Authors - Deepali Dhaka, Saima Saleem, Monica Mehrotra Abstract - Spammers disseminating obscene content on Twitter have been studied and detected using various hybrid features and machine-learning approaches in past. To have greater insight into data prevailing in form of text on platforms like Twitter, their correct vector representation is paramount. Our goal is to understand what encoding techniques are more suitable for representing long text documents. We proposed a novel deep learning model consisting of a Universal Sentence Encoder (USE) as a feature extractor and an artificial neural network (ANN) as a classifier. We transform all the sentence vectors representing the tweets of a user into a document vector. These vectors are used as high-quality features to be processed by the artificial neural network for classification. To check the effectiveness of our proposed model, different sentence embedding techniques such as Doc2Vec, Infersent, and SentenceBert have been used and compared with the proposed model. Experimental results show that the pro-posed model outperforms all of them in terms of recall, precision, f1-score, and AUROC. Our results show that a simple ANN combined with USE based deep learning approach can be a robust solution for the detection of spammers on Twitter.
Open Zoom