Finding English and translated Arabic documents similarities using GHSOM

Abstract

The idea of finding similar news across Arabic and English sources is that to provide the audience with multiple views of the broadcasted news because reading the news from a single source may not always reflects on what happening around the world due different background, cultures and opinions of the readers and writers. To achieve this goal there are many techniques have been used to cluster the documents with similar themes. In this paper, we analyze the similarity of the views on the news written in the news translations form Arabic and English texts using Self-organizing Map (SOM). However, we have found there are some difficulties in SOM that affect its performance. In order to improve the problems of performance, we have used a Growing Hierarchical Self-organizing Map (GHSOM). The main advantage of such a mapping is the ease by which a user gains an idea regarding the structure of the data by analyzing the map. Thousands of news documents have been collected from Arabic and English news sources from the web in order to train both algorithms. Form experiments, the results show that using GHSOM is better in terms of clustering documents with the same opinions.