Sistemas baseados em mapas auto-organizÃveis para organizaÃÃo automÃtica de documentos texto

AUTOR(ES)

Renato Fernandes CorrÃa

DATA DE PUBLICAÇÃO

2008

RESUMO

This work proposes and evaluates hybrid systems for automatic text document organization based on Self-Organizing Maps (SOM). The aim is to design a system that combines SOM with other clustering algorithms, in order to generate document maps for large text document collections of good quality at a low computational cost. The posprocessing of a neural network SOM trained with the vectors that represent documents of a collection generates a document map. Document maps of good quality are those that represent well the relations of content-based similarity between documents A document map organizes a text document collection in accordance with the contentbased similarity, and it has application in improving of the processes of information retrieval, exploration, browsing and text mining on a collection. Several works in the literature of neural networks have used SOM to create document maps. However, the training of SOM networks is still an expensive computational task for large text document collections. Some methods considered in literature to construct document maps more quickly reduce drastically the quality of the generated map. Moreover, hybrid systems combining SOM with other clustering algorithms are not investigated enough in literature. These facts had motivated the present work. The results show that the careful combination of traditional clustering algorithms like Kmeans and Leader with SOM networks is able to produce very efficient hybrid systems. For this reason, a hybrid system was proposed, in order to implement an automatic process to generate document maps of good quality at a low computational cost. These hybrid systems represent a advance in the field of document organization systems, as well as SOM-based neural hybrid systems, by providing important results for several practical applications in design of systems as: search engines, systems for digital libraries and systems for text mining

ASSUNTO(S)

algoritmos de agrupamento sistemas hÃbridos inteligentes artificial neural networks clustering algorithms hybrid intelligent systems redes neurais artificiais information retrieval recuperaÃÃo de informaÃÃo ciencia da computacao

Documentos Relacionados

• GeraÃÃo de trajetÃrias de estados por mapas auto-organizÃveis com topologia dinÃmica
• Um framework para transformaÃÃo automÃtica de documentos e extraÃÃo de informaÃÃes baseado em XML
• Sistema inteligente de organizaÃÃo de documentos.
• Mapas auto-organizaveis na classificação de dados em quimica analitica
• Um ambiente para geraÃÃo automÃtica de biblioteca de componentes de comunicaÃÃo em sistemas embarcados distribuÃdos