Document Classification using Neural Networks

The paper starts with the need for classification. Then the reasons why neural networks are suitable for document classification are explained. The paper continues with the details of the most commonly used topologically organized network model proposed by Kohonen (1982), referred to as the self-organizing map (SOM). The general idea proposed is to display the contents of a document library by representing similar documents in similar regions of the map. Without knowledge of the type of and the organization of the documents it is difficult to get satisfying results without multiple training runs. So the paper discusses the possibility to use a a hierarchical structure of independent SOMS, referred to as GHSOM, where for every unit of a map, a SOM is added to the next layer. The essential steps in the classification system are given. The paper concludes with applications of document classification using neural networks.