Document classification using self-organizing maps

The number of internet web pages are growing at a high rate. Automated search engines are becoming insuffienct in returning appropiate results to the search queries. The directory sites can't keep up with evaluation of all web pages, therefore the quality and scope of their directories are decreased. Furthermore, links are becoming out of date. On the other hand, the number of the documents saved in computers are increasing. As a result, automatic classification of the web pages and documents takes more attention. In this study, it is aimed to classify the documents according to their contents. For this purpose, a classification system is developed that is based on the Self- Organizing Map (SOM) algorithm, which is an effective unsupervised artificial neural network method for high-dimensional data. The results obtained from self-organizing maps are compared with hierarchical classification, an effective classification method. For both methods, the significant and distinctive words within each document are found by using a labeling algorithm. Before the classification process, some preprocessing steps are applied, these are stopword removal, removing very low and very high frequently used words, indexing words, calculating weight vectors, equalizing the dimension of the weight vectors, and normalization. In experimental studies, two different document libraries are being used. The first library is prepared by collecting random news abstracts from an online news site and the second library is prepared by gathering different course contents from web pages of different universities. Both of libraries are being successfully classified. Furthermore, documents with different contents can also be classified by using this developed system.