Regensburg 2007 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 4: Statistical physics of complex networks I

DY 4.2: Vortrag

Montag, 26. März 2007, 12:15–12:30, H3

Applying direct weighted networks to recommendation systems — •Stefano Battiston, Frank Walter, and Frank Schweitzer — Chair of Systems Design, ETH Zurich, Kreuzplatz 5, 8032 Zurich, Switzerland

Models of large evolving social networks find today numerous applications in on-line web services, in particular for recommendation systems. We model an evolving weighted directed network in which agents ask for recommendations to their peers about a set of items. A node represents an agent with expertise in a specific domain and with heterogenous preferences (summarized in a profile), while the weight of a link represents the trust of an agent towards another one. Differently from other network models, here the weights of the outgoing links of a node evolve according to a utility function of the node. By means of mean field approximations, we derive an expression for the performance of the system as a function of the frequency and heterogeneity of profiles across agents. We find that the critical parameters for the performance of the system are network density, preference heterogeneity among agents, and expertise sparseness. The mean-field approximation leads to accurate predictions in a broad range of these parameters and provides useful constraints for the design of real applications.