A Novel Twitter Spam Detection Technique by Integrating Inception Network with Attention based LSTM

Online Social Networking sites have become a well-known way for web surfers to connect and meet. Twitter got to be a well-known micro blogging site that clients post and associate with messages known as tweets. As this networking site gains its popularity, spammers target Twitter to spread spam posts. Hence, several spam detection techniques have been proposed by analysts to create Twitter a spam-free stage. Be that as it may, the accessible machine learning algorithms cannot effectively distin- guish spammers on Twitter because of reasonable information controls by unsolicited clients to elude spam discovery. As a result, here, we present an incipient approach predicated on a deep learning technique that leverages a text-predicated feature to detect spammers. A novel architecture that contains a one-dimensional dimension reduction inception module stacked with LSTM along with an attention layer is introduced here. Within the proposed model, the inception module extricates the features from the vectors after GloVe word embedding, and then LSTM is utilized to get the context representations. An Attention layer is also used in this model to give attention to the data outputted from LSTM module. At long last, the sigmoid classifier is utilized to classify the labels as spam or ham. Here, the execution of our proposed model is compared with four machine learning-based and two deep learning-based approaches, exhibiting our approach acquired the best results with an F1-score of 95.74, accuracy of 95.75, and precision of 95.58.