A bi-objective hub-and-spoke approach for reconfiguring Web communities

• Autor(a) principal: Colaço, Susana
• Data de Publicação: 2008
• Outros Autores: Pato, Margarida Vaz
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.5/1428
• topic: multi-objective; heuristics; Web communities; Web communities

Web communities in general grow naturally, thus creating unbalanced network structures where a few domains centralise most of the linkups. When one of them breaks down, a significant part of the community might be unable to communicate with the remaining domains. Such a situation is highly inconvenient, as in the case of wishing to pursue distribution policies within the community, or for marketing purposes. In order to reduce the damages of such an occurrence, the Web community should be reconfigured, in such a way that a complete sub-network of main domains -the hubs - is identified and that each of the other domains of the community - the spokes - is doubly linked at least with a hub. This problem can be modellised through a bi-objective optimisation problem, the Web Community Reconfiguring Problem, which will be presented in this paper. A bi-objective mixed binary formulation will also be shown, along with a brief description of GRASP, tabu search and hybrid heuristics which were developed to find feasible solutions to the problem, possibly efficient solutions to the bi-objective problem. A computational experiment is reported, involving comparison of these metaheuristics when applied to several Web communities, obtained by crawling the Web and using epistemic boundaries and to other randomly generated ones. The heuristics revealed excellent quality for the small dimension cases whose efficient solutions were roughly all determined. As for the other medium and higher dimension instances, the heuristics were successful in building a wide variety of feasible solutions that are candidate efficient solutions. The best behaviour was attained with the GRASP and the GRASP and tabu hybrid search. Comparison of some metrics before and after reconfiguration confirmed that the final structures are more balanced in terms of degree distribution reinforcing the connecting effect imposed by the reconfiguration process.