Meeting 3 02 2004

Meeting 3 02 2004

Present: Avesani, Massa

= Traverso =

Speak with him on Thursday or Friday

= MontagneSicure =

survey For Thursday! Todo!
For project documentation, produce


 * survey on trust metrics (exhaustive and structured) + motivations for trust metrics (cut and paste from proposal) (see AnalyzedTrustMetrics)
 * paper on trust analysis of epinions

wireless and research
think about research use of wireless services (sms, mms, moblog, ...)

THINK!

moblog open source server

http://easymoblog.sourceforge.net/

http://pansapien.com/geekstuff/moblog/

trust metrics
goal: ranking itineraries based on propagated trust

1. design information input

1. design changes to interfaces to get the new information

1. design which information to log (and which offline test we will run)

1. design how to publish info [foaf+trust (http://www.perceive.net/schemas/20021119/trust/ with context "country skiing"), xfn, rss, rdf, ...]

1. design trust metric (see link on cocoacodingwiki that now is down!)

1. implement trust metric in Java

deadline: september 2004

= Charles =

check the state of cocoablog, state of the art (what is there, what is missing)

= Conor =

* cocoa as a RS, scientific results and papers (with charles)

what is possible to extract from cocoa logs? ratings? trust? this could become a public dataset as eachmovie!

* Blog su cocoa

* Trust metrics * recommendation peer to peer

* smart radio (dublin)

* Phd candidates

* readings groups and discussions

= Papers / Phd activity =

see TrustPapers

papers about moleskiing and with a first trust metric
TODO!

Deadline March 29, 2004

AH2004 	 http://www.ah2004.org/

companion papers on epinions (relationship us and trust)
TODO!

Deadline: May 30 2004 (abstract)

Coopis 12th International Conference on COOPERATIVE INFORMATION SYSTEMS (CoopIS 2004) 	 http://www.cs.rmit.edu.au/fedconf/

evaluation of recommender systems
slides dublino

online evaluation is better

considering time

(see WorkingPaper)

Trust and Collaborative Filtering: A Linear algebraic framework
(see WorkingPaper)

trust metrics survey
(see AnalyzedTrustMetrics)

Book trust metrics
see BookTrustMetrics

possible ideas for new papers
* considering one rating statement or one trust statement as a cognitive bit, which cognitive bit gives more information? (experiments on epinions)

* are ther significant differences in global trust (same for everyone) and local trust (personalized)? or every user likes in the same way the other users? is jimmy reputed 0.9 liked by everyone 0.9? (test on advogato). That is, is a personalized trust computation useful or not? could we give to everyone the global trust and personalized (since requires computation) offered only behind payment or self made by every user?

= I should focus my research on =

how to combine trust and ratings
avesani proposed iterative way a la EM (expectation maximization)

using trust you predict some ratings, you fill the ratings in the matrix, you how have more ratings and you can use them to fill the trust matrix (only when the user profile agree with the trust profile), with this new filled trust matrix you can better predict more ratings, ... [iterating].

You can keep under control the similarity/trust coefficient and monitor if they converge or not.

global versus local trust metrics
computational requirements

are ther significant differences in global trust (same for everyone) and local trust (personalized)? or every user likes in the same way the other users? is jimmy reputed 0.9 liked by everyone 0.9? (test on advogato)

= do some presentations =

present pagerank (using raghavan lessons) http://www.stanford.edu/class/cs276a/handouts/lecture12-6in1.pdf http://www.stanford.edu/class/cs276a/handouts/lecture11-6in1.pdf

--

Grin Conferences for deciding the level of conferences

[avesani can you write the url here?]