Deep Learning Approach for Classification of Tweets in Detecting Cyber Truculent

Abstract

Identification and Classification of truculent tweets is a serious problem. Studies have shown how truculent tweets on Twitter have caused anxiety and in extreme cases death of victims. Various researches have applied Machine Learning approaches while others Deep Learning in classifying toxic sentences. Impressive results have been obtained using Deep Learning approaches, the focus of the research is to develop an extended model for deep learning classification using cyber tweets. The Model developed in this research work used labelled dataset (twitter_parsed_dataset.csv), Maximum Entropy was used to extract keywords and entities. One-Dimensional Convolutional Neural Network (1d-CNN) was used to detect truculent in tweets. The experimental result from the developed model consider four preprocess datasets for classification, the Unigram, Bigram,  Trigram and N-gram. The result that was obtained for Accuracy 96.1%, Precision 93.6%, Recall 73.7% and F1-Score 83.8% for different test during classification. The result obtained from the developed model for accuracy, 96.1% was compared with related work Banerjee’s accuracy of 93.97%. The developed model can be used for auto detection of truculent messages in cyberspace.