Descoberta de perfis de consumo a partir de uma arquitetura de data warehouse voltada para dados da web

Abstract

Environments for Data Warehouse (DW) are typically used to manipulate data bases concrete operational. However, with the growth of relevant information available on the Web about the interests and desires of consumer users, the need arises to use the DW architecture of the Web, which requires that both the mechanisms of data extraction as later viewing are rethought. This work demonstrated the possibility of using a DW typical architecture within a Web application oriented analysis consumption profiles. The application is a social network of consumption, which aims to bring together people who have similar buying profiles. Data extraction occurs through a registration form to purchase, where users provide their information consumption. Visualization is performed by generating communities that allow exchange of experiences of users who have similar consumption profiles. To perform discovery of consumption profiles were evaluated three unsupervised learning algorithms: K-means , Expectation Maximization and Farthest First. Through experiments we observed a similar behavior between the approaches when the consumer profiles are similar. With well defined profiles , the EM algorithm achieved better results .