Anti Bully

• 

  Example bot reply to a harassment tweet

• 

  Example bot reply to a non-harassment tweet

Inspiration

Cyberbullying is an extremely prevalent issue in the modern digital world. Yet many social media platforms such as Twitter don’t have intelligent cyberbullying detection systems so they don’t effectively address this issue. Because of this, we wanted to create a classifier using artificial intelligence to determine whether a tweet is cyberbullying or not.

What it does

Our project is a twitter bot that uses a machine learning classifier to detect cyberbullying and automatically reply to the tweet to warn the user not to bully. We chose Twitter for this project because it was simple to search through tweets and reply with a bot, but in the future, we could expand this project for other social media platforms.

How we built it

We used Python for the entire project, utilizing the Twitter API for searching for tweets and replying automatically. We used a Naive Bayes machine learning classifier (uses conditional probability concepts) to calculate the probability that a given tweet is harassment or not.

Challenges we ran into

The first main challenge we ran into was finding a good dataset. Not many well labeled cyberbullying datasets exist, and we spent many hours trying to find a good dataset. We also had difficulty getting the classifier to work (tokenizing the input tweets) and dealing with unknown (previously unencountered words). Furthermore, since we were using a free version of the twitter API, there was a hard limit on the number of API requests we could make, causing many delays in testing.

Accomplishments that we're proud of

• Getting twitter API to work
• Classifier performs quite well on test tweets the judges came up with (despite a pretty badly mislabeled dataset)

What we learned

We learned how to use the Twitter API from scratch, as well as how to code an effective naive bayes classifier. We also realized that finding good data was necessary for an accurate algorithm.

What's next for Anti Bully

In the future, we plan on expanding the model to other social media platforms besides just Twitter (such as facebook, reddit, and instagram). We also hope to find a better dataset and achieve even higher accuracy.

Built With

• machine-learning
• naive-bayes
• python
• twitter