Browsing by Author "Pappa, Gisele Lobo"

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    Discretizador heurístico para o contexto de classificação hierárquica.
    (2016) Galvão, Leandro Ribeiro; Merschmann, Luiz Henrique de Campos; Silla Júnior, Carlos Nascimento; Pappa, Gisele Lobo; Ferreira, Almeida Ferreira
    Diferentes tipos de problemas de classificação podem ser encontrados na literatura, cada qual possuindo seu nível de complexidade. Diversos algoritmos de aprendizado de máquina requerem atributos discretos e nesses casos o pré-processamento da base de dados né necessário. Na literatura, os trabalhos apresentam diversos métodos de discretização, porém até o momento, não há nenhum método de discretização supervisionado projetado para ser utilizada em conjunto com classificadores hierárquicos globais. Neste trabalho é proposto um método supervisionado de discretização capaz de lidar com bases do contexto de classificação hierárquica. Esse método corresponde a uma heurística, denominada Agglomerative Discretization Heuristic for Hierarchical Classification - ADH2C, que foi projetada para ser utilizada em conjunto com classificadores hierárquicos globais. A avaliação da qualidade da discretização realizada pela heurística ADH2C foi feita a partir de experimentos comparativos com métodos de discretização não-supervisionados Equal-Width (EW) e Equal-Frequency (EF). A qualidade da discretização foi medida por meio do desempenho preditivo pelo classificador hierárquico Global Model Naive Bayes (GMNB) utilizando-se 9 bases de dados de bioinformática pré-processadas pelos métodos de discretização EW, EF e ADH2C. Os experimentos realizados neste trabalho mostraram que para a maioria das bases de dados utilizadas, o classificador GMNB alcançou o melhor desempenho preditivo (hF) quando utilizou as bases de dados pré-processadas pela heurística ADH2C. A melhora no desempenho preditivo do GMNB, utilizando as bases de dados pré-processadas pela heurística ADH2C, evidencia sua aplicabilidade no contexto de classificação hierárquica monorrótulo.
  • No Thumbnail Available
    Item
    Exploring multiple evidence to inferusers’ location in twitter.
    (2015) Rodrigues, Erica Castilho; Assunção, Renato Martins; Pappa, Gisele Lobo; Renno, Diogo
    Online social networks are valuable sources of information to monitor real-time events, such as earthquakes and epidemics. For this type of surveillance, users’ location is an essential piece of information, but a substantial number of users choose not to disclose their geographical location. However, characteristics of the users' behavior, such as the friends they associate with and the types of messages published may hint on their spatial location. In this paper, we propose a method to infer the spatial location of Twitter users. Unlike the approaches proposed so far, it incorporates two sources of information to learn geographical position: the text posted by users and their friendship network. We propose a probabilistic approach that jointly models the geographical labels and Twitter texts of users organized in the form of a graph representing the friendship network. We use the Markov random field probability model to represent the network, and learning is carried out through a Markov Chain Monte Carlo simulation technique to approximate the posterior probability distribution of the missing geographical labels. We show the accuracy of the algorithm in a large dataset of Twitter users, where the ground truth is the location given by GPS. The method presents promising results, with little sensitivity to parameters and high values of precision.
  • No Thumbnail Available
    Item
    A methodology for photometric validation in vehicles visual interactive systems.
    (2012) Faria, Alexandre Wagner Chagas; Menotti, David; Pappa, Gisele Lobo; Lara, Daniel da Silva Diogo; Araújo, Arnaldo de Albuquerque
    This work proposes a methodology for automatically validating the internal lighting system of an automobile by assessing the visual quality of each instrument in an instrument cluster (IC) (i.e., vehicle gauges, such as speedometer, tachometer, temperature and fuel gauges) based on the user’s perceptions. Although the visual quality assessment of an instrument is a subjective matter, it is also influenced by some of its photometric features, such as the light intensity distribution. This work presents a methodology for identifying and quantifying non-homogeneous regions in the lighting distribution of these instruments, starting from a digital image. In order to accomplish this task, a set of 107 digital images of instruments were acquired and preprocessed, identifying a set of instrument regions. These instruments were also evaluated by common drivers and specialists to identify their non-homogenous regions. Then, for each region, we extracted a set of homogeneity descriptors, and also proposed a relational descriptor to study the homogeneity influence of a region in the whole instrument. These descriptors were associated with the results of the manual labeling, and given to two machine learning algorithms, which were trained to identify a region as being homogeneous or not. Experiments showed that the proposed methodology obtained an overall precision above 94% for both regions and instrument classifications. Finally, a meticulous analysis of the users’ and specialist’s image evaluations is performed