Introduction
What is cyberbullying?
- The use of digital technologies to intimidate, harass, threaten, or otherwise embarrass another person or group.
- It involves using technology to deliberately and repeatedly intimidate, threaten, embarrass, or harm others.
Why detect cyberbullying?
- Real-time Detection
- Safety
- Consistency / Reduced Bias
- Data-Driven Insights
- Detection of Subtle and Hidden Patterns
Challenges in detecting cyberbullying
- Context matters
- Words or phrases like “You're dumb” could be playful among friends, but malicious in other scenarios.
- Subjective nature of language
- Words may hold different meanings depending on the person's background, region, or community.
- Spellchecking / Formatting
- For instance, a word like “idiot” might be written as “!d10t,” making it hard for traditional keyword-based systems to recognize.
- Nuanced Expressions like Sarcasm and Irony
- The model might interpret this phrase as positive, even though it was meant sarcastically.
Examples
| Non-Flagged Messages | Flagged Messages |
|---|---|
| Going to make some pancakes.....Don't hve any strawberries ....🍓🍓🍓🍓but I hve bananas .....🍌🍌🍌... | RT @Mooseoftorment Call me sexist, but when I go to an auto place, I'd rather talk to a guy |
| RT @ahtweet: @freebsdgirl How dare you have feelings is a fantastic way to dehumanize someone. | @KamelNasrEldien Islam is about as perfect as Nazism. |
| RT @Mumdeep: Can you just kick Kat out for being a strategic arsehole? #mkr | RT @TheRealCiaranW A lady is currently driving this bus... Funnily enough she's hit the curb twice #NotSexist #stereotypical |
| @Taxis__ don't really know, most of those people are dumb. | RT @TheMeninism: guys who sell drugs get more girls than guys with degrees, but that says more about women than men ☕️🐸 |
Project Overview
- We implemented two different machine learning algorithms, CNN-BiLSTM and Random Forest Classifier. Its intended use is within online forums and social media sites. It will be run over posts and comments and flag messages as cyberbullying appropriately.
- We define cyberbullying as the use of technology to harass, threaten, embarrass, or target another. Cyberbullying can include but is not limited to racism, sexism, homophobia, antisemitism, and religious discrimination.
- Designing software to detect cyberbullying prevents further cyberbullying in online spaces, which allows users to enjoy their experience using these platforms without fear of harassment. Users benefit by being in a safe environment and companies benefit by offering that environment to users.