In recent times, online harassment due to cyber-bullying is significantly increased with the growth of social media users. Cyber-bullying is a technique to harass users using electronic messages. Many researchers attack this problem using natural language processing. Most of them detect whether a message is a bully or not. In this paper, multiple deep learning models are introduced to detect not only bullying messages but also the annotation of cyber-bullying. Annotation detection of cyber-bullying assigns a proper description in which category a message belongs. The advantage of annotation detection is to warn the user by giving an alert message with proper annotation when the user sends or posts a message on social media. If this feature is combined with popular social network sites like Facebook, Twitter, WhatsApp, etc., this can be an additional filter to alert the user that they are going to post or send a bullied message of which type. Social media messages are unstructured as it includes text, URL link, emojis, abbreviations, etc. Most of the previous works are conducted to detect bullying messages only considering important words in the text, neglecting the other attributes in the message like URL links, emojis, and abbreviations. In this paper, an advanced pre-processing technique is proposed by considering some of the attributes in the messages like URL, abbreviation, number, emojis, etc., to detect bullying messages. In this work, six models, i.e., three deep learning models combined with two different word-embedding models have been employed for annotation detection. The performances of each of these six models are measured twice, by employing traditional pre-processing, and proposed advanced pre-processing. The experimental results show that the advanced pre-processing works better in the case of all six models.
