<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0710">
  <Title>Classifying Amharic News Text Using Self-Organizing Maps</Title>
  <Section position="7" start_page="76" end_page="76" type="evalu">
    <SectionTitle>
6 Summary and Conclusions
</SectionTitle>
    <Paragraph position="0"> A set of experiments investigated text retrieval of selected Amharic news items using Self-Organizing Maps, an unsupervised learning neural network method. 101 training set items, 25 queries, and 105 test set items were selected. The content of each news item was taken as the basis for document indexing, and the content of the specific query was taken for query indexing. A term-document matrix was generated and the occurrence of terms per document was registered. This original matrix was changed to a weighted matrix using the log-entropy scheme. The weighted matrix was further reduced using SVD. The length of the query vector was also reduced using the global weight vector obtained in weighing the original matrix.</Paragraph>
    <Paragraph position="1"> The ANN model using unnormalised vector space had a precision of 10.5%, whereas the best ANN model using reduced dimensional vector space performed at a 60.0% level for the test set. For this configuration we also tried to classify the data around a query content, taken that query as class label. The results obtained then were 72.8% for the training set and 69.5% for the test set, which is encouraging.</Paragraph>
  </Section>
class="xml-element"></Paper>