Published on January 19, 2008
Controversial Users demand Local Trust Metrics: an Experimental Study on Epinions.com Community Paolo Massa PhD student in ICT ITC/iRST and University of Trento Blog: http://moloko.itc.it/paoloblog/ (Joint Work with Paolo Avesani) (Thanks Epinions.com for providing data) Slides licenced under CreativeCommons AttributionShareAlike (see last slide for more info)
Summary ■What is Epinions.com ■Trust Networks and Trust Metrics Local vs Global Trust Metrics ■Controversial Users ■Experiments and Results ■Conclusions
Motivation In a society, some peers are unknown to you (ebay, p2p, ...) Q: “Should I trust peer A?” [decentralization>relevant] Most papers assume a peer has a unique quality value (there are good peers and bad peers, goal is to spot bad) IRREALISTIC assumption (Evidence from real online community of 150.000 users). Consequence: we need Local Trust Metrics (personalized) [But most papers propose Global Metrics]
Epinions.com What is Epinions.com? Community web site where users can Write reviews about items and give them ratings Express their Web of Trust (“Users whose reviews and ratings you have consistently found to be valuable”) Express their Block List (“Users whose reviews and ratings ... offensive, inaccurate, or in general not valuable”) Reviews of TRUSTed users are more visible Reviews of DISTRUSTed users are hidden
Epinions.com Dr.P profile page Dr.P's Web of Trust (Block List is hidden) Do you trust or distrust Dr.P? Ratings given by Dr.P
Real uses of Trust News sites: Slashdot.org, Kuro5hin.org, ... Emarketplaces: Ebay.com, Epinions.com, Amazon.com, ... P2P networks: eDonkey, Gnutella, JXTA Jobs sites: LinkedIn, Ryze, ... Friendster, Tribes, Orkut and other “social” sites. Opensource Developers communities: Advogato.org (Affero.org) Hospitalityclub.org, couchsurfing.com: hosting in your house unknown people? Bookcrossing and lending stuff sites. Network of personal weblogs (the blogroll is your trust list) Semantic Web: FOAF (FriendOfAFriend) is an RDF format that allows to express social relationships (~10 millions files) and XFN microformat PageRank (Google) ... MyWeb2.0 (Yahoo!)
Trust networks (are graphs) Aggregate all the trust statements to produce a trust network. A node is a user (example: Dr.P). A direct edge is a trust statement 0 Mena Ben In Epinions, 0.2 Properties of Trust: just 1 and 0! 0.9 weighted (0=distrust, 1=max trust) 0.6 subjective 1 Dr.P Doc asymmetric contextdependent Trust Metric (TM): ? Uses existing edges for predicting values of trust for nonexisting edges, thanks to trust propagation (if you trust someone, then you have some degree of trust in anyone that person trusts).
TM perspective: Local or Global 1 1 Mary Mena Bill How much Bill can be trusted? 0 On average (by the community)? By Mary? ME 1 Doc And by ME? Global Trust Metrics: “Reputation” of user is based on number and quality of incoming edges. Bill has just one predicted trust value (0.5). PageRank (Google), eBay, Slashdot, ... Works badly for controversial people Local Trust Metrics Trust is subjective > consider personal views (trust “Bill”?) Local can be more effective if people are not standardized.
Controversial Users Intuitively: a Controversial User is TRUSTED by some users DISTRUSTED by some users Do you want an example?
Controversial Users: an example 1 0 1 0 1 0 1 0 1 0 (....) (....) 1 0 100M people 100M people If you don't know Bush, should you trust Bush? T(Bush)=0.5? Make sense? Here global metrics don't.
Controversial Users: an example 1 0 1 0 1 0 1 1 1 0 R 1 0 D 1 1 (....) (....) 1 0 100M people 100Mpeople Local Metric makes more sense. Your trust in Bush depends on your trusted users! T(R,Bush)=1 T(D,Bush)=0
Controversial Users on Epinions Controversial users are normal in societies How many controversial users on Epinions.com? But first, two definitions of Controversiality: Controversiality Level of A: number of users that disagree with the majority = Min(#trust, #distrust) Contr. “Percentage” of A = (TD) / (T+D) in [1, 1] CP(A)=1 if A is trusted by everyone (loved!) CP(A)=1 if A is distrusted by everyone (hated!) CP(A)=0 if A is trusted by n users and distrusted by n users
Experiment Epinions.com dataset Real Users: ~150K Edges (Trust / Distrust): 841K (717K / 124K) ~85K received at least one judgement (trust or distrust) 17.090 (>20%) are at least 1controversial (at least 1 user disagrees with the majority) > Non negligible portion! 1.247 are at least 10controversial 144 are at least 40controversial 1 user is 212controversial! (~400 trust her, 212 distrust her)
Experiment Comparing 2 metrics about accuracy in trust/distrust prediction. Global: ebaylike. Trust(A)=#trust/(#trust+#distrust) Local: MoleTrust, based on Trust Propagation from current user (simple and fast) Cycles are a problem > Order peers based on distance from source user Trust of users at level k is based only on trust of users at level k1 (and k) Trust propagation horizon & decay
Experiment How do we compare metrics? Leaveoneout: Remove an edge in Trust Network and try to predict it. Then compute error as absolute difference between Real and Predicted value. Also differentiating over trust or distrust statements
Exp. on Controversiality Level y=error made by TM predicting edges on users with x Error controversiality level. Predicting Distrust is more Ebay Controversiality level difficult. Ebay error on Distrust ~ 0.6 Mole2 error on Distrust ~ 0.4 Error Error on Trust is similar because (#trust >> #distrust) MoleTrust2 Controversiality level
Exp. on Controversiality Percentage CP~0 = Controversial User Error Ebay = 0.5 on Contr.Us Error Error MoleTrust2 smaller but not as small as we Ebay Controversiality percentage would like: can we reach 0? Other experiments in paper: Error on Trust Edges. Error Error on Distrust Edges (very important to correctly predict these ones!) MoleTrust2 Controversiality percentage
Other experiments MoleTrust with different propagation horizons 2, 3, 4 Computing Coverage.
Conclusions In complex societies, it is normal that someone likes you and someone dislikes you. Most Papers make assumption of unique quality value for a peer (and propose an algo for predicting it) This is IRREALISTIC! (I know this is intuitive but still ...)
Conclusions (2) As a consequence, we need Local Trust Metrics. Local TMs are computationally much more expensive than Global TMs! > Possibly, you run it locally for yourself on your mobile or on your browser (should be fast!) Local TMs exploits less information > reduced coverage. Global Metrics fine in noncontroversial domains: possibly ok on Ebay, surely not ok on sites about (political?) opinions Trust networks are everywhere! More research is needed: Yahoo! (with MyWeb2.0) and Google are there. More real testbeds, more proposals of Local TMs, more comparisons, ...
Licence of this slides These slides are released under Creative Commons AttributionShareAlike 2.5 You are free: * to copy, distribute, display, and perform the work * to make derivative works * to make commercial use of the work Under the following conditions: Attribution. You must attribute the work in the manner specified by the author or licensor. Share Alike. If you alter, transform, or build upon this work, you may distribute the resulting work only under a license identical to this one. * For any reuse or distribution, you must make clear to others the license terms of this work. * Any of these conditions can be waived if you get permission from the copyright holder. Your fair use and other rights are in no way affected by the above. More info at http://creativecommons.org/licenses/bysa/2.5/
The End The End. Thanks for your attention! Questions?
Controversial Users demand Local Trust Metrics: an Experimental Study on Epinions.com Community Paolo Massa and Paolo Avesani ITC-iRST Via Sommarive 14 - I ...
Controversial users demand local trust metrics: an experimental study on Epinions.com community
Controversial users demand local trust metrics: An experimental study on epinions.com community (2005)
Controversial Users demand Local Trust ... an Experimental Study on Epinions.com Community ... Epinions.com community, controversial users are a ...
Controversial Users Demand Local Trust Metrics: An Experimental Study on Epinions.com Community. Paolo Massa, Paolo Avesani. In today’s connected world ...
CiteSeerX - Scientific documents that cite the following paper: Controversial users demand local trust metrics: an experimental study on epinions.com community
Epinions trust. About this network ... Controversial users demand local trust metrics: an experimental study on epinions.com community.
Epinions product ratings network dataset ... Controversial users demand local trust metrics: an experimental study on epinions.com community.
Trust Metrics on Controversial Users on ResearchGate, the professional network for scientists.