot
It closes the proposed algorithm trusted and also faces some problems, such as trust data sparsity.
Although recommender system can alleviate " information overload " problem under big data, it is faced with choosing for some sternnesses
War.First, sparsity problem.How data sparsity problem is effectively solved, is the main problem that collaborative filtering faces.
Second, Malicious recommendation problem.Conventional recommendation algorithm is mainly foundation score data to calculate similarity between user, this effective phase
Premise like degree is that score data is true, reliable.But in practical application scene, this premise often hardly results in guarantor
Card.Third recommends the low problem of precision.Hidden factor model based on matrix decomposition is because its algorithm scalability is good and flexibility height etc.
Various features are the main models it is presently recommended that system, are had been widely used.
Summary of the invention
Present invention aims at the social networks network informations and existing data sparsity problem that are directed to user, propose one
The new personalized recommendation method based on the trust degree of association of kind.
In order to achieve the above-mentioned object of the invention, the present invention uses a kind of personalized recommendation method based on trusting relationship, optional
Ground the described method comprises the following steps:
Step 1: using the formula of matrix decomposition, by decompose to user-project history ratings data matrix
To two matrixes, and learn their implicit features matrix;
Step 2: according to the implicit features matrix in step 1, the relationship of user rating and the definition of loss function are obtained;
Step 3: according to the loss function in step 2, loss function is solved using gradient descent method;
Step 4: it includes global trusting value and local trust value that trust value is calculated in the way of trust network digraph;
Step 5: fusion trust value calculates the similarity between user using Pearson correlation coefficient;
Step 6: recommendation quality is evaluated using mean absolute deviation and root-mean-square error;
Optionally, the method that step 2 obtains the definition of the implicit features matrix and loss function of two split-matrixes includes
Following steps:
Step 1: using user-project history ratings data matrix, and assume that matrix implies K feature, matrix decomposition
Formula are as follows:
R=PQT (1)
Wherein P ≈ Rm×K, Q ≈ Rn×K, K is the quantity of the hidden feature of user and project, and m is number of users, and n is project
Quantity, R are m * n matrixes.
Step 2: after obtaining two implicit features matrixes, relationship of the user to the prediction grading of project are as follows:
Wherein PuWithThe transposed vector of the implicit features of the implicit features vector sum project of user is respectively represented,It is
User items prediction grading.
Step 3: by minimizing RMSE come learning characteristic matrix, loss function is defined as:
Optionally, step 3 is divided into following steps using gradient descent method solution loss function:
Step 1: parameter P is calculated separatelyukAnd qkiPartial derivative, calculation formula are as follows:
Step 2: the loss function of each parameter is most fast along positive direction decline, and iteration more new formula is available;
Optionally, step 4 calculates trust value in the way of trust network digraph and is divided into following steps:
Step 1: each user only possesses a global trusting value, global trusting value in being in current trust network
Calculation formula are as follows:
Wherein, td (u) represents the in-degree of user u in trust network, and min (td (w)) is represented in trust network figure, institute
There is a smallest in-degree in user node, (td (w) represents in trust network maximum in-degree in all user nodes to max.
Step 2: local trust value, formula are calculated using the calculation method of belief propagation are as follows:
Wherein, TukIndicate degree of belief of the user u to user k, TkvUser k is indicated to the degree of belief of user v, N (u) is to use
The neighbours of family u collect, tuvIt is the indirect trust values for the user u and user v being calculated by Trust transitivity.
Step 3: after calculating global trusting value and local trust value, using the method for weighted sum, obtain user u with
The final trust value of user v:
tuv=(1- β) Gu+βLuv (8)
Optionally, Pearson correlation coefficient described in step 5 are as follows:
The beneficial effects of the present invention are: the fast development of Internet of Things (IoT) and e-commerce brings for people's lives
It is many convenient.Internet of Things application program generates a large amount of services and user data.It is necessary to design a kind of suitable Internet of Things service use
The personalized recommendation technology at family improves user experience.The present invention is by establishing effective trust metrics model, by the social activity of user
Information integration proposes a kind of recommended method based on trusting relationship that associate(d) matrix decomposes into proposed algorithm.The present invention is full
Under conditions of foot is recommended, in conjunction with users to trust relationship and user's similitude, so that recommendation effect greatly improves.
Detailed description of the invention
Scheme the flow chart first is that a kind of personalized recommendation method based on trusting relationship of the present invention.
Specific embodiment
The embodiment of the present invention is described in detail with reference to the accompanying drawing.
Embodiment one
One group of new data set is given, next it is grasped according to corresponding to step in the recommended method based on degree of belief
Make, comprising:
Step 1: user-project history ratings data matrix is decomposed;
User-project grading matrix R is obtained to project I history ratings data by user U, and assumes that matrix implies K
Then R is decomposed into two matrixes P, Q by feature
R=PQT (1)
Wherein P ≈ Rm×K, Q ≈ Rn×K, K is the quantity of the hidden feature of user and project, and m is number of users, and n is project
Quantity, R are m * n matrixes.The i row of P indicates the degree of correlation of user i and k feature, the j row expression project j of Q and k feature
The degree of correlation.
Step 2: the relationship and loss function of user rating are obtained;
After obtaining two implicit features matrixes, user i predicts that the grading on project j is equal to vector Pik×Qkj.We set
It setsIt is prediction grading of the user u on object item i, andIt is really to grade.WithBetween relationship be
Wherein PuWithThe transposed vector of the implicit features of the implicit features vector sum project of user is respectively represented,It is
User items prediction grading.
Learn P, Q eigenmatrix by minimizing RMSE.Loss function is defined as:
Fundamental matrix decomposition algorithm directly optimizes loss function, but may cause over-fitting.Therefore in loss function
Middle addition regularizer is to avoid overfitting.After adding regularizer, loss function is newly defined as
Wherein | | p | |2With | | Q | |2It is the quadratic sum of all elements.
Step 3: loss function is solved using gradient descent method;
Calculate separately parameter PukAnd qkiPartial derivative
Then, the loss function of each parameter declines most fast along direction, and iteration more new formula is available
Wherein α is the constant for indicating learning rate, and the value of α needs to obtain by repeating experiment.Learning rate α influence matrix point
Solve convergence rate and the training time of model.In fluctuation range, α value is bigger, and the convergence rate of model is faster, required training
Time is shorter.Finally, updated by the way that specific the number of iterations or one small positive threshold value of setting is arranged as end iteration
Stop condition.After obtaining final matrix P and matrix Q, PQTThe new grading matrix that all elements are all filled can be obtained,
New grading is the prediction grading that object uses on object item.
Assuming that there are object function f (x) and the function has local minimum.Declined by gradient and obtains Local Minimum
The step of value, is as follows:
A) two small positive number α and ε are set, α is iterative step, and ε is off condition.
B) gradient of current point is calculated
C) iterative formula is usedCarry out more new variables x.
If d) target function value variation is less than ε, otherwise iteration stopping goes to step b.
Usual f (x) can converge to local minimum.α is factorial effect f (x) convergence rate.If α is too small, covering speed
Degree can be very slow;If α is too big, search will be expanded to gradually near smallest point, rather than converge to smallest point.
Step 4: trust value is calculated in the way of trust network digraph;
Global trusting value is popularity or status in entire trust network, that is, each user is being in current trust
In network, only possess a global trusting value, the calculating of global trusting value such as formula
Wherein, td (u) represents the in-degree of user u in trust network, and value represents the number of users for trusting user u,
The quantity directly represents the global reputation of user u, and min (td (w)) is represented in trust network figure, in all user nodes
The smallest in-degree, it can be understood as the least user of trusting relationship in trusting relationship figure, (td (w) represents trust network to max
In, maximum in-degree in all user nodes, it can be understood as the target user most trusted by user in trusting relationship figure, it is global
Degree of belief GuValue be in the section of [0,1].
The trust value between user u and user v is calculated using the calculation method of belief propagation for local trust value.
Wherein, TukIndicate degree of belief of the user u to user k, TkvUser k is indicated to the degree of belief of user v, N (u) is to use
The neighbours of family u collect, tuvIt is the indirect trust values for the user u and user v being calculated by Trust transitivity.
But have ignored the relationship of trust value Yu belief propagation path.Therefore the calculation method of improved local trust are as follows:
Wherein, d is represented in trust network, and user u connect the length of shortest path with user v, that is, passes through letter
Appoint the shortest distance propagated and reach user v.
After calculating global trusting value and local trust value, using the method for weighted sum, user u and user v are obtained
Final trust value
tuv=(1- β) Gu+βLuv (8)
Wherein, GuIndicate the local trust value of user u and user v, LuvIndicate the global trusting value of user u.
Step 5: the recommended method of fusion trust value and similarity;
Using Pearson correlation coefficient formula, the scoring similitude between user is calculated, formula is as follows
Wherein, ruiIndicate score value of the user u to project i, rviIndicate user v commenting to the score value project i of project i
Score value, IuvIndicate the common scoring item collection of user u and user v, IuIndicate the scoring item collection of user u, IvIndicate user v's
Scoring item collection,Indicate the average value of all scoring items of user u,Indicate the average value of all scoring items of user v
Weigh trusting relationship and influence of the similarity relationships to recommendation results of scoring.It obtains new between user u and user v
Similarity ωuv
In view of the transmitting of trust value can reduce with the growth in path, therefore, the introducing impact factor a in formula,
Its calculation method is such as
Wherein, tuvrIndicate that the trust value of user u and user v on r paths, road (u, v) indicate in trust network,
User u is connected to the set in all paths user v, | road (u, v) | indicate shortest path length between user u to user v
The calculation formula of impact factor b
Wherein, n is that user u and user v give a mark the quantity of project jointly, n1It is at least grading to project in system
Amount, n2It is most marking quantity in system to project
Step 6: recommendation quality is evaluated using mean absolute deviation and root-mean-square error;
Reflect the inclined of prediction result and actual conditions by calculating the mean absolute deviation between predicted value and true value
Difference, MAE calculated and RMSE value are smaller, then it represents that the error between the corresponding predicted value of model and true score value is got over
Small, the precision that represent recommendation results is higher.
The calculation formula of root mean square error RMSE
The calculation formula of mean absolute error MAE is as follows
Wherein T indicates test set scoring record number, RijIndicate true scoring of the user i to project jIndicate user i to item
The prediction score value of mesh j
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
The equivalent structure or processes is waited to convert that bright specification and accompanying drawing content are done, are applied directly or indirectly in correlative technology field,
It similarly include in scope of patent protection of the invention.
Claims (6)
1. a kind of personalized recommendation method based on trusting relationship, which is characterized in that the described method comprises the following steps:
Step 1: using the formula of matrix decomposition, by being decomposed to obtain two to user-project history ratings data matrix
A matrix, and learn their implicit features matrix;
Step 2: according to the implicit features matrix in step 1, the relationship of user rating and the definition of loss function are obtained;
Step 3: according to the loss function in step 2, loss function is solved using gradient descent method;
Step 4: it includes global trusting value and local trust value that trust value is calculated in the way of trust network digraph;
Step 5: fusion trust value calculates the similarity between user using Pearson correlation coefficient;
Step 6: recommendation quality is evaluated using mean absolute deviation and root-mean-square error.
2. a kind of personalized recommendation method based on trusting relationship according to claim 1, which is characterized in that in step 2
The definition of the implicit features matrix and loss function for obtaining two split-matrixes the following steps are included:
Step 1: using user-project history ratings data matrix, and assume that matrix implies K feature, the public affairs of matrix decomposition
Formula are as follows:
R=PQT (1)
Wherein P ≈ Rm×K, Q ≈ Rn×K, K is the quantity of the hidden feature of user and project, and m is number of users, and n is the number of entry, R
It is m * n matrix;
Step 2: after obtaining two implicit features matrixes, relationship of the user to the prediction grading of project are as follows:
Wherein PuWithThe transposed vector of the implicit features of the implicit features vector sum project of user is respectively represented,It is user
Project forecast grading;
Step 3: by minimizing RMSE come learning characteristic matrix, loss function is defined as:
3. a kind of personalized recommendation method based on trusting relationship according to claim 1, which is characterized in that step 3 institute
State using gradient descent method solve loss function the following steps are included:
Step 1: parameter P is calculated separatelyukAnd qkiPartial derivative, calculation formula are as follows:
Step 2: the loss function of each parameter is most fast along positive direction decline, and iteration more new formula is available;
4. a kind of personalized recommendation method based on trusting relationship according to claim 1, which is characterized in that in step 4
It is described trust value point is calculated in the way of trust network digraph the following steps are included:
Step 1: each user only possesses a global trusting value, the meter of global trusting value in being in current trust network
Calculate formula are as follows:
Wherein, td (u) represents the in-degree of user u in trust network, and min (td (w)) is represented in trust network figure, and institute is useful
The smallest in-degree in the node of family, (td (w) represents in trust network maximum in-degree in all user nodes to max;
Step 2: local trust value, formula are calculated using the calculation method of belief propagation are as follows:
Wherein, TukIndicate degree of belief of the user u to user k, TkvUser k is indicated to the degree of belief of user v, N (u) is user u
Neighbours' collection, tuvIt is the indirect trust values for the user u and user v being calculated by Trust transitivity;
Step 3: after calculating global trusting value and local trust value, using the method for weighted sum, user u and user v are obtained
Final trust value:
tuv=(1- β) Gu+βLuv (8)
5. a kind of personalized recommendation method based on trusting relationship according to claim 1, which is characterized in that in step 5
The similarity between user is calculated using Pearson correlation coefficient:
6. a kind of correlation rule merging method based on endless form according to claim 1, which is characterized in that step 5
It is middle that recommendation quality is evaluated using mean absolute deviation and root-mean-square error:
Pass through the deviation for calculating the mean absolute deviation between predicted value and true value to reflect prediction result and actual conditions, institute
The MAE of calculating and RMSE value are smaller, then it represents that the error between the corresponding predicted value of model and true score value is with regard to smaller, generation
Table recommendation results precision it is higher;
The calculation formula of root mean square error RMSE:
The calculation formula of mean absolute error MAE:
Wherein T indicates test set scoring record number, RijIndicate true scoring of the user i to project jIndicate user i to project j
Prediction score value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910633177.8A CN110334286A (en) | 2019-07-10 | 2019-07-10 | A kind of personalized recommendation method based on trusting relationship |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910633177.8A CN110334286A (en) | 2019-07-10 | 2019-07-10 | A kind of personalized recommendation method based on trusting relationship |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110334286A true CN110334286A (en) | 2019-10-15 |
Family
ID=68146763
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910633177.8A Pending CN110334286A (en) | 2019-07-10 | 2019-07-10 | A kind of personalized recommendation method based on trusting relationship |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110334286A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111198991A (en) * | 2020-01-03 | 2020-05-26 | 长沙理工大学 | Collaborative filtering recommendation method based on trust level and expert user |
CN111460318A (en) * | 2020-03-31 | 2020-07-28 | 中南大学 | Collaborative filtering recommendation method based on explicit and implicit trusts |
CN111814059A (en) * | 2020-08-24 | 2020-10-23 | 安徽大学 | Matrix decomposition recommendation method and system based on network representation learning and community structure |
CN111881342A (en) * | 2020-06-23 | 2020-11-03 | 北京工业大学 | Recommendation method based on graph twin network |
CN113486259A (en) * | 2021-07-06 | 2021-10-08 | 天津大学 | Recommendation method based on bidirectional sparse trust |
CN117333203A (en) * | 2023-12-01 | 2024-01-02 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109101667A (en) * | 2018-09-29 | 2018-12-28 | 新乡学院 | A kind of personalized recommendation method based on explicit trust and implicit trust |
CN109408734A (en) * | 2018-09-28 | 2019-03-01 | 嘉兴学院 | A kind of collaborative filtering recommending method of fuse information Entropy conformability degree and dynamic trust |
-
2019
- 2019-07-10 CN CN201910633177.8A patent/CN110334286A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109408734A (en) * | 2018-09-28 | 2019-03-01 | 嘉兴学院 | A kind of collaborative filtering recommending method of fuse information Entropy conformability degree and dynamic trust |
CN109101667A (en) * | 2018-09-29 | 2018-12-28 | 新乡学院 | A kind of personalized recommendation method based on explicit trust and implicit trust |
Non-Patent Citations (2)
Title |
---|
田保军 等: "融合信任和基于概率矩阵分解的推荐算法", 《计算机应用》 * |
苏佳会: "基于信任关系的个性化推荐方法研究", 《硕士电子期刊》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111198991A (en) * | 2020-01-03 | 2020-05-26 | 长沙理工大学 | Collaborative filtering recommendation method based on trust level and expert user |
CN111460318A (en) * | 2020-03-31 | 2020-07-28 | 中南大学 | Collaborative filtering recommendation method based on explicit and implicit trusts |
CN111460318B (en) * | 2020-03-31 | 2022-09-30 | 中南大学 | Collaborative filtering recommendation method based on explicit and implicit trusts |
CN111881342A (en) * | 2020-06-23 | 2020-11-03 | 北京工业大学 | Recommendation method based on graph twin network |
CN111814059A (en) * | 2020-08-24 | 2020-10-23 | 安徽大学 | Matrix decomposition recommendation method and system based on network representation learning and community structure |
CN113486259A (en) * | 2021-07-06 | 2021-10-08 | 天津大学 | Recommendation method based on bidirectional sparse trust |
CN117333203A (en) * | 2023-12-01 | 2024-01-02 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
CN117333203B (en) * | 2023-12-01 | 2024-04-16 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110334286A (en) | A kind of personalized recommendation method based on trusting relationship | |
CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
CN109190030B (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
CN107784124B (en) | LBSN (location based service) hyper-network link prediction method based on space-time relationship | |
CN106649657A (en) | Recommended system and method with facing social network for context awareness based on tensor decomposition | |
CN109948066B (en) | Interest point recommendation method based on heterogeneous information network | |
CN105760649B (en) | A kind of credible measure towards big data | |
CN107038184B (en) | A kind of news recommended method based on layering latent variable model | |
CN105787100A (en) | User session recommendation method based on deep neural network | |
CN109062962A (en) | A kind of gating cycle neural network point of interest recommended method merging Weather information | |
CN110990718A (en) | Social network model building module of company image improving system | |
CN104794367A (en) | Hospitalizing resource scoring and recommending method based on latent factor model | |
CN107291840A (en) | A kind of user property forecast model construction method and device | |
CN106354783A (en) | Social recommendation method based on trust relationship implicit similarity | |
CN110674857A (en) | Advertisement click classification method based on multi-scale stacking network | |
CN107003834B (en) | Pedestrian detection device and method | |
CN105761154A (en) | Socialized recommendation method and device | |
CN111475739B (en) | Heterogeneous social network user anchor link identification method based on meta-path | |
CN113868537B (en) | Recommendation method based on multi-behavior session graph fusion | |
CN107016566A (en) | User model construction method based on body | |
CN115618108A (en) | Cigarette brand recommendation method based on knowledge graph in new retail mode | |
CN104484365B (en) | In a kind of multi-source heterogeneous online community network between network principal social relationships Forecasting Methodology and system | |
CN105279699A (en) | Recommendation method combining multi-class untrust relation based on collaborative filtering | |
CN106777123A (en) | A kind of group based on two-way tensor resolution model recommends method | |
CN116304289A (en) | Information chain recommendation method and device for supply chain based on graphic neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191015 |