Analyzed trust metrics

TODO: reorganize this page, possibly by automatically collecting pages in the category trust metrics with their implementation status. Template?

What are the compared trust metrics?

Here we list of trust metrics we plan to compare. For each one we've collected the name, the reference paper, the status of code completion, the input, the output, the assumptions and notes.

Coded metrics

We have implemented the following metrics in our code, which is available under the GNU General Public Licence (GPL):

• mole trust metric
• guakamole
• ebay

to be tested

• Advogato

Analyzed metrics

Zero at the moment, I need to bring order to all the material I accumulated...

A concise categorization of trust metrics, from trustcomp

Metrics not yet analyzed

Name: Richardson et al's trust metric

Reference:

• Trust Management for the Semantic Web http://www.cs.washington.edu/homes/mattr/doc/iswc2003/abstract.html
• comment here http://moloko.itc.it/paoloblog/archives/2003/11/14/trust_management_for_the_semantic_web.html

Input = two variants:

• A general trust graph (may include cycles), with weighted edges (weight denotes trust).
• A trust matrix (trust of A in B for each A,B) and a belief vector (belief of A in a statement, for each A).

Output =

• A transitive closure of the graph, with weighted edges.
• A merged trust matrix (merged trust of A in B for each A,B) and a merged belief vector.

Notes =

• For the graph variants, two functions are defined: concatenation (e.g. multiplication) and aggregation (i.e. maximum, minimum, average).
• For the matrix variant, a Markov model is used.

Name: Mui et al's trust metric.

Reference: Mui, ... : A computational model of Trust and Reputation (imatch MIT) http://www.cdm.lcs.mit.edu/ftp/lmui/computational%20models%20of%20trust%20and%20reputation.pdf

Code completion =

Input =

Output =

Assumptions =

Notes =

Name: Aberer et al's trust metric.

Reference: Aberer, ... : Managing Trust in a Peer-2-Perrs Information System. http://citeseer.nj.nec.com/aberer01managing.html

Code completion =

Input =

Output =

Assumptions =

Notes =

Name: Pagerank (Eigentrust)

Reference: Page,Brin: Bringing order to the web, Eigentrust...

Code completion =

Input = Social network (a directed graph of links/trust)

Output = A rank for every single page depending on the incoming/outgoing links and relative importance

Assumptions = a link from a relevant peer is more important than a link from an unknown peer.

Notes = one of the algorithms behind google! It computes how much a page is "important" based on the number and quality of ingoing and outgoing links. It computes a global value, equal for every peer: i.e. the value of CNN is the same for me and for you even if you happen to prefer Indymedia

Name: Josang's trust metric.

Reference: many publications here http://sky.fit.qut.edu.au/~josang/publications.html

Code completion =

Input =

Output =

Assumptions =

Notes = Impressive java applet demos here:

• http://sky.fit.qut.edu.au/~josang/sl/ (new)
• http://security.dstc.edu.au/spectrum/trustengine/ (old)

Name: eBay trust metric.

Reference: (some paper that analyze the auctions on eBay)

Code completion =

Input = statement (+1, 0, -1) made by seller/buyer on buyer/seller after every successful auction. It uses the date information

Output =

Assumptions =

Notes = it is a very trivial metric (just weights every "link" the same, and only reports some statistics for every user (sum of trust over 6 months, past month, past week). Even if so simple, there are reports that say it is to difficult to understand for the average user.

Name: yet another eBay trust metric.

Reference: (O'Donovan, Smyth Extracting and Visualising Trust from Online Auction Comments)

Code completion =

Input = a set of feedback comments

Output = a trust graph of the auction (using a modified version of paul mitton's PieSpy graph generator)

Assumptions =

Notes = this one takes in the feedback comments and the user ids, uses NLP and a lexicon of positive and negative terms and generates a trust model for the transactors. Uses real ebay data crawled from the egyptian antiques pages.

Name: Newsmonster (Fionna)

Reference = Desing of the OpenPrivacy Distributed Reputation System. Webpage, 30/5/2002. Burton, K. A; ht