Word2vec and LSTM-based Offensive Content Detection

초록

The volume of offensive content on Twitter is increasing drastically. These contents are informal and unstructured. The existing systems based on traditional approaches are unable to detect offensive content accurately. Therefore, this paper presents a Word2vec- and LSTM-based detection technique to automatically identify and block access to the offensive content. The proposed technique obtains higher accuracy of 91% in offensive text classification than the current state-of-art approaches. This shows that our system successfully distinguishes offensive and non-offensive contents from normal tweets.