| Summary: | 碩士 === 國立中正大學 === 資訊管理系研究所 === 106 === Social network such as Facebook and Twitter have promoted communication between people , but some unusual social user overuse the social media like cyberbullying. Cyberbullying in social network will cause more negative effect than traditional bullying. For example, many tweets on Twitter can be read publicly by registered users around the world , so that cyberbullying messages can spread quickly.
When a cyber criminal sends a text message to attack other social users, such behavior may involve cyberaggression and cyberbullying. In the past, the cyberbullying literature focused on a single text message to make the expert distinguish whether the cyber bullying or not. This situation is more like the criteria of cyber aggression. The criteria of the cyberbullying must include the imbalance of power and repetition, in order to consider the power of imbalance and repetition, the expert should not only use text message to identify cyberbullying, they must understand the whole information on the social network such as likes or replies in a tweet and the tweet that social user received recently.
This study collects tweets and user information on the Twitter, and create three different feature group (text, user, social). Using Random Forest, Logistic Regression, and Support Vectors three classifiers to build cyberbullying model. With the comparison of models established by three different classifiers, the best predict model is selected as the research discussed.
|
|---|
