A comparison between symbolic and non­symbolic machine learning techniques in automated annotation of the "Keywords" field of SWISS­PROT

Luciana F. SchroederAna L. C. BazzanJoão ValiatiPaulo M. EngelSérgio Ceroni

The aim of this work is to carry out a comparison between symbolic and non­symbolic approaches regarding the task of automated annotation of the field called Keywords in SWISS­PROT. The non­symbolic technique employed was a feedforward artificial neural network (ANN), while the symbolic ones was CN2. Using an ANN trained with the well­known Backpropagation algorithm over previously annotated data from public databases like SWISS­PROT, a classifier was built up that maps attributes of a specific protein to keywords encountered in SWISS­PROT and TrEMBL databases. The symbolic counterpart, CN2, builds a specific classifier for each keyword. Regarding the non­symbolic approach, the resulted classifier is much more compact than the symbolic counterpart. However, the symbolic one had a slightly better performance and is also more "readable" to the end user. The performance of the obtained classifier was evaluated using data taken out SWISS­PROT (for training) and TrEMBL for validation.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: