CN109213951A - A kind of proposed algorithm calculated based on trust with matrix decomposition - Google Patents
A kind of proposed algorithm calculated based on trust with matrix decomposition Download PDFInfo
- Publication number
- CN109213951A CN109213951A CN201810614359.6A CN201810614359A CN109213951A CN 109213951 A CN109213951 A CN 109213951A CN 201810614359 A CN201810614359 A CN 201810614359A CN 109213951 A CN109213951 A CN 109213951A
- Authority
- CN
- China
- Prior art keywords
- trust
- user
- matrix decomposition
- social
- 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
- 239000011159 matrix material Substances 0.000 title claims abstract description 50
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 claims abstract description 33
- 238000010606 normalization Methods 0.000 claims abstract description 5
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000035945 sensitivity Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 8
- 238000002474 experimental method Methods 0.000 description 7
- 238000012549 training Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 230000006399 behavior Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000004132 cross linking Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000011273 social behavior Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Computational Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Human Resources & Organizations (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of based on the proposed algorithm for trusting calculating and matrix decomposition.Matrix decomposition recommended models are now widely used personalized recommendation technologies, however generally existing Deta sparseness, the problems such as cold start-up and anti-attack ability are poor.The present invention is using the social homogeney principle in social networks as theoretical foundation, interest preference according to user is the heuristic thought that result is influenced by its own speciality and friend, proposes the matrix decomposition proposed algorithm that a kind of simple and effective trust metrics method and a trust are reinforced.Trust calculating process and comprehensively considers the factors such as global trusting, local trust, Trust transitivity, multichannel trust combination.During the user modeling based on matrix decomposition, the latent factor vector of active user is extended using neighbor user is trusted, normalisation constraint is carried out to parametric solution objective function using the trusting relationship between user simultaneously, to greatly alleviate Deta sparseness and cold start-up problem, and there is good anti-attack ability.
Description
Technical field:
The invention belongs to Artificial smart fields, and in particular to trust extended method and letter based on user's credit worthiness
Appoint the matrix decomposition Collaborative Recommendation model reinforced, is opened for solving data sparsity problem in recommender system and the cold of new user
Dynamic problem.
Background technique:
In the epoch of this information explosion, recommender system is as a kind of effective Information Filtering Technology in various e-commerce
It is widely used in platform and social networks, such as Amazon, eBay, Netflix etc..Since Netflix Prize hundred in 2009
Since ten thousand U.S. dollar contests, the Collaborative Recommendation technology based on matrix decomposition has obtained the wide of industry with its outstanding representation in contest
General concern.The core concept of matrix decomposition thinks that the interest of user is only influenced by a few deciding factor, therefore passes through
Sparse and higher-dimension user-project rating matrix is decomposed into the matrix of two low-dimensionals, i.e. user characteristics matrix and item characteristic
Then matrix can predict scoring of the user to project with the low-dimensional matrix of reconstruct, can reduce computation complexity in this way
While improve recommend performance.
However in actual environment, user-project rating matrix is often very sparse, and the extreme sparsity of rating matrix makes
The preference pattern of user can not accurately be obtained by obtaining matrix disassembling method, to influence the precision of score in predicting and recommendation.In consideration of it,
Researchers propose various improvement strategies one after another, such as matrix fill-in, the matrix point of the matrix decomposition with biasing, context-aware
Solution, confederate matrix decomposition etc..However above method does not solve the data sparsity problem in Collaborative Recommendation fundamentally, because
There is no change for the data source used for them.
According to " comforming " property in social influence theory, on the one hand behavior of the user in social environment is by user itself
Interest preference determine, on the other hand then will receive the influence of surrounding population.In recent years, with the rise of social networks and fluffy
The exhibition of breaking out, a large amount of user data is produced in social networks, and trusting relationship is exactly one of them.Trusting relationship between user and
Interbehavior between user-project is two relatively independent data sources, improves recommender system using social trusting relationship
Performance is mainspring of the invention.
There is also cold start-up problems for traditional collaborative filtering, have seldom those even without history scoring number
According to new user, this method is often difficult to it and makes correct recommendation.The introducing that social activity is trusted can alleviate this problem, root
According to social selection and the homogeney principle in social influence theory, often there is similar hobby between friend, so can
To utilize the preference pattern of the trusting relationship auxiliary building user between user.
Anti-attack ability is also that recommender system needs one of the problem of paying close attention to, and the Collaborative Recommendation technology based on history scoring is aobvious
So there are security risks.Malicious user early period becomes its similar neighborhood by imitating the behavior of target user, and the later period passes through implementation
Malicious act damages target user.Collaborative Recommendation based on trusting relationship can be to avoid this problem, as long as malice
User is not belonging to the trusted circle of friends of target user, they are impossible to generate any shadow to the score in predicting of target user
It rings.
Based on the above reasons, I, which invents, proposes a kind of recommended method trusted and reinforced, using it is low when efficient matrix decomposition
Model expands recommended models using homogeney principle is trusted using trusting relationship as additional data source as main frame
Exhibition and constraint effectively solve the problems, such as data sparsity problem, cold start-up, improve the performance and attack resistance energy of recommender system comprehensively
Power.
Summary of the invention:
Firstly, propose a kind of trust calculation method based on user's overall situation credit worthiness, to the social relationships of extending user,
Alleviate the sparsity problem of trust data in social recommendation；Then, it proposes a kind of matrix decomposition model trusted and reinforced, uses
Top-K trusts user and is extended to the feature vector of active user, solve the problems, such as score data sparsity and new user
Cold start-up problem；Finally, further being standardized using the trusting relationship between user to the objective optimization function of matrix decomposition
Constraint improves the accuracy of score in predicting and recommendation.
Specific steps of the invention are as follows:
1. the trust of credit worthiness sensitivity calculates
It is well known that the data sparsity problem of trusting relationship being introduced in solution Collaborative Recommendation system, however
In actual application environment, trusting relationship itself is also very sparse, the trust of many its maintenances of the few user of quantity of scoring
Relationship is often also seldom, so the performance of traditional social recommendation algorithm is unsatisfactory.In order to alleviate trust data in social recommendation
Sparsity problem, the present invention proposes a kind of trust calculation method of credit worthiness sensitivity, carrys out the indirect trust between reasoning user
Spend Sij, calculation formula is as follows:
Wherein, Trustee (ui) indicate user uiDirect trust user collection, rep (uk) indicate user uiGlobal prestige
Degree.In social networks, there are many kinds of the measures of user's credit worthiness, such as based on the centrality degree index of social network of Crosslinking Structural
Degree, Closeness, Betweenness and Eigenvector and PageRank, the graph theorys calculation method such as Hits.
Simply but meaning is abundant for the above trust extended method principle, and calculating process considers the part letter between user comprehensively
The effect of the relationship of appointing and the global credit worthiness of user in trust transfer, and the summation operation in molecule item embodies
The composition principle that multichannel is trusted, the degree of belief being calculated are in [0,1] section.
2. the social recommendation based on matrix decomposition
In view of outstanding representation of the PMF model in Collaborative Recommendation, the present invention continues using main frame of the PMF as score in predicting
Frame, while based on social homogeney principle, make full use of the social trusting relationship between user to be extended matrix decomposition model
And constraint.
By social homogeney principle it is found that in social networks the behavior of people be not only to be determined by the interest preference of its own
Determine, and will receive the influence of its surrounding population, especially its friend trusted.Therefore, the present invention is utilizing matrix decomposition
When model models user characteristics, using social trusting relationship as auxiliary information, to the potential feature vector of active user
It is extended, user's scoring conditional probability distribution after extension is as follows:
Wherein, u indicates scoring mean value, biAnd bjRespectively indicate user uiWith project vjBiasing, Tk(i) user u is indicatedi's
Top-K trusts user's collection, xlIndicate uiThe 1st trust user in uiContribution weight during feature modeling, the parameter is by instructing
Practice procedural learning to obtain.The figure description of Trust-MF modeling process is as shown in Figure 1.
The present invention and existing method have essential distinction, have method and trust social activity as additional information to score in predicting
Calculating process be adjusted, not modifying to user vector itself, and the method for the present invention is to user characteristics vector
Extension, by original subscriber's vector UiBecome Ui', whereinSo, just hidden during user modeling
The process of Trust transitivity is contained.
According to probability distribution principle, the logarithmic form of score in predicting Posterior probability distribution is as follows:
Wherein, N (i) is and user uiUser's collection with trusting relationship, C is the unrelated constant term of parameter.Maximize with
Upper posterior probability is equivalent to seek the minimum of following objective function:
Wherein,Here for simplicity, if λU=λV=λ.Objective function last
Be the normalisation constraint factor based on trusting relationship, it is therefore an objective to keep the feature vector trusted between user similar as far as possible: when with
Family uiAnd ulBetween degree of belief it is relatively high when, their feature vector UiAnd UlShould be more similar, i.e., difference should very little；
When the degree of belief between them is relatively low, feature vector UiAnd UlBetween can have larger gap.
The above objective function be it is non-convex, be difficult to acquire optimal solution, ask it used here as stochastic gradient descent method (SGD)
Locally optimal solution.Remember current score in predicting errorFor eij, hyper parameter is iterated by traversing all known scorings
It updates.The training sample R given for oneij, each parameter update rule it is as follows:
bi=bi+γ(eij-λbi) formula (7)
bj=bj+γ(eij-λbj) formula (8)
Wherein, T (i) is user uiTrust user collection, B (i) be trust user uiUser collection, γ is learning rate, λSWith
λ is normalization factors, and k is the number for trusting user.
Experimental result of the invention:
Experiment test selection Epinions general data collection, the data set is by 49,289 users to 139,738 projects
664,823 scorings composition, while include user statement 478,183 trusting relationships.Experiment is compared to be tested using two kinds
Index: mean absolute error (MAE) and root-mean-square error (RMSE).
In order to illustrate Trust-MF algorithm proposed by the present invention in the improvement for recommending aspect of performance, in identical experiment
Under environment, the present invention is compared with a series of algorithms of social recommendation currently popular with the proposed algorithm based on matrix decomposition
Compared with, including:
TrustCF: in order to which validation matrix decomposes the superiority of recommended method, we, which design, realizes one based on user
The Collaborative Recommendation algorithm of trusting relationship is recommended wherein using based on the KNN method for trusting user；
BPMF: the matrix decomposition algorithm with biasing, it increases the biasing of user and project on the basis of PMF algorithm
Regulatory factor models user.The algorithm is only recommended using user-project rating matrix as data source；
SoRec: the social recommendation algorithm decomposed based on confederate matrix, the algorithm will be scored square using matrix decomposition technology
Battle array and trust matrix are decomposed simultaneously to model to user, and it is empty that user characteristics vector is had shared during matrix decomposition
Between；
TrustPMF: the algorithm decomposes rating matrix and trust matrix using PMF technology, with SoRec algorithm
Unlike, there is no shared user vector spaces for the algorithm, but using the trusting relationship between user as the pact of objective function
Shu Xiang, with this come the connection established between scoring behavior and trusting relationship；
TrustSVD: the algorithm is extended svd algorithm using trust information, and it is auxiliary as additional information to trust user
Carry out user modeling is helped, while regularization term of enhancing trust constrains objective function.This method and the present invention have it is similar it
Place, but there is essential distinction in the calculation method of users to trust degree, and the two is to the extension side of the potential feature vector of user
Formula is also entirely different.
Performance between the different proposed algorithms of experiment 1. compares
Raw data set is randomly divided into 80% training set and 20% test set, under identical experimental situation, base
In the latent factor quantity of different scales, experiment evaluation and test is carried out to each proposed algorithm using cross validation method.Experimental result is such as
Shown in table 1, last line indicates the improvement degree of the relatively existing method top performance of the present invention in table.
From table 1 it follows that under the latent factor quantity of different scales, the score in predicting of Trust-MF of the present invention
It can be superior to other comparison algorithms.Generally speaking, in conjunction with the matrix decomposition proposed algorithm of trusting relationship (SoRec, TrustPMF,
TrustSVD performance) is better than the bPMF algorithm performance for only depending on score data, this illustrates that social networks have user behavior
There is certain influence, also illustrates that trusting relationship can be used as the accuracy that auxiliary information improves user modeling.In addition, bPMF algorithm
Performance is recommended to be better than TrustCF algorithm, this illustrates that Model-based recommended method ratio Memory-based recommended method performance is excellent
More, also the Deta sparseness of the original trust network of side light has seriously affected the performance recommended.In addition, because TrustCF
Algorithm only depends on trusting relationship, not use matrix disassembling method, so the change of latent factor quantity to its performance not
It has any impact.From can also be seen that the quantity of latent factor in matrix decomposition is bigger in table, to the accurate of user modeling
Property it is higher, recommend performance also better, but with the increase of latent factor quantity, computation complexity can also be increase accordingly.
Performance between 1 algorithms of different of table compares
Recommendation performance under the different training set scales of experiment 2. compares
In order to further can be carried out comprehensive comparison to the present invention and comparison the recommendatory of algorithm, we are respectively different
Recommendatory test can be carried out to its under training set scale.It specifically, is exactly by raw data set according to different ratio cut partitions
For training set and test set, to simulate the Sparse degree of different scales, in matrix decomposition recommended models by it is potential because
Subnumber is fixed as 20, and experimental result is as shown in table 2.
From Table 2, it can be seen that being continuously increased with training set scale, the recommendation performance of all recommended methods is all
It improves.Data than it is sparse in the case where, trust reinforcement proposed algorithm (TrustCF, SoRec, TrustPMF,
TrustSVD it) has a clear superiority than only relying upon the bPMF algorithm of score information.Wherein, TrustCF algorithm is due to based on letter
Neighbours are appointed to recommend, so being influenced less by score data scale.Trust-MF of the present invention is no matter under which kind of data scale
All there is significant advantage, performance is recommended to be substantially better than other comparison algorithms, especially in the case where data are extremely sparse, property
Energy advantage becomes apparent.To find out its cause, it is the validity of user model extension and objective function optimization in Trust-MF algorithm,
And these are all attributed to the reasonability and validity of degree of belief calculation method RepTrust proposed by the present invention.Accurate degree of belief
Measurement is so that the accuracy of user modeling improves, so as to improve recommendation precision.The excavation of implicit trust relationship extends existing
Trust network greatly alleviates the sparsity problem of data.
Recommendation performance under the different training set scales of table 2 compares
The recommendation performance for testing 3. couples of different type users compares
A major challenge of traditional Collaborative Recommendation system is exactly to be difficult to the user with few scoring history to make effectively
Recommend.In order to verify the present invention and comparison algorithm for the recommendation ability of such user, scored according to history several to all users
Classify, then observe performance of the algorithms of different in different type user group, experimental result is as shown in Figure 2.Wherein
All users are divided into 5 groups: " 1-10 ", " 11-20 ", " 21-30 ", " 31-40 ", " > 40 ", and the numerical value in grouping indicates that user comments
The range of dosis refracta.
From Fig. 2 (a) and 2 (b) as can be seen that the performance ranking of each algorithm is kept not substantially in different type user
Become.Generally speaking, trust the proposed algorithm TrustCF, SoRec, TrustPMF and TrustSVD of reinforcement and be based only upon scoring number
According to algorithm bPMF compare, by user score number influenced it is smaller.Compared with comparing algorithm, Trust-MF of the present invention is to different type
The recommendation ability of user all has a clear superiority, and especially for the few user of scoring quantity, here it is usually said " cold
Starting " user.And the recommendation performance of TrustCF method kept stable in all types of users, it is equally because of it
Only depend on trust information recommended it is unrelated with the scoring quantity of user.The experimental result not only demonstrates of the invention steady
It is qualitative, and illustrate that the cold start-up problem in Collaborative Recommendation can be effectively relieved in the present invention.
It summarizes:
The society that the friend that the present invention is trusted with user in social networks can have an impact the Social behaviors of user is same
For matter principle as theoretical foundation, the interest preference according to user is the heuristic think of that result is influenced by itself speciality and friend
Think, propose a kind of completely new social recommendation algorithm based on matrix decomposition, user's mould in proposed algorithm is decomposed to classical matrix
Type is extended, while carrying out normalisation constraint to parametric solution objective function, to greatly alleviate the number in Collaborative Recommendation
According to the cold start-up problem of sparsity problem and new user, and there is good anti-attack ability.
Further it is proposed that users to trust degree calculation method it is simple and effective, the overall situation is considered in calculating process comprehensively
Several deciding factors such as prestige, local trust, Trust transitivity and multichannel trust.Comparative experiments based on Epinions data set
It is obvious that result of study shows that the relatively existing social recommendation algorithm of the recommendation performance of the method for the present invention and matrix decomposition recommendation method have
Advantage, especially in the case where Sparse and cold start-up.
In some social networks, user can not only safeguard that it trusts the list of friend, but also can construct it not
The user list of trust.The present invention is in the calculating process of users to trust degree, although not referring to the distrust relationship between user,
As long as be the definition of wherein user's credit worthiness is revised as trust the number of users of the user and distrust the user number of users it
Difference, the present invention can be suitable for the social recommendation system with distrust relationship, therefore the present invention has general applicability.
Detailed description of the invention:
Fig. 1 is to trust the matrix decomposition recommended models figure reinforced
Fig. 2 is that the recommendation performance of different type user compares figure.
Claims (3)
1. a kind of based on the collaborative recommendation method for trusting calculating and matrix decomposition, it is characterised in that: the letter including credit worthiness sensitivity
Appoint computing module and trust the matrix decomposition module reinforced, in which:
The trust computing module of the credit worthiness sensitivity for the discovery of implicit trust relationship, and then extends trust network, solves
The sparsity problem of trust data；
It is described to trust the matrix decomposition module reinforced, for being based on social homogeney principle, using trusting relationship to traditional square
Battle array decomposes recommended method and is extended and constrains, to improve recommendation performance.
2. the trust calculation method of credit worthiness sensitivity according to claim 1, it is characterized in that: between considering user comprehensively
The effect of local trust relationship and the global credit worthiness of user in trust transfer, and fully demonstrated multichannel trust
Composition principle.
3. according to claim 1 trust the matrix disassembling method reinforced, it is characterized in that: making full use of social trusting relationship
Major tuneup is carried out to matrix decomposition recommended models, specifically includes: user's latent factor vector is extended and is excellent to target
Change function carry out normalisation constraint, thus solve traditional Collaborative Recommendation data sparsity problem and new user cold start-up ask
Topic, and enhance the anti-attack ability of system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810614359.6A CN109213951A (en) | 2018-06-07 | 2018-06-07 | A kind of proposed algorithm calculated based on trust with matrix decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810614359.6A CN109213951A (en) | 2018-06-07 | 2018-06-07 | A kind of proposed algorithm calculated based on trust with matrix decomposition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109213951A true CN109213951A (en) | 2019-01-15 |
Family
ID=64991843
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810614359.6A Pending CN109213951A (en) | 2018-06-07 | 2018-06-07 | A kind of proposed algorithm calculated based on trust with matrix decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109213951A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110119465A (en) * | 2019-05-17 | 2019-08-13 | 哈尔滨工业大学 | Merge the mobile phone application user preferences search method of LFM latent factor and SVD |
| CN110134835A (en) * | 2019-05-15 | 2019-08-16 | 哈尔滨工业大学 | A kind of users to trust modeling method towards online education platform |
| CN110955829A (en) * | 2019-11-19 | 2020-04-03 | 江西财经大学 | Interest point recommendation method and system fusing credibility and measurement factor matrix decomposition |
| CN112395197A (en) * | 2020-11-19 | 2021-02-23 | 中国平安人寿保险股份有限公司 | Data processing method, data processing device and electronic equipment |
| CN112667885A (en) * | 2020-12-04 | 2021-04-16 | 四川长虹电器股份有限公司 | Matrix decomposition collaborative filtering method and system for coupling social trust information |
-
2018
- 2018-06-07 CN CN201810614359.6A patent/CN109213951A/en active Pending
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110134835A (en) * | 2019-05-15 | 2019-08-16 | 哈尔滨工业大学 | A kind of users to trust modeling method towards online education platform |
| CN110134835B (en) * | 2019-05-15 | 2022-10-28 | 哈尔滨工业大学 | User trust modeling method facing online education platform |
| CN110119465A (en) * | 2019-05-17 | 2019-08-13 | 哈尔滨工业大学 | Merge the mobile phone application user preferences search method of LFM latent factor and SVD |
| CN110119465B (en) * | 2019-05-17 | 2023-06-13 | 哈尔滨工业大学 | Mobile phone application user preference retrieval method integrating LFM potential factors and SVD |
| CN110955829A (en) * | 2019-11-19 | 2020-04-03 | 江西财经大学 | Interest point recommendation method and system fusing credibility and measurement factor matrix decomposition |
| CN110955829B (en) * | 2019-11-19 | 2023-05-23 | 江西财经大学 | Point-of-interest recommendation method and system integrating trust and metric factor matrix decomposition |
| CN112395197A (en) * | 2020-11-19 | 2021-02-23 | 中国平安人寿保险股份有限公司 | Data processing method, data processing device and electronic equipment |
| CN112667885A (en) * | 2020-12-04 | 2021-04-16 | 四川长虹电器股份有限公司 | Matrix decomposition collaborative filtering method and system for coupling social trust information |
| CN112667885B (en) * | 2020-12-04 | 2022-08-16 | 四川长虹电器股份有限公司 | Matrix decomposition collaborative filtering method and system for coupling social trust information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| CN109213951A (en) | A kind of proposed algorithm calculated based on trust with matrix decomposition | |
| CN105740430B (en) | A kind of personalized recommendation method of mosaic society's information | |
| Sun et al. | Improved monarch butterfly optimization algorithm based on opposition‐based learning and random local perturbation | |
| Zhao et al. | Community extraction for social networks | |
| CN103971161B (en) | Hybrid recommendation method based on Cauchy distribution quantum-behaved particle swarm optimization | |
| CN108596774A (en) | Socialization information recommendation algorithm based on profound internet startup disk feature and system | |
| Navgaran et al. | Evolutionary based matrix factorization method for collaborative filtering systems | |
| CN110222838B (en) | Document sorting method and device, electronic equipment and storage medium | |
| Sun et al. | Boosting ant colony optimization via solution prediction and machine learning | |
| Zou et al. | Reinforcement learning to diversify top-n recommendation | |
| Zhang et al. | A novel power-driven fractional accumulated grey model and its application in forecasting wind energy consumption of China | |
| CN104239496A (en) | Collaborative filtering method based on integration of fuzzy weight similarity measurement and clustering | |
| Yao et al. | Accurately detecting protein complexes by graph embedding and combining functions with interactions | |
| CN105761154A (en) | Socialized recommendation method and device | |
| CN103699573A (en) | UGC (User Generated Content) tag clustering method and UGC tag clustering device for social platforms | |
| CN107909498B (en) | Recommendation method based on area below maximized receiver operation characteristic curve | |
| Zhong et al. | An Improved Differential Evolution Algorithm Based on Dual‐Strategy | |
| Huang et al. | A global network alignment method using discrete particle swarm optimization | |
| CN106971053A (en) | A kind of recommendation method based on mixing collaborative filtering | |
| Shirmohammady et al. | PPI‐GA: A Novel Clustering Algorithm to Identify Protein Complexes within Protein‐Protein Interaction Networks Using Genetic Algorithm | |
| Fan et al. | An improved quantum clustering algorithm with weighted distance based on PSO and research on the prediction of electrical power demand | |
| CN108021985A (en) | A kind of model parameter training method and device | |
| CN110910953A (en) | Key protein prediction method based on protein-domain heterogeneous network | |
| CN105678430A (en) | Improved user recommendation method based on neighbor project slope one algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190115 |