We have designed a multi-agent dynamic systems that move in a virtual space according to repulsive and attractive forces that are defined from the complex network structure. In our approach we consider that each agent does not have access to the information about the general structure of the network, because it is non attainable to have a complete representation of the network inside each agent, but only searches for its first order connections. The links to each of its neighbors conditions the movement of the agent, pulling it by attractive forces. This dynamical system reaches an stable global state where agents tend to form clusters that correspond to high order connections in the network. We apply this approach to Amazon's similar product's network, based in the "client who bought this also bought that" feature looking for hidden product communities that break through the immediate categorization of products given a catalog. We report preliminary results of simulations carried out in Netlogo.
展开▼