Home | Search | Help  
Home Page Università di Genova
DISI -> People -> PostDoc -> Matteo Dell'Amico -> PhD -> PhD Activity

Last updated December 16th, 2007 by Matteo Dell'Amico.

Leave of Absence

Phd Courses and Credits

Teaching Assistant Activity

Research Activity

  • Supervisor: Giovanni Chiola
  • Research interests: Peer-to-peer networks and middleware, reputation and trust, incentives to cooperation, complex and social networks, anonimity in P2P networks.

Publications

Conferences, Seminars, Short Visits

  • Matteo Dell'Amico, Licia Capra. SOFIA: Social Filtering for Niche Markets. 9/10/2007, UCLMobiSys seminar, London, Great Britain.
  • Matteo Dell'Amico. Mapping Small Worlds: Making Computers Understand the Six Degrees of Separation. 24/08/2007, UCLMobiSys seminar, London, Great Britain.
  • Seventh IEEE International Conference on Peer-to-Peer Computing (P2P2007), September 2-5, 2007, Galway, Ireland.
  • Web-MINDS: Riunione finale. DISI, Università di Genova. 27-28/10/2006.
    My talk: Valutare la reputazione usando reti sociali (slides).
  • Matteo Dell'Amico. Neighbourhood Maps: Decentralised Ranking in Small-World P2P Networks. 28/9/2006, UCLMobiSys seminar, London, Great Britain. (slides).
  • Second European Conference on Complex Systems 2006 (ECCS '06). 25-29/9/2006. Oxford, Great Britain.
  • Fourth International Conference on Trust Management (iTrust 2006). 16-19/5/2006. Pisa, Italy.
  • 20th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2006). 25-29/4/2006, Rhodes Island, Greece.
  • Seminario sulle reti complesse. Alessandria, Università del Piemonte Orientale, 4/4/2006.
    My talk: Preferential attachment e clustering: Estendere l'algoritmo di Barabasi-Albert (slides).
  • First Summer School on Aspects of Complexity. University of Bologna Residential Center Bertinoro (Forlì), Italy, 18-28/07/2005.
  • Meeting annuale del progetto WebMinds. Salerno, 20-22/06/2005.
    My talk: Mappe dei dintorni: calcolo efficiente di relazioni fra nodi in reti small-world.
  • One day workshop on Game Theory and Engineering. Politecnico di Milano, 09/06/2005.

Annual Reports

  • 2007. Work on network layout started in the previous year yielded very satisfying results, and yielded a paper that was presented in the IEEE P2P2007 conference. This method can be combined with the neighborhood maps technique to obtain even better results when estimating reputation metrics.
    During the six months spent at University College London, I worked on defences from Sybil attacks (also known as "profile injection" or "shilling") in collaborative filtering systems. The Sybil attack consists in creating a very high number of fake identities corresponding to just one legitimate user, in order to obtain a higher weight in the dynamics of the network. We developed a technique, called "social filtering", that combines information about tastes with reputation derived from a web of trust. Our experimental results show that social filtering is effective in defending from the Sybil attack, and that addition of information from the web of trust can help in obtaining results that have a higher quality even in the absence of malicious attack.
    An overview of various papers I have read is available on CiteULike.
  • 2006. Progress report (pdf)
    In this year I worked on methods for evaluating reputation in P2P networks, starting from a network of recommendations between peers (_web of trust_).
    Webs of trust can be considered as complex social networks. Exploiting common properties of this kind of networks can be crucial in obtaining efficient and useful means for evaluating reputation.
    _Neighbourhood maps_ exploit the _small-world_ and _fast-mixing_ properties in order to give an approximation of interesting metrics such as distance between nodes and PageRank. I am now working in extending the precision and the class of metrics that can be evaluated using that technique.
    High clustering can be explained by considering nodes as placed in an _n_-dimensional layout, where closeness reflects affinity between nodes (_i.e._, higher probabilty of being linked). A simple extension to the Barabási-Albert preferential attachment model can create networks that are both _scale-free_ and _small-world_ (power law degree distribution, high clustering, small diameter).
    Since an underlying layout can be a good explanation of the high clustering of such a network, I am working on techniques taken from graph-drawing algorithms in order to give a layout to a given network and using that layout to find short paths in webs of trust. Preliminary experimentations show promising results.
  • 2005. Thesis proposal (pdf)
    During this year I studied the problem of incentives to cooperation in peer-to-peer applications; this approach exists in order to combat _free riding_, the behaviour of exploiting resources without contributing anything in return.
    Two of the most promising approaches are _direct_ and _indirect_ reciprocation.
    In direct reciprocation, the quality of service (QoS) a peer gives to other nodes is proportional to the QoS they gave to them. This approach is implemented in successful networks such as BitTorrent and Edonkey, and is viable when nodes have a high probability of having repeated interactions.
    Indirect reciprocation is based on _reputation_, that is a judgement from the whole network to a peer. While this approach is more powerful, it is challenging to be implemented efficently in large-scale P2P networks.
    Game theory, social networks and link analysis ranking are connected topics that I studied. I made available on the web an essential bibliography.
  The Department | Research | Education | People | Contacts