MS Research Progress Presentation

Progress ofProgress of
Research WorkResearch Work
03/16/15 1

User Similarity Computation forUser Similarity Computation for
Collaborative Filtering UsingCollaborative Filtering Using
Dynamic Implicit TrustDynamic Implicit Trust
03/16/15 2

Presented byPresented by
Falguni Roy
MSSE- 0209
Supervised bySupervised by
Sheikh Muhammad Sarwar
03/16/15 3
User Similarity Computation for CF
Using Dynamic Implicit Trust

ContentsContents
 About Recommender System
 Motivation
 Literature Review
 Proposed Methodology
 Dataset
 Publication
 Remarkable Reviews
03/16/15 4User Similarity Computation for
CF Using Dynamic Implicit Trust

AboutAbout RecommenderRecommender
SystemSystem

• Most popular forms of web information
customization system
• Used in E-commerce and entertainment based
websites
– amazon.com
– Netflix.com
• Aim: To predict the 'rating' or 'preference' that
user would give to an item

Category of RSCategory of RS
• Based on the information filtering approach, there
are two category of recommender system [1]
 Content based filtering, and
 Collaborative based filtering

Content Based FilteringContent Based Filtering

Collaborative Filtering(CF)Collaborative Filtering(CF)
03/16/15 9
• Most popular approach [2,3]
User Similarity Computation for

Category of CFCategory of CF
• Based on Methodology
• Model Based Method
– item recommendation by developing a model
– regression, Bayesian network, rule-based
and clustering
• Memory Based Method
– a rating matrix
– some statistical techniques applied on the
rating matrix.

• Based on Similarity
03/16/15 11
Category of CF (Cont’)Category of CF (Cont’)

Trust based Recommender SystemTrust based Recommender System
• Guo et al. defines trust in recommender system as
“Trust is defined as one's belief towards the ability
of others in providing valuable ratings“ [4]
• Express the integrity in the relationship between
two entities.
• Trust used to scale similarity

Trust PropertiesTrust Properties
• According to trust theory [5]
 Asymmetry
 Transitivity
 Dynamicity
 Context Dependence

Type of TrustType of Trust
• On the basis of trust computation
 Explicit Trust
 Implicit Trust

Explicit TrustExplicit Trust
• Trust value is calculated by pre-existing
social link between users
• The link is defined as
 By defining web of trust
 Assign a trust statement

Implicit TrustImplicit Trust
• Derived from user-item rating matrix by
analyzing rating pattern, rating value and
historical behavior of ratings of the users
• Trustworthiness of a user is determined by the
prediction accuracy of a user in the past [6]

MotivationMotivation
Problem of existing Trust based RSProblem of existing Trust based RS
 Don’t concern about users’ changing interestsDon’t concern about users’ changing interests
 Treats users’ similarity as symmetricTreats users’ similarity as symmetric
03/16/15 17
User Similarity Computation for CF Using Dynamic Implicit Trust

Literature ReviewLiterature Review
03/16/15 18

• Qusai Shambour et al. [7] (TM1)
Computed trust based on mean squared distance (MSD)
and propagated trust based, on the MoleTrust matric
• Lathia et al. [9] (TM2)
Proposed a trust method to define degree of trust by
tracking the value of ratings provided by other users
Trust is defined as the average of provided values over all
the rated items
• Papagelis et al. [10] (TM3)
Define trust through user similarity computed by Pearson
correlation coefficient

• Hwang et al. [8] (TM4)
Computed trust by averaging the prediction error on co-
rated items
• O'Donovan et al. [6] (TM5)
A rating provided by users is correct if the absolute
difference between the predicted rating and the actual
rating is smaller than a threshold
Profile level trust and item level trust

A comparison of different trust metrics inA comparison of different trust metrics in
terms of trust properties [12][13]terms of trust properties [12][13]
Methods Asymmetry Transitivity Dynamicity Context
Dependence
TM1 [7] No Yes No No
TM3 [10] No Yes, iff s> ϴ
ϴ = 0.707
No No

ProposedProposed
MethodologyMethodology
03/16/15 22

Trust Based Similarity ComputationTrust Based Similarity Computation
• The proposed system consists of the following modules:
 Trust Computation module (TC),
Similarity Computation module (SC) and
Combined Trust and Similarity Computation module (CTSC)

Similarity Computation ModuleSimilarity Computation Module
• Extract a neighborhood of similar minded users for the target
user
• Similarity is calculated by integrating Pearson Correlation
Coefficient (PCC) and Jaccard similarity method [8] defined
as JPCC

Trust Computation ModuleTrust Computation Module
• Implicit trust is populated by defining the similarity or degree
of similarity between the users [12]
• Proposed a new method for determining the implicit trust
between the users as an integration of Mean Square Difference
(MSD) and Confidence and consider users’ changing interests
to support trust properties.

DatasetDataset

PublicationPublication
03/16/15 27

03/16/15 28
THE 4TH INTERNATIONAL CONFERENCE ON ANALYSIS OFTHE 4TH INTERNATIONAL CONFERENCE ON ANALYSIS OF
IMAGES, SOCIAL NETWORKS, AND TEXTSIMAGES, SOCIAL NETWORKS, AND TEXTS
(Yekaterinburg, Russia)(Yekaterinburg, Russia)
 AIST 2015: 24% Acceptance rate24% Acceptance rate
 Will be published in the Springer’s
Communications in Computer and
Information Science series
 The conference will be held on 9th
through Saturday, 11th
of April 2015
at Russia

Remarkable ReviewsRemarkable Reviews
03/16/15 29

ReviewsReviews
03/16/15 30
 The paper addresses an important problemaddresses an important problem in recommender systems,
that is, how to incorporate trust into the process. The authors give a
comprehensive overview of the related workcomprehensive overview of the related work. Then, two components
of the TJPCC model are introduced one by one. The proposed final
TJPCC has the key properties typical for trust. Four existing baselines
are used for comparison. According to the experimental results, the
proposed TJPCC method achieves better results.
 Pros
– The method is simple to implementThe method is simple to implement
– The problem is very relevant.The problem is very relevant.
– Good overview of the literatureGood overview of the literature,
 Cons
– Use more data sets for evaluationUse more data sets for evaluation

ReviewsReviews
 The paper starts with a detailed introduction, followed
by a broad section on related work for trust based and
dynamic recommender systems. Then, the method is
presented as a combination of similarity and trust
scores. Overall, the paper is well written and interestingthe paper is well written and interesting
to read. The descriptions are elaborateto read. The descriptions are elaborate. I recommend
this paper for publication

ReviewsReviews
03/16/15 32
 The authors present a new similarity between the
users using a novel dynamic trust definition
among users. The new trust definition satisfies
desirable properties such as being asymmetric,
transitive and dynamic, and experimental results
on movie database IMDB are favorable. However,
more experimental resultsmore experimental results that will furtherthat will further
strengthen the proposed method would bestrengthen the proposed method would be
preferablepreferable.

Findings for Next StepFindings for Next Step
 Usage of more data sets for evaluation
 More exploration of the properties of the
proposed algorithm
 Experimenting the proposed method with
different evolution metrics

ReferencesReferences
1. R. Burke, “Hybrid web recommender systems," in The adaptive web, pp. 377-408, Springer, 2007
2. M. J. Pazzani, “A framework for collaborative, content-based and demographic fitering," Articial Intelligence Review, vol. 13,
no. 5-6, pp. 393-408, 1999.
3. F. Cacheda, V. Carneiro, D. Fernandez, and V. Formoso, “Comparison of collaborative fitering algorithms: Limitations of
current techniques and proposals for scalable, high-performance recommender systems," ACM Transactions on the Web
(TWEB), vol. 5, no. 1, p. 2, 2011.
5. Guibing Guo. “Integrating trust and similarity to ameliorate the data sparsity and cold start for recommender systems”. In
Proceedings of the 7th ACM conference on Recommender systems, pages 451-454. ACM, 2013
6. Soude Fazeli, Babak Loni, Alejandro Bellogin, Hendrik Drachsler, and Peter Sloep. “Implicit vs. explicit trust in social matrix
factorization”. In Proceedings of the 8th ACM Conference on Recommender systems, pages 317-320. ACM, 2014.
7. John O'Donovan and Barry Smyth. Trust in recommender systems. In Proceedings of the 10th international conference on
Intelligent user interfaces, pages 167-174. ACM, 2005.
8. Q. Shambour and J. Lu, “A trust-semantic fusion-based recommendation approach for e-business applications,” Decision
Support Systems, vol. 54, no. 1, pp. 768–780, 2012.
9. Chein-Shung Hwang and Yu-Pin Chen. “Using trust in collaborative ltering recommendation”. In New trends in applied articial
intelligence, pages 1052-1060. Springer, 2007.
10. N. Lathia, S. Hailes, and L. Capra. “Trust-based collaborative filtering”. In Trust Management II, 2008.
11. M. Papagelis, D. Plexousakis, and T. Kutsuras. Alleviating the sparsity problem of collaborative filtering using trust inferences.
In Trust management. 2005.
12. Guibing Guo, Jie Zhang, Daniel Thalmann, Anirban Basu, and Neil Yorke-Smith. “From ratings to trust: an empirical study of
implicit trust in recommender systems.” 2014
13. Soude Fazeli, Babak Loni, Alejandro Bellogin, Hendrik Drachsler, and Peter Sloep. “Implicit vs. explicit trust in social matrix
factorization”. In Proceedings of the 8th ACM Conference on Recommender systems, pages 317-320. ACM, 2014.

MS Research Progress Presentation

More Related Content

Similar to MS Research Progress Presentation (20)

Recently uploaded (20)

MS Research Progress Presentation

Editor's Notes