Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in...
Transcript of Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in...
![Page 1: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/1.jpg)
Social RegularizationSEMINAR RECOMMENDER SYSTEMS - 10.01.17
Maurício Camargo / Guilherme Holdack / Zafar Mahmood
1/10/2017 1
![Page 2: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/2.jpg)
Agenda◦ Introduction to Social Recommendation
◦ Social Recommendation Methods
◦ SoRec: Social Recommendation Using Probabilistic Matrix Factorization
◦ SocialMF: A Matrix Factorization Technique with Trust Propagation for Recommendation in Social Networks
◦ RSR: Recommended Systems With Social Regularization
◦ Conclusions and Comparison
1/10/2017 2
![Page 3: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/3.jpg)
1. Intro to Social Recommendation
• Traditional recommender systems assume that users areindependent and identically distributed (i.i.d. assumption);
• However, online users are inherently connected via various typesof relations such as friendships and trust relations;
• Users in social recommender systems are connected, providingsocial information.
Definition
1/10/2017 3
![Page 4: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/4.jpg)
• Narrow Definition: any recommendation with online socialrelations as an additional input, i.e., augmenting an existingrecommendation engine with additional social signals.
• Broad Definition: recommender systems recommending anyobjects in social media domains such as items (the focus under thenarrow definition), tags , people, and communities.
The narrow definition is used in the context of this presentation.
1. Intro to Social Recommendation
Narrow x Broad Definition
1/10/2017 4
![Page 5: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/5.jpg)
• Connected users are more likely to share similar interests in topics than two randomly selected users;
• In the physical world, we usually ask suggestions from our friends (tend to be similar and also know our tastes);
• Provides an independent source of information about online users (specially useful on Cold Start);
• Exploiting social relations can potentially improve recommendation performance.
1. Intro to Social Recommendation
Reasons to use
1/10/2017 5
![Page 6: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/6.jpg)
• In addition to the rating matrix in traditional recommender systems, there is also a second matrix to map the relations:
1. Intro to Social Recommendation
Representation
1/10/2017 6
![Page 7: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/7.jpg)
•Memory based: for social recommendation, it takes both the rating information and social information to find similar users (ex: TidalTrust, MoleTrust, TrustWalker).
•Model based: uses matrix-factorization methods which also take into account the social relations. A unified framework can be stated as:
2. Social Recommendation Methods
Overview of the Methods
1/10/2017 7
![Page 8: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/8.jpg)
2. Social Recommendation Methods
Model Based Methods• Co-factorization methods: performs a co-factorization in the
useritem matrix and the user-user social relation matrix by sharing the same user preference latent factor (Ex: SoRec and LOCABAL).
• Ensemble methods: a missing rating for a given user is predicted as a linear combination of ratings from the user and the social network (Ex: STE, mTrust).
• Regularization methods: For a given user, regularization methodsforce his preference to be closer to that of users in his socialnetwork. (Ex: SocialMF and Social Regularization).
1/10/2017 8
![Page 9: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/9.jpg)
SoRec: Social Recommendation Using Probabilistic Matrix FactorizationMaurício Camargo
1/10/2017 9
![Page 10: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/10.jpg)
SoRec
MotivationProblems with current recommender systems:
• Ignores the social interactions or connections among users;
• Bad results on users who have made very few ratings or even none at all;
• Some existing approaches fail to handle very large datasets;
In reality, we always turn to friends we trust for movie, music or book recommendations;
1/10/2017 10
![Page 11: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/11.jpg)
SoRec
Current ScenarioCollaborative Filtering
• Memory Based:
- user-based and item-based approaches;
- trust-based recommender systems – also use trust to calculate similarity (does not scale well).
• Model-based:
- clustering model, aspect models and the latent factor model.
- considers users independent and identically distributed
No model-based approach to deal with social relations.
1/10/2017 11
![Page 12: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/12.jpg)
SoRec
Proposed SolutionSoRec (Social Recommendation):
• predict the missing values of the user-item by employing two different data sources.
• factorize the social network graph and user-item matrix simultaneously using UTZ and UTV
U – low-dimensional user latent feature space
Z – factor matrix in the social network graph
V – low-dimensional item latent feature space
1/10/2017 12
![Page 13: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/13.jpg)
SoRec
How it works1 - By analysing both the social relations and the ratings, we get two diferent tables:
1/10/2017 13
![Page 14: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/14.jpg)
SoRec
How it works2 – Both resulting tables can be factorized into its latent features:
U2TZU1
TV
1/10/2017 14
U1 and U2 – low-dimensional user latent feature space
Z – factor matrix in the social network graph
V – low-dimensional item latent feature space
![Page 15: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/15.jpg)
SoRec
How it works3 – The trick is to force both factorizations to share the same U:
UTZ
UTVSame matrix
1/10/2017 15
U – low-dimensional user latent feature space
Z – factor matrix in the social network graph
V – low-dimensional item latent feature space
U will be influenced by the user x itemratings AND its social network.
![Page 16: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/16.jpg)
SoRec
How it worksTo improve the model:
• trust value should decrease if user i trusts lots of users;
• trust value should be increase if user k is trusted by lots of users.
= outdegree of node vi
= indegree of node vk
The original equation becomes:
1/10/2017 16
![Page 17: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/17.jpg)
SoRec
How it worksFor SoRec, the general equation:
- denotes the Frobenius norm
Becomes:
1/10/2017 17
![Page 18: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/18.jpg)
SoRec
How it worksOther notation:
The minimum can be found through gradient descent:
derivative of the logistic function
1/10/2017 18
In order to reduce the model complexity: λU = λV = λZ
![Page 19: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/19.jpg)
SoRec
How it works
1/10/2017 19
Even though user 4 does not rate any items, the approach still can predict reasonable ratings.
![Page 20: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/20.jpg)
SoRec
How it works – pseudocodeInput: The rating information r, the social information c, the number of latent factors
k, and (regularization parameters)
Output: The user preference matrix U and the item characteristic matrix V
1: Initialize U, V and Z randomly (with k factors)
2: while Not convergent do
3: Calculate ∂J ∂U , ∂J ∂V and ∂J ∂Z
4: Update U ← U − γu ∂J ∂U
5: Update V ← V − γv ∂J ∂V
6: Update Z ← Z − γz ∂J ∂Z
7: Evaluate LossFunction
8: end while
1/10/2017 20
![Page 21: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/21.jpg)
SoRec
Complexity Analysis
Computational time of the method is linear with respect to the number of observations in the two sparse matrices. Thus, the approach can scale on large datasets.
1/10/2017 21
Where and are the numbers of nonzero entries in matrices R and C.
Total computational complexity in one iteration is:
![Page 22: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/22.jpg)
SoRec
Experimental AnalysisEpinions was selected as the data source
• well known knowledge sharing site and review site.
• Users submit their opinions on topics such as products, companies, movies, or reviews issued by other users.
• Users can also assign products or reviews integer ratings from 1 to 5.
• Members maintain a “trust” and a “block (distrust)” list
• 40,163 users who have rated at least one of a total of 139,529 different items. The total number of reviews is 664,824
Density = (very sparce)
1/10/2017 22
![Page 23: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/23.jpg)
SoRec
Experimental AnalysisEpinions was select as the data source
1/10/2017 23
![Page 24: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/24.jpg)
SoRec
Experimental AnalysisComparison to other methods
MMF - Maximum Margin Matrix FactorizationPMF - Probabilistic Matrix FactorizationCPMF - Constrained Probabilistic Matrix Factorization
On average, the approach improves the accuracy by 11.01%, 9.98%, and 7.82% relative to MMMF, PMF and CPMF, respectively.
1/10/2017 24
![Page 25: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/25.jpg)
SoRec
Experimental AnalysisImpact of
1/10/2017 25
![Page 26: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/26.jpg)
SoRec
Experimental AnalysisPerformance on Different Users
1/10/2017 26
![Page 27: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/27.jpg)
SoRec
Experimental AnalysisEfficiency Analysis
1/10/2017 27
![Page 28: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/28.jpg)
SoRec
Conclusion and Future WorkConclusion
• Experimental results: the approach outperforms the other state-of-the-art collaborative filtering algorithms.
• Complexity analysis: it is scalable to very large datasets.
• Can also be used to predict connections on social network.
Future Work:
• Investigate whether the distrust information is useful to increase the prediction quality, and how to incorporate it.
• Consider the diffusion process between users.
1/10/2017 28
![Page 29: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/29.jpg)
References
Ma, Hao, et al. "Sorec: social recommendation using probabilistic matrix factorization." Proceedings of the 17th ACM conference on Information and knowledge management. ACM, 2008.
Social Recommendation: A Review Jiliang Tang · Xia Hu · Huan Liuhttps://pdfs.semanticscholar.org/fff4/4f028044dd6ee79b7c9c26a90a23dc8d4438.pdf
MATRIX FACTORIZATION TECHNIQUES FOR RECOMMENDER SYSTEMS https://datajobs.com/data-science-repo/Recommender-Systems-%5BNetflix%5D.pdf
P. Massa and P. Avesani. Trust-aware collaborative filtering for recommender systems. In Proceedings of CoopIS/DOA/ODBASE, pages 492–508, 2004.
1/10/2017 29
![Page 30: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/30.jpg)
![Page 31: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/31.jpg)
![Page 32: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/32.jpg)
![Page 33: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/33.jpg)
![Page 34: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/34.jpg)
![Page 35: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/35.jpg)
![Page 36: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/36.jpg)
![Page 37: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/37.jpg)
![Page 38: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/38.jpg)
![Page 39: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/39.jpg)
![Page 40: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/40.jpg)
![Page 41: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/41.jpg)
![Page 42: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/42.jpg)
![Page 43: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/43.jpg)
![Page 44: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/44.jpg)
![Page 45: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/45.jpg)
![Page 46: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/46.jpg)
![Page 47: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/47.jpg)
![Page 48: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/48.jpg)
![Page 49: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/49.jpg)
![Page 50: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/50.jpg)
![Page 51: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/51.jpg)
![Page 52: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/52.jpg)
![Page 53: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/53.jpg)
![Page 54: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/54.jpg)
![Page 55: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/55.jpg)
![Page 56: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/56.jpg)
![Page 57: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/57.jpg)
![Page 58: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/58.jpg)
![Page 59: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/59.jpg)
![Page 60: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/60.jpg)
![Page 61: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/61.jpg)
![Page 62: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/62.jpg)
![Page 63: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/63.jpg)
![Page 64: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/64.jpg)
![Page 65: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/65.jpg)
![Page 66: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/66.jpg)
![Page 67: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/67.jpg)
![Page 68: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/68.jpg)
![Page 69: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/69.jpg)
![Page 70: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/70.jpg)
![Page 71: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/71.jpg)
![Page 72: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/72.jpg)
![Page 73: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/73.jpg)
![Page 74: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/74.jpg)
![Page 75: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/75.jpg)
![Page 76: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/76.jpg)
![Page 77: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/77.jpg)
![Page 78: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/78.jpg)
![Page 79: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/79.jpg)
![Page 80: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/80.jpg)
![Page 81: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/81.jpg)
![Page 82: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/82.jpg)
![Page 83: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/83.jpg)
![Page 84: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/84.jpg)
![Page 85: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/85.jpg)
g’(x) = e−x /(1 + e−x )² .
![Page 86: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/86.jpg)
Inputs: observed ratings R, users U, items V and trust information TOutput: the latent feature vectors
1: U and V initialization - samples from normal noises with zero mean
2: while not converged do
3: update U:
4: update V:
5: Evaluate LossFunction
![Page 87: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/87.jpg)
![Page 88: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/88.jpg)
![Page 89: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/89.jpg)
![Page 90: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/90.jpg)
![Page 91: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/91.jpg)
![Page 92: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/92.jpg)
![Page 93: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/93.jpg)
![Page 94: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/94.jpg)
![Page 95: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/95.jpg)
![Page 96: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/96.jpg)
![Page 97: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/97.jpg)
![Page 98: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/98.jpg)
![Page 99: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/99.jpg)
![Page 100: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/100.jpg)
![Page 101: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/101.jpg)
![Page 102: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/102.jpg)
![Page 103: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/103.jpg)
![Page 104: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/104.jpg)
Propagation of Trust and Distrust.Guha, R. , Kumar R., Prabhakar, R., Tomkins, A.
https://pdfs.semanticscholar.org/3911/6b28f1a94e7d0aec082fb325ffdeae430012.pdf
Social-aware Matrix Factorization for Recommender Systems, 2013.Weidele, D.
https://kops.uni-konstanz.de/bitstream/handle/123456789/29251/Weidele_0-259317.pdf
A Generative Bayesian Model for Item and User Recommendation in Social Rating Networks with Trust
RelationshipsC. Gianni, Manco G., Ortale R.
http://www.academia.edu/23622275/A_Generative_Bayesian_Model_for_Item_and_User_Recommendation_i
n_Social_Rating_Networks_with_Trust_Relationships
![Page 105: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/105.jpg)
Recommended System with Social Regularization
Hao Ma, Dengyoung Zhou, Chao Liu, Micheal R.Lyu, Irwin King
Microsoft Research & Chinese University of Hong Kong 2011
Zafar Mahmood
1/10/2017 105
![Page 106: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/106.jpg)
Outline
• Motivation and Introduction
• Trust and Social Aware System Difference
• Traditional Systems
• Problem Definition
• Matrix Factorization
• Social Regularization
• Data Sets
• Comparisons and Results
• Pseudo Code
• Conclusion and Future work
• References
1/10/2017 106
![Page 107: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/107.jpg)
Introduction and Motivation
• Widely studied for information retrieval
• For production Recommendation, used in Amazon, Itunes, Netflix etc
• We always ask friends for recommendation in different products
• We Used Trust aware Systems
• Previous methods ignores social relationship in process,
1/10/2017 107
![Page 108: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/108.jpg)
Trust Aware And Social Friends (1)
• Different Approaches
• “Trust aware” doesn’t have to know each other, … SoundCloud ,twitter etc
• Based on the Assumption that user have similar taste
• “Social aware” to interact and connect with their friends in the real life, ... facebook etc
• Need to incorporate social information
1/10/2017 108
![Page 109: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/109.jpg)
Traditional Systems
Collaborative Filtering
• Neighborhood Approaches (User or Items)
• Model Based approaches
1/10/2017 109
![Page 110: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/110.jpg)
Problem Definition
Predict the missing terms of user-item matrix by Incorporate the social network information
• Bidirectional social Connection (User – Item Matrix)
• Unidirectional trust Connection
1/10/2017 110
![Page 111: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/111.jpg)
Low Rank Matrix Factorization
• We have User and Item Matrix, approx rating matrix by multiplying l-rank factors
• Traditionally, we use Single Value Decomposition (SVD) for minimization of R
• Due to sparsity we only need factorize the observed rating in matrix
• So, we use Indicator function for missing value's ----> I = {1,0}
• when user rated the item = 1 , else = 0
1/10/2017
Extremely Sparse …................................................ (1)
….........................................................(2)
…....…....................(3)
111
![Page 112: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/112.jpg)
Now to avoid overfitting, we add normalization
1/10/2017
…..........................(4)
• Now we can use Gradient Approach to Find the minimum
112
![Page 113: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/113.jpg)
Social Regularization
Two models are used for social Regularization
• Average Based Model
• Individual Based Model
1/10/2017 113
![Page 114: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/114.jpg)
Average Based Model
We always ask our friend for recommendation using ( …. 4 ) Matrix Factorization
1/10/2017
….......................(5)
In social Network, Facebook etc
114
![Page 115: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/115.jpg)
• In (…...5) we have given the average taste users friends, which doesn’t seems right, due to diverse taste nature …. changing it by introducing a similarity function
• As similarity is more accurate than our previous approach,
• Now to find the local minima, we just take the derivative
1/10/2017
….....................(6)
115
![Page 116: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/116.jpg)
1/10/2017 116
![Page 117: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/117.jpg)
Individual-based Regularization
• Previously, we used similarity average of friends
• In reality users have diverse taste, so this could cause information loss so, add another regularization term,
• Constraint between user and their friends, individually
Now putting in equation ( ….... 5)
1/10/2017
…...... (iii)
117
![Page 118: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/118.jpg)
Als
• Also deal with 2nd degree friends
• Like U(i) and U(g) are not friends but indirectly minimizing the distance between the feature vectors ….. ( expanding.....(iii) )
• Now for local minima we again use the gradient descent ( …...... 6 )
1/10/2017
…............................... (6)
118
![Page 119: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/119.jpg)
1/10/2017 119
![Page 120: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/120.jpg)
Similarity Function
We have User's rating, for similarity two methods are used.
• Two popular methods raging [0,1] Vector Space Similarity (VSS), ignore the individual rating behavior
• Pearson Correlation Coefficient (PCC) [-1,1], considers individual Rating behavior
1/10/2017
To map in [0,1] we will
do f(x) = (x + 1)/ 2
120
![Page 121: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/121.jpg)
Datasets
two data-sets
• Douban
• Rating and Recommendation about movies, books, music
• Provides information about social friends
• In Movie Group, Users = 129,490, Movies = 58,541,.. total rated cells in matrix = 16,830,839
1/10/2017 121
![Page 122: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/122.jpg)
Epinions
• Visitors read review of other users for and item selection
• Each user Maintain Trust list
• Users = 51,670; items = 83,509, ….. total rating cells in matrix = 631,064
1/10/2017 122
![Page 123: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/123.jpg)
Comparison’s
Comparison with previous three other different methods
• NMF
• For image analysis, also used in Collaborative Filtering
• Probabilistic Matrix Factorization (PMF)
• User-item matrix for recommendation
• RECOMMENDATION WITH SOCIAL TRUST ENSEMBLE (RSTE) *
• Trust aware recommendation user’s rating
In Douban and Epinions , lemda ( 0.001 )
Alpha = 0.001 …. on Douban
Beta = 0.01 …. on Epinions
1/10/2017
Parameters
123
![Page 124: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/124.jpg)
Result By Doubian
• Results given by Different Previous Methods and Our Present Method SR_1 and SR_2,
• SR_1 = Average Based Model
• SR_2 = Individual Based Model
1/10/2017 124
![Page 125: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/125.jpg)
Results By Epinions
1/10/2017 125
![Page 126: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/126.jpg)
Impact Of Parameters
We keep the values of beta low
• Only uses second model
• Douban
1/10/2017 126
![Page 127: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/127.jpg)
Epinions
1/10/2017 127
![Page 128: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/128.jpg)
Impact Of Similarity Functions
• Also test the similarity function by few alterations ( random & set all to 1)
• As we used PCC and VSS for evaluation
1/10/2017 128
![Page 129: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/129.jpg)
1/10/2017 129
![Page 130: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/130.jpg)
1/10/2017 130
![Page 131: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/131.jpg)
Conclusion and Future Work
• Two general algorithms are proposed that imposed social regularization using PCC and VSS
• Quite generic method also can be applied to trust aware recommendation problems
• Comparison shows it outperforms the state of the art RSTE method
• Make it more better if we have user's information about Clicking behavior and Tagging Records
• To make it more realistic we can use categorical cluster wise approach
1/10/2017 131
![Page 132: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/132.jpg)
References
• D. D. Lee and H. S. Seung. Learning the parts of objects by non-negative matrix factorization. Nature, 401(6755):788–791, Oct. 1999.
• R. Salakhutdinov and A. Mnih. Probabilistic matrix factorization. In Advances in Neural Information Processing Systems, volume 20, 2008.
• H. Ma, I. King, and M. R. Lyu. Learning to recommend with social trust ensemble. In Proc. Of SIGIR ’09, pages 203–210, Boston, MA, USA, 2009
1/10/2017 132
![Page 133: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/133.jpg)
Conclusions and Comparison
1/10/2017 133
SoRec SocialMF SRSModel Based V V V
Method Co-factorization Regularization methods Regularization methods
Dataset - Epinions03 V V
Dataset - Epinions02 V
Dataset - Douban V
Dataset - Flixster V
Error Metric - RMSE V V
Error Metric - MAE V V
![Page 134: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/134.jpg)
Conclusions and Comparison
1/10/2017 134
Best MAE Best RMSE ContextSoRec – Epinions 0.8932 --- Dimensionality = 10
99% Training Data
SocialMF – Epinions [2] --- 1.075 80% Training Data
SociaIMF - Flixter -- 0.815 5-fold CV.
SRS – Epinions 0.8256 1.0739 PCC, Individual Method90% Training Data
SRS - Douban 0.5543 0.6988 PCC , Individual Method80% Training Data
![Page 135: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/135.jpg)
Conclusions and Comparison
1/10/2017 135
Social recommendation may also perform worse than tradicionalrecommender systems:
• social network composed of valuable friends, casual friends andevent friends; users are not necessarily all that similar;
• social relations mixed with useful and noise connections;
• users with fewer ratings are likely to also have fewerconnections.
Issues on Social Recommendation
![Page 136: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/136.jpg)
QUESTIONS
1/10/2017 136
![Page 137: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/137.jpg)
Backup Slides - SRS
1/10/2017 137
![Page 138: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/138.jpg)
NMF
• Originally Used for image Analysis, But now widely used in Collaborative Filtering (For recommendation uses User Item matrix )
• algorithm for non-negative matrix factorization that is able to learn parts of faces and semantic features of text.
• This is in contrast to other methods, such as principal components analysis and vector quantization,
• that learn holistic, not parts-based, representations. Non-negative matrix factorization is distinguished from the other methods by its use of non-negativity constraints. These constraints lead to a parts-based representation because they allow only additive, not subtractive, combinations.
• When non-negative matrix factorization is implemented as a neural network, parts-based representations emerge by virtue of two properties: the firing rates of neurons are never negative and synaptic strengths do not change sign.
1/10/2017 138
![Page 139: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/139.jpg)
PMF (Probabilistic Matrix Factorization)
• model which scales linearly with the number of observations and, more importantly, performs well on the large, sparse, and very imbalanced Netflix dataset
• users who have rated similar sets of movies are likely to have similar preferences
• When the predictions of multiple PMF models are linearly combined with the predictions of Restricted Boltzmann Machines models, we achieve an error rate of 0.8861, that is nearly 7% better than the score of Netflix’s own system.
1/10/2017 139
![Page 140: Social Regularization - Universität Hildesheim · 2020-06-25 · The narrow definition is used in the context of this presentation. 1. Intro to Social Recommendation Narrow x Broad](https://reader035.fdocuments.us/reader035/viewer/2022070912/5fb419edf1cd6e356248b7aa/html5/thumbnails/140.jpg)
RSTE (RECOMMENDATION WITH SOCIAL TRUST ENSEMBLE )
• Aiming at modeling recommender systems more accurately and realistically, we propose a novel probabilistic factor analysis framework, which naturally fuses the users’ tastes and their trusted friends’ favors together.
• term Social Trust Ensemble (RSTE) to represent the formulation of the social trust restrictions on the recommender systems.
• Uses the epinion Dataset
1/10/2017 140