May 17, 2005
| Peer to peer |
peer_to_peer
|
| PhD |
phd
|
| Trust and Reputation |
trust_and_reputation
|
In the following some comments about the Program of the second day.
Epidemics spreading
Alessandro VESPIGNANI
Indiana University, Bloomington, U.S.A.
Great Lecture! For now, Vespignani is my preferred lecturer. The lecture was very interesting and spoke about how to model deseases and their spreading and how you can immunitize nodes in order to stop the virus (immunize at random is not effective in general, immunizing the hub [nodes with a lot of connections] works very well [you immunize a tiny fraction of the nodes and the virus is stopped] but you have to have a global knowledge of the network that is often not the case. Taking node at random, asking to name a "connection" and immunize that connection works almost as well as immunizing the hubs but you don't need global knowledge: in fact you are probably going to reach the hubs with this strategy.
He was speaking about sex networks, with the (to be expected) note that, when asked explicitly abou the number of sexual connections, men tend to overstimate and women tend to understimate. He said that there are already data on sexual networks from Sweden, US, UK, Zimbabwe, Uganda. Remind for myself: I need to have a look at them and download the data.
[Everyone here is speaking of Mean Field, but I have no idea what this is. Most of the people here are physicists. I had a check on the Mean Field Wikipedia page but I didn't get it too much. I'll have to ask someone, even if this is a bit like asking "what do you mean by 2+2 after 5 years of math".]
Official visit - the Director General of UNESCO Mr. K. MATSUURA
I arrived late and it was boring.
Percolation on networks II
Shlomo HAVLIN
Bar-Ilan University, Ramat-Gan, Israel
Too mathematical for me. Certainly Havlin knows a lot about these things.
Models of scale free graphs
Guido CALDARELLI
Very good speaker as well! The slides were perfect. There was some speaking about fractals, about fractals networks and pseudo-fractal graphs. He spoke about cosing.org where there also some datasets to play with. He made some fun remarks at the end about the sexual networks but I can't remember them. I need to get the slides that were perfect.
Search in random networks
Lada ADAMIC
HP Laboratories, Palo Alto, U.S.A.
I read most of her papers and it was great to see that there is a real person behind the name. Her lecture was interesting and not too hard to follow. Tomorrow she will speak about search on scale-free networks. I need to understand better what are these cutoffs in powerlaw distributions.
Network visualization
Katy BORNER
Indiana Unviersity, Bloomington, U.S.A.
In the next days we will have 3 computer sessions with her in which we will play with datasets and she also asked to bring our datasets. I think it will be fun. I read a paper by her some time ago about activeWorld and visualizations of footprints in a virtual library: very interesting! Anyway this talk was very different from the previoous talk, it was much more graphical, sometime phychological and surely not at all a formulas-dense one!
I will try to get the slides of this as well. Interesting picture of sociogram from a 1934 (!) paper by Moreno. Red and green has swapped meaning in China, compared to Italy.
I suggested her to look at the TrustArt wiki page I somehow maintain, there could be some interesting visualizations she had missed.
Game theory
Fernando VEGA-REDONDO
Universidad de Alicante, Spain
He was great in condensing the all field in one hour and in doing it in a very intriguing way, basically only by examples. Very interesting and at the end he started explained how game theory relates to networks (formation). I took 6 pages of notes. The last line I wrote on my notes is "Economy is the science of greediness and egoism" (you know, all these assumptions about humans being rational, i.e. interested just in maximizing their instantaneous utility not caring for all the rest...)
October 15, 2004
| Peer to peer |
peer_to_peer
|
| Privacy |
privacy
|
| Social Software |
social_software
|
When millions of users will have Desktop.google.com installed, Google will simply release a new version in which the user can check a box and say "Share the files in my disk" (maybe only files in a certain directory). This will create in a second an enormous P2P (peer-to-peer) network, in which you can search for files directly on other users' disks. What do you think? Make sense?
UPDATE: If I were Google, I let users choose also "share your files only with your friends on Orkut". In this way Orkut would becoma a uber-useful network (now is a bit pointless), and Google Corporation will have all the worlds users for all the services. And increase what it knows.
October 13, 2004
| Free software |
free_software
|
| Peer to peer |
peer_to_peer
|
| PhD |
phd
|
| Social Software |
social_software
|
Citeseer is less useful today than how it was 2 years ago. It seems they stop the crawling looking for papers. [I have a project about adding "web of trust" to citeseer so that every user can express a degree of interest in another users' kept bibliography) but it seems I never have the time to seriously start it.] Anyway this post is to cite 2 interesting related projects: LionShare and Eprints ... (read below for links and details)
LionShare P2P project (see the search screenshot on the wiki) allows people to stop thinking P2P is illegal by default. It is Java Open Source code.
"LionShare P2P project is an innovative effort to facilitate legitimate file-sharing among individuals and educational institutions around the world. By using Peer to Peer (P2P) technology and incorporating features such as authentication, directory servers, and owner controlled sharing of files, LionShare promises secure file-sharing capabilities for the easy exchange of image collections, video archives, large data collections, and other types of academic information. In addition to authenticated file-sharing capabilities, the developing LionShare technology will also provide users with resources for organizing, storing, and retrieving digital files."
(found via the iper-interesting Italian WikiLab).
The other interesting project is more mature and is called Eprints.org - Self-Archiving and Open Access (OA) Eprint Archives. The software is called GNU eprints and it is of course free software. At present, there are 141 known archives running EPrints software worldwide. It is a sort of p2p network where peers are libraries (also University libraries) that certificates that the papers they host are real papers from their scholars. I think that every single researcher can have her own instance of the peer but I think noone is doing it. [Anyway I didn't investigate too much the project and I could be wrong]. My University Library is one of the 141, good! Among the softwares, there is CiteBase whose goal is, I guess, very similar to citeseer.
You can also admire some powerlaws (are they ubiquitous?) in these graphs (Java applets).
The different eprints peers keep them up to date using the Open Archives Initiative Protocol for Metadata Harvesting.
Ok, I must admit I didn't check the projects in great details so if you do it (or even install them!), leave some comments (if you feel like). Thanks!
February 27, 2004
I realised today I didn't write yet an entry about my PhD Research Proposal "Trust-aware Decentralized Recommender Systems" (TaDRS).
So here it is the PDF file. If you have any comment or criticism, I'll be happy to hear from you.
The PhD research proposal is a little bit outdated (29th May 2003) but I didn't have a blog at that time. Enjoy and let me know what you think.
UPDATE:
Abstract
This PhD thesis addresses the following problem: exploiting of trust information in order to enhance the accuracy and the user acceptance of current Recommender Systems (RS). RSs suggest to users items they will probably like. Up to now, current RSs mainly generate recommendations based on users' opinions on items. Nowadays, with the growth of online communities, e-marketplaces, weblogs and peer-to-peer networks, a new kind of information is available: rating expressed by an user on another user (trust). We analyze current RS weaknesses and show how use of trust can overcome them. We proposed a solution about exploiting of trust into RSs and underline what experiments we will run in order to test our solution.
November 14, 2003
| Peer to peer |
peer_to_peer
|
| Social Software |
social_software
|
| Trust and Reputation |
trust_and_reputation
|
I've just finished reading a very interesting paper Trust Management for the Semantic Web by Matthew Richardson, Rakesh Agrawal, and Pedro Domingos.
There are a lot of interesting ideas and theoretical analysis about trust and propagation (it defines or explains some concepts such as path algebra, generalized transitive closure algorithms, well-formed decomposable path problems, global invariance, cycle-indifferent combination function, ...).
Among the many interesting ideas, I liked a lot the following simple idea partially inspired by pagerank (I think).
Let me give you first a short explanation of how pagerank (one of the algorythms behind google.com) works.
PageRank ideally performs a random walk from webpage to webpage following a link at random and remembering how many times it passed in a web page; this number is the pagerank of the web page. Of course a random surfer will pass in Yahoo.com more often than in my homepage and in fact yahoo.com will have a high pagerank while my homepage will have a lower one.
At every step there is a (small) probability to jump to a random web page, so that the random surfer does not get stuck in a clique of highly connected web pages and every webpage has some chance to be taken into account.
It is proven that the random surfer can start in whatever webpage and the final pageranks will be the same; only the time needed to converge will be different.
If the description was not clear, you can read the the explanation given by Google or by Webworkshop.net or by Ian Rogers. Otherwise you can read the original paper or you can even construct a web graph and try pagerank.
So the authors of Trust Management for the Semantic Web, similarly, propose:
Imagine a random knowledge-surfer hopping from user to user in search of beliefs. At each step, the surfer probabilistically selects a neighbor to jump to according to the current user's distribution of trusts. (...) Further, choosing which user to jump to, the random surfer will, with probability delta (in [0,1]) ignore the trusts and instead jump directly back to the original user, i.
So here the very simple idea:
the random surfer starts in i (the user who want to predict how much she should trust other unknown peers), then it jumps to other users depending on the trust distribution (following with higher probability, higher trust links).
But (and here there is the difference!) with probability delta it will not follow a trust link but instead jump directly to i (our starting point user).
In this way you can choose how much you want the opinion of i taken into account and basically how much the random surfer is allowed to go far away from i!!!
As in the paper about pagerank, there is also the probabilistic interpretation and the interpretation as operations on matrixes.
Actually on pagerank, doing the random walk is equivalent to compute the first eigenvector of the link matrix.
Ok, if you trust me, you should read it!
November 05, 2003
| FOAF |
foaf
|
| Peer to peer |
peer_to_peer
|
| Social Software |
social_software
|
| Trust and Reputation |
trust_and_reputation
|
My Erdös Number is 5.
Erdös numbers have been a part of the folklore of mathematicians throughout the world for many years.
Essentially, the Erdos Number Project studies research collaboration among mathematicians.
If you are curious about your Erdos number, you can compute it.
If you have written a paper with Paul Erdös, you have Erdos Number 1. If you are a coauthor of a coauthor, you have Erdös number 2. And so on.
Here is the mail I receive when I asked to compute my Erdos Number.
Erdos number
Your Erdos number is at most 5, via this path of length 4 to Aguzzoli:
50 #9662 Erdos, P.; Hajnal, A.; Shelah, S. On some general properties of
chromatic numbers. Topics in topology (Proc. Colloq., Keszthely, 1972),
pp. 243--255. Colloq. Math. Soc. Janos Bolyai, Vol. 8, North-Holland,
Amsterdam, 1974.
85i:03125 Makowsky, J. A.; Shelah, S. Positive results in abstract model
theory: a theory of compact logics. Ann. Pure Appl. Logic 25 (1983), no.
3, 263--299.
819 550 Makowsky, J. A.; Mundici, D. Abstract equivalence relations.
Model-theoretic logics, 717--746, Perspect. Math. Logic, Springer, New
York, 1985.
96d:14048 Aguzzoli, Stefano; Mundici, Daniele. An algorithmic
desingularization of $3$-dimensional toric varieties. Tohoku Math. J. (2)
46 (1994), no. 4, 557--572.
S. Aguzzoli, P.Avesani, P.Massa
"Collaborative Case-Based Recommendation Systems"
Lecture Notes in Computer Science, Lecture Notes in Artificial Intelligence
Volume 2416, 2002.
October 30, 2003
| Maryland |
maryland
|
| Peer to peer |
peer_to_peer
|
| PhD |
phd
|
| Recommender Systems |
recommender_systems
|
| Trust and Reputation |
trust_and_reputation
|
As part of my PhD program, I'll spend the next 3 months at the University of Maryland. I'll stay here until January 17, 2003.
I also opened up a photo gallery in which I'll post photos taken here.
