Text Dissimilarities Predictions Using Convolutional Neural Networks and Clustering

Text mining is still interesting problem which can be solved using different methods and it can bring many surprising results. Our goal is to analyze text segments of some long text and find segments which have a different stylometry in comparison to the other. We developed two-steps method: (1) clustering of segments, and (2) classification of segments using convolutional neural networks. The method was tested on ten Arabic and ten English long texts. Our new method contributes to the previous results about texts. It classify a text into two classes: reliable or suspicious text and in many cases it confirms the previous evaluation.