The Effects of Corpus Type and Size on the Classification of News

Abstract

A tremendous amount of data is produced and stored in every second. This increase causes problems in controlling, storing and analyzing the data. These data have to be classified in order to access information rapidly. We researched word embedding tools, the effects of corpus types and size on the success of classification. We present the large volume data set of Turkish news that will be classified, unlabeled 2 million and 20 million Turkish tweets to researchers. Word embedding tools GloVe and fastText success were inspected on 40 thousand Turkish news that were labeled 13 different classes, 2 millions of tweets and 20 millions of tweets.