TY  - THES
T1  - Decentralized link analysis in peer-to-peer web search networks
A1  - Parreira,Josiane Xavier
Y1  - 2009/11/17
N2  - Analyzing the authority or reputation of entities that are connected by a graph structure and ranking these entities is an important issue that arises in the Web, in Web 2.0 communities, and in other applications. The problem is typically addressed by computing the dominant eigenvector of a matrix that is suitably derived from the underlying graph, or by performing a full spectral decomposition of the matrix. Although such analyses could be performed by a centralized server, there are good reasons that suggest running theses computations in a decentralized manner across many peers, like scalability, privacy, censorship, etc. There exist a number of approaches for speeding up the analysis by partitioning the graph into disjoint fragments. However, such methods are not suitable for a peer-to-peer network, where overlap among the fragments might occur. In addition, peer-to-peer approaches need to consider network characteristics, such as peers unaware of other peers' contents, susceptibility to malicious attacks, and network dynamics (so-called churn). In this thesis we make the following major contributions. We present JXP, a decentralized algorithm for computing authority scores of entities distributed in a peer-to-peer (P2P) network that allows peers to have overlapping content and requires no a priori knowledge of other peers' content. We also show the benets of JXP in the Minerva distributed Web search engine. We present an extension of JXP, coined TrustJXP, that contains a reputation model in order to deal with misbehaving peers. We present another extension of JXP, that handles dynamics on peer-to-peer networks, as well as an algorithm for estimating the current number of entities in the network. This thesis also presents novel methods for embedding JXP in peer-to-peer networks and applications. We present an approach for creating links among peers, forming semantic overlay networks, where peers are free to decide which connections they create and which they want to avoid based on various usefulness estimators. We show how peer-to-peer applications, like the JXP algorithm, can greatly benet from these additional semantic relations. 
KW  - Peer-to-Peer-Netz
KW  - World Wide Web
KW  - Hyperlink
KW  - Graph
KW  - Algorithmus
CY  - Saarbrücken
PB  - Universitäts- und Landesbibliothek
AD  - Postfach 151141, 66041 Saarbrücken
UR  - http://scidok.sulb.uni-saarland.de/volltexte/2009/2562
ER  -